agora inbox for [email protected]
help / color / mirror / Atom feed[PATCH v20 1/6] Split vacuumdb to create vacuuming.c/h
477+ messages / 2 participants
[nested] [flat]
* [PATCH v20 1/6] Split vacuumdb to create vacuuming.c/h
@ 2025-08-30 12:39 Álvaro Herrera <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Álvaro Herrera @ 2025-08-30 12:39 UTC (permalink / raw)
---
src/bin/scripts/Makefile | 4 +-
src/bin/scripts/meson.build | 28 +-
src/bin/scripts/vacuumdb.c | 1048 +----------------------------------
src/bin/scripts/vacuuming.c | 978 ++++++++++++++++++++++++++++++++
src/bin/scripts/vacuuming.h | 95 ++++
5 files changed, 1119 insertions(+), 1034 deletions(-)
create mode 100644 src/bin/scripts/vacuuming.c
create mode 100644 src/bin/scripts/vacuuming.h
diff --git a/src/bin/scripts/Makefile b/src/bin/scripts/Makefile
index f6b4d40810b..019ca06455d 100644
--- a/src/bin/scripts/Makefile
+++ b/src/bin/scripts/Makefile
@@ -28,7 +28,7 @@ createuser: createuser.o common.o $(WIN32RES) | submake-libpq submake-libpgport
dropdb: dropdb.o common.o $(WIN32RES) | submake-libpq submake-libpgport submake-libpgfeutils
dropuser: dropuser.o common.o $(WIN32RES) | submake-libpq submake-libpgport submake-libpgfeutils
clusterdb: clusterdb.o common.o $(WIN32RES) | submake-libpq submake-libpgport submake-libpgfeutils
-vacuumdb: vacuumdb.o common.o $(WIN32RES) | submake-libpq submake-libpgport submake-libpgfeutils
+vacuumdb: vacuumdb.o vacuuming.o common.o $(WIN32RES) | submake-libpq submake-libpgport submake-libpgfeutils
reindexdb: reindexdb.o common.o $(WIN32RES) | submake-libpq submake-libpgport submake-libpgfeutils
pg_isready: pg_isready.o common.o $(WIN32RES) | submake-libpq submake-libpgport submake-libpgfeutils
@@ -50,7 +50,7 @@ uninstall:
clean distclean:
rm -f $(addsuffix $(X), $(PROGRAMS)) $(addsuffix .o, $(PROGRAMS))
- rm -f common.o $(WIN32RES)
+ rm -f common.o vacuuming.o $(WIN32RES)
rm -rf tmp_check
export with_icu
diff --git a/src/bin/scripts/meson.build b/src/bin/scripts/meson.build
index 80df7c33257..a4fed59d1c9 100644
--- a/src/bin/scripts/meson.build
+++ b/src/bin/scripts/meson.build
@@ -12,7 +12,6 @@ binaries = [
'createuser',
'dropuser',
'clusterdb',
- 'vacuumdb',
'reindexdb',
'pg_isready',
]
@@ -35,6 +34,33 @@ foreach binary : binaries
bin_targets += binary
endforeach
+vacuuming_common = static_library('libvacuuming_common',
+ files('common.c', 'vacuuming.c'),
+ dependencies: [frontend_code, libpq],
+ kwargs: internal_lib_args,
+)
+
+binaries = [
+ 'vacuumdb',
+]
+foreach binary : binaries
+ binary_sources = files('@[email protected]'.format(binary))
+
+ if host_system == 'windows'
+ binary_sources += rc_bin_gen.process(win32ver_rc, extra_args: [
+ '--NAME', binary,
+ '--FILEDESC', '@0@ - PostgreSQL utility'.format(binary),])
+ endif
+
+ binary = executable(binary,
+ binary_sources,
+ link_with: [vacuuming_common],
+ dependencies: [frontend_code, libpq],
+ kwargs: default_bin_args,
+ )
+ bin_targets += binary
+endforeach
+
tests += {
'name': 'scripts',
'sd': meson.current_source_dir(),
diff --git a/src/bin/scripts/vacuumdb.c b/src/bin/scripts/vacuumdb.c
index fd236087e90..b1be61ddf25 100644
--- a/src/bin/scripts/vacuumdb.c
+++ b/src/bin/scripts/vacuumdb.c
@@ -14,92 +14,13 @@
#include <limits.h>
-#include "catalog/pg_attribute_d.h"
-#include "catalog/pg_class_d.h"
#include "common.h"
-#include "common/connect.h"
#include "common/logging.h"
-#include "fe_utils/cancel.h"
#include "fe_utils/option_utils.h"
-#include "fe_utils/parallel_slot.h"
-#include "fe_utils/query_utils.h"
-#include "fe_utils/simple_list.h"
-#include "fe_utils/string_utils.h"
-
-
-/* vacuum options controlled by user flags */
-typedef struct vacuumingOptions
-{
- bool analyze_only;
- bool verbose;
- bool and_analyze;
- bool full;
- bool freeze;
- bool disable_page_skipping;
- bool skip_locked;
- int min_xid_age;
- int min_mxid_age;
- int parallel_workers; /* >= 0 indicates user specified the
- * parallel degree, otherwise -1 */
- bool no_index_cleanup;
- bool force_index_cleanup;
- bool do_truncate;
- bool process_main;
- bool process_toast;
- bool skip_database_stats;
- char *buffer_usage_limit;
- bool missing_stats_only;
-} vacuumingOptions;
-
-/* object filter options */
-typedef enum
-{
- OBJFILTER_NONE = 0, /* no filter used */
- OBJFILTER_ALL_DBS = (1 << 0), /* -a | --all */
- OBJFILTER_DATABASE = (1 << 1), /* -d | --dbname */
- OBJFILTER_TABLE = (1 << 2), /* -t | --table */
- OBJFILTER_SCHEMA = (1 << 3), /* -n | --schema */
- OBJFILTER_SCHEMA_EXCLUDE = (1 << 4), /* -N | --exclude-schema */
-} VacObjFilter;
-
-static VacObjFilter objfilter = OBJFILTER_NONE;
-
-static SimpleStringList *retrieve_objects(PGconn *conn,
- vacuumingOptions *vacopts,
- SimpleStringList *objects,
- bool echo);
-
-static void vacuum_one_database(ConnParams *cparams,
- vacuumingOptions *vacopts,
- int stage,
- SimpleStringList *objects,
- SimpleStringList **found_objs,
- int concurrentCons,
- const char *progname, bool echo, bool quiet);
-
-static void vacuum_all_databases(ConnParams *cparams,
- vacuumingOptions *vacopts,
- bool analyze_in_stages,
- SimpleStringList *objects,
- int concurrentCons,
- const char *progname, bool echo, bool quiet);
-
-static void prepare_vacuum_command(PQExpBuffer sql, int serverVersion,
- vacuumingOptions *vacopts, const char *table);
-
-static void run_vacuum_command(PGconn *conn, const char *sql, bool echo,
- const char *table);
+#include "vacuuming.h"
static void help(const char *progname);
-
-void check_objfilter(void);
-
-static char *escape_quotes(const char *src);
-
-/* For analyze-in-stages mode */
-#define ANALYZE_NO_STAGE -1
-#define ANALYZE_NUM_STAGES 3
-
+static void check_objfilter(void);
int
main(int argc, char *argv[])
@@ -145,10 +66,6 @@ main(int argc, char *argv[])
int c;
const char *dbname = NULL;
const char *maintenance_db = NULL;
- char *host = NULL;
- char *port = NULL;
- char *username = NULL;
- enum trivalue prompt_password = TRI_DEFAULT;
ConnParams cparams;
bool echo = false;
bool quiet = false;
@@ -168,13 +85,18 @@ main(int argc, char *argv[])
vacopts.process_main = true;
vacopts.process_toast = true;
+ /* the same for connection parameters */
+ memset(&cparams, 0, sizeof(cparams));
+ cparams.prompt_password = TRI_DEFAULT;
+
pg_logging_init(argv[0]);
progname = get_progname(argv[0]);
set_pglocale_pgservice(argv[0], PG_TEXTDOMAIN("pgscripts"));
- handle_help_version_opts(argc, argv, "vacuumdb", help);
+ handle_help_version_opts(argc, argv, progname, help);
- while ((c = getopt_long(argc, argv, "ad:efFh:j:n:N:p:P:qt:U:vwWzZ", long_options, &optindex)) != -1)
+ while ((c = getopt_long(argc, argv, "ad:efFh:j:n:N:p:P:qt:U:vwWzZ",
+ long_options, &optindex)) != -1)
{
switch (c)
{
@@ -195,7 +117,7 @@ main(int argc, char *argv[])
vacopts.freeze = true;
break;
case 'h':
- host = pg_strdup(optarg);
+ cparams.pghost = pg_strdup(optarg);
break;
case 'j':
if (!option_parse_int(optarg, "-j/--jobs", 1, INT_MAX,
@@ -211,7 +133,7 @@ main(int argc, char *argv[])
simple_string_list_append(&objects, optarg);
break;
case 'p':
- port = pg_strdup(optarg);
+ cparams.pgport = pg_strdup(optarg);
break;
case 'P':
if (!option_parse_int(optarg, "-P/--parallel", 0, INT_MAX,
@@ -227,16 +149,16 @@ main(int argc, char *argv[])
tbl_count++;
break;
case 'U':
- username = pg_strdup(optarg);
+ cparams.pguser = pg_strdup(optarg);
break;
case 'v':
vacopts.verbose = true;
break;
case 'w':
- prompt_password = TRI_NO;
+ cparams.prompt_password = TRI_NO;
break;
case 'W':
- prompt_password = TRI_YES;
+ cparams.prompt_password = TRI_YES;
break;
case 'z':
vacopts.and_analyze = true;
@@ -380,66 +302,9 @@ main(int argc, char *argv[])
pg_fatal("cannot use the \"%s\" option without \"%s\" or \"%s\"",
"missing-stats-only", "analyze-only", "analyze-in-stages");
- /* fill cparams except for dbname, which is set below */
- cparams.pghost = host;
- cparams.pgport = port;
- cparams.pguser = username;
- cparams.prompt_password = prompt_password;
- cparams.override_dbname = NULL;
-
- setup_cancel_handler(NULL);
-
- /* Avoid opening extra connections. */
- if (tbl_count && (concurrentCons > tbl_count))
- concurrentCons = tbl_count;
-
- if (objfilter & OBJFILTER_ALL_DBS)
- {
- cparams.dbname = maintenance_db;
-
- vacuum_all_databases(&cparams, &vacopts,
- analyze_in_stages,
- &objects,
- concurrentCons,
- progname, echo, quiet);
- }
- else
- {
- if (dbname == NULL)
- {
- if (getenv("PGDATABASE"))
- dbname = getenv("PGDATABASE");
- else if (getenv("PGUSER"))
- dbname = getenv("PGUSER");
- else
- dbname = get_user_name_or_exit(progname);
- }
-
- cparams.dbname = dbname;
-
- if (analyze_in_stages)
- {
- int stage;
- SimpleStringList *found_objs = NULL;
-
- for (stage = 0; stage < ANALYZE_NUM_STAGES; stage++)
- {
- vacuum_one_database(&cparams, &vacopts,
- stage,
- &objects,
- vacopts.missing_stats_only ? &found_objs : NULL,
- concurrentCons,
- progname, echo, quiet);
- }
- }
- else
- vacuum_one_database(&cparams, &vacopts,
- ANALYZE_NO_STAGE,
- &objects, NULL,
- concurrentCons,
- progname, echo, quiet);
- }
-
+ vacuuming_main(&cparams, dbname, maintenance_db, &vacopts, &objects,
+ analyze_in_stages, tbl_count, concurrentCons,
+ progname, echo, quiet);
exit(0);
}
@@ -466,885 +331,6 @@ check_objfilter(void)
pg_fatal("cannot vacuum all tables in schema(s) and exclude schema(s) at the same time");
}
-/*
- * Returns a newly malloc'd version of 'src' with escaped single quotes and
- * backslashes.
- */
-static char *
-escape_quotes(const char *src)
-{
- char *result = escape_single_quotes_ascii(src);
-
- if (!result)
- pg_fatal("out of memory");
- return result;
-}
-
-/*
- * vacuum_one_database
- *
- * Process tables in the given database.
- *
- * There are two ways to specify the list of objects to process:
- *
- * 1) The "found_objs" parameter is a double pointer to a fully qualified list
- * of objects to process, as returned by a previous call to
- * vacuum_one_database().
- *
- * a) If both "found_objs" (the double pointer) and "*found_objs" (the
- * once-dereferenced double pointer) are not NULL, this list takes
- * priority, and anything specified in "objects" is ignored.
- *
- * b) If "found_objs" (the double pointer) is not NULL but "*found_objs"
- * (the once-dereferenced double pointer) _is_ NULL, the "objects"
- * parameter takes priority, and the results of the catalog query
- * described in (2) are stored in "found_objs".
- *
- * c) If "found_objs" (the double pointer) is NULL, the "objects"
- * parameter again takes priority, and the results of the catalog query
- * are not saved.
- *
- * 2) The "objects" parameter is a user-specified list of objects to process.
- * When (1b) or (1c) applies, this function performs a catalog query to
- * retrieve a fully qualified list of objects to process, as described
- * below.
- *
- * a) If "objects" is not NULL, the catalog query gathers only the objects
- * listed in "objects".
- *
- * b) If "objects" is NULL, all tables in the database are gathered.
- *
- * Note that this function is only concerned with running exactly one stage
- * when in analyze-in-stages mode; caller must iterate on us if necessary.
- *
- * If concurrentCons is > 1, multiple connections are used to vacuum tables
- * in parallel.
- */
-static void
-vacuum_one_database(ConnParams *cparams,
- vacuumingOptions *vacopts,
- int stage,
- SimpleStringList *objects,
- SimpleStringList **found_objs,
- int concurrentCons,
- const char *progname, bool echo, bool quiet)
-{
- PQExpBufferData sql;
- PGconn *conn;
- SimpleStringListCell *cell;
- ParallelSlotArray *sa;
- int ntups = 0;
- bool failed = false;
- const char *initcmd;
- SimpleStringList *ret = NULL;
- const char *stage_commands[] = {
- "SET default_statistics_target=1; SET vacuum_cost_delay=0;",
- "SET default_statistics_target=10; RESET vacuum_cost_delay;",
- "RESET default_statistics_target;"
- };
- const char *stage_messages[] = {
- gettext_noop("Generating minimal optimizer statistics (1 target)"),
- gettext_noop("Generating medium optimizer statistics (10 targets)"),
- gettext_noop("Generating default (full) optimizer statistics")
- };
-
- Assert(stage == ANALYZE_NO_STAGE ||
- (stage >= 0 && stage < ANALYZE_NUM_STAGES));
-
- conn = connectDatabase(cparams, progname, echo, false, true);
-
- if (vacopts->disable_page_skipping && PQserverVersion(conn) < 90600)
- {
- PQfinish(conn);
- pg_fatal("cannot use the \"%s\" option on server versions older than PostgreSQL %s",
- "disable-page-skipping", "9.6");
- }
-
- if (vacopts->no_index_cleanup && PQserverVersion(conn) < 120000)
- {
- PQfinish(conn);
- pg_fatal("cannot use the \"%s\" option on server versions older than PostgreSQL %s",
- "no-index-cleanup", "12");
- }
-
- if (vacopts->force_index_cleanup && PQserverVersion(conn) < 120000)
- {
- PQfinish(conn);
- pg_fatal("cannot use the \"%s\" option on server versions older than PostgreSQL %s",
- "force-index-cleanup", "12");
- }
-
- if (!vacopts->do_truncate && PQserverVersion(conn) < 120000)
- {
- PQfinish(conn);
- pg_fatal("cannot use the \"%s\" option on server versions older than PostgreSQL %s",
- "no-truncate", "12");
- }
-
- if (!vacopts->process_main && PQserverVersion(conn) < 160000)
- {
- PQfinish(conn);
- pg_fatal("cannot use the \"%s\" option on server versions older than PostgreSQL %s",
- "no-process-main", "16");
- }
-
- if (!vacopts->process_toast && PQserverVersion(conn) < 140000)
- {
- PQfinish(conn);
- pg_fatal("cannot use the \"%s\" option on server versions older than PostgreSQL %s",
- "no-process-toast", "14");
- }
-
- if (vacopts->skip_locked && PQserverVersion(conn) < 120000)
- {
- PQfinish(conn);
- pg_fatal("cannot use the \"%s\" option on server versions older than PostgreSQL %s",
- "skip-locked", "12");
- }
-
- if (vacopts->min_xid_age != 0 && PQserverVersion(conn) < 90600)
- {
- PQfinish(conn);
- pg_fatal("cannot use the \"%s\" option on server versions older than PostgreSQL %s",
- "--min-xid-age", "9.6");
- }
-
- if (vacopts->min_mxid_age != 0 && PQserverVersion(conn) < 90600)
- {
- PQfinish(conn);
- pg_fatal("cannot use the \"%s\" option on server versions older than PostgreSQL %s",
- "--min-mxid-age", "9.6");
- }
-
- if (vacopts->parallel_workers >= 0 && PQserverVersion(conn) < 130000)
- {
- PQfinish(conn);
- pg_fatal("cannot use the \"%s\" option on server versions older than PostgreSQL %s",
- "--parallel", "13");
- }
-
- if (vacopts->buffer_usage_limit && PQserverVersion(conn) < 160000)
- {
- PQfinish(conn);
- pg_fatal("cannot use the \"%s\" option on server versions older than PostgreSQL %s",
- "--buffer-usage-limit", "16");
- }
-
- if (vacopts->missing_stats_only && PQserverVersion(conn) < 150000)
- {
- PQfinish(conn);
- pg_fatal("cannot use the \"%s\" option on server versions older than PostgreSQL %s",
- "--missing-stats-only", "15");
- }
-
- /* skip_database_stats is used automatically if server supports it */
- vacopts->skip_database_stats = (PQserverVersion(conn) >= 160000);
-
- if (!quiet)
- {
- if (stage != ANALYZE_NO_STAGE)
- printf(_("%s: processing database \"%s\": %s\n"),
- progname, PQdb(conn), _(stage_messages[stage]));
- else
- printf(_("%s: vacuuming database \"%s\"\n"),
- progname, PQdb(conn));
- fflush(stdout);
- }
-
- /*
- * If the caller provided the results of a previous catalog query, just
- * use that. Otherwise, run the catalog query ourselves and set the
- * return variable if provided.
- */
- if (found_objs && *found_objs)
- ret = *found_objs;
- else
- {
- ret = retrieve_objects(conn, vacopts, objects, echo);
- if (found_objs)
- *found_objs = ret;
- }
-
- /*
- * Count the number of objects in the catalog query result. If there are
- * none, we are done.
- */
- for (cell = ret ? ret->head : NULL; cell; cell = cell->next)
- ntups++;
-
- if (ntups == 0)
- {
- PQfinish(conn);
- return;
- }
-
- /*
- * Ensure concurrentCons is sane. If there are more connections than
- * vacuumable relations, we don't need to use them all.
- */
- if (concurrentCons > ntups)
- concurrentCons = ntups;
- if (concurrentCons <= 0)
- concurrentCons = 1;
-
- /*
- * All slots need to be prepared to run the appropriate analyze stage, if
- * caller requested that mode. We have to prepare the initial connection
- * ourselves before setting up the slots.
- */
- if (stage == ANALYZE_NO_STAGE)
- initcmd = NULL;
- else
- {
- initcmd = stage_commands[stage];
- executeCommand(conn, initcmd, echo);
- }
-
- /*
- * Setup the database connections. We reuse the connection we already have
- * for the first slot. If not in parallel mode, the first slot in the
- * array contains the connection.
- */
- sa = ParallelSlotsSetup(concurrentCons, cparams, progname, echo, initcmd);
- ParallelSlotsAdoptConn(sa, conn);
-
- initPQExpBuffer(&sql);
-
- cell = ret->head;
- do
- {
- const char *tabname = cell->val;
- ParallelSlot *free_slot;
-
- if (CancelRequested)
- {
- failed = true;
- goto finish;
- }
-
- free_slot = ParallelSlotsGetIdle(sa, NULL);
- if (!free_slot)
- {
- failed = true;
- goto finish;
- }
-
- prepare_vacuum_command(&sql, PQserverVersion(free_slot->connection),
- vacopts, tabname);
-
- /*
- * Execute the vacuum. All errors are handled in processQueryResult
- * through ParallelSlotsGetIdle.
- */
- ParallelSlotSetHandler(free_slot, TableCommandResultHandler, NULL);
- run_vacuum_command(free_slot->connection, sql.data,
- echo, tabname);
-
- cell = cell->next;
- } while (cell != NULL);
-
- if (!ParallelSlotsWaitCompletion(sa))
- {
- failed = true;
- goto finish;
- }
-
- /* If we used SKIP_DATABASE_STATS, mop up with ONLY_DATABASE_STATS */
- if (vacopts->skip_database_stats && stage == ANALYZE_NO_STAGE)
- {
- const char *cmd = "VACUUM (ONLY_DATABASE_STATS);";
- ParallelSlot *free_slot = ParallelSlotsGetIdle(sa, NULL);
-
- if (!free_slot)
- {
- failed = true;
- goto finish;
- }
-
- ParallelSlotSetHandler(free_slot, TableCommandResultHandler, NULL);
- run_vacuum_command(free_slot->connection, cmd, echo, NULL);
-
- if (!ParallelSlotsWaitCompletion(sa))
- failed = true;
- }
-
-finish:
- ParallelSlotsTerminate(sa);
- pg_free(sa);
-
- termPQExpBuffer(&sql);
-
- if (failed)
- exit(1);
-}
-
-/*
- * Prepare the list of tables to process by querying the catalogs.
- *
- * Since we execute the constructed query with the default search_path (which
- * could be unsafe), everything in this query MUST be fully qualified.
- *
- * First, build a WITH clause for the catalog query if any tables were
- * specified, with a set of values made of relation names and their optional
- * set of columns. This is used to match any provided column lists with the
- * generated qualified identifiers and to filter for the tables provided via
- * --table. If a listed table does not exist, the catalog query will fail.
- */
-static SimpleStringList *
-retrieve_objects(PGconn *conn, vacuumingOptions *vacopts,
- SimpleStringList *objects, bool echo)
-{
- PQExpBufferData buf;
- PQExpBufferData catalog_query;
- PGresult *res;
- SimpleStringListCell *cell;
- SimpleStringList *found_objs = palloc0(sizeof(SimpleStringList));
- bool objects_listed = false;
-
- initPQExpBuffer(&catalog_query);
- for (cell = objects ? objects->head : NULL; cell; cell = cell->next)
- {
- char *just_table = NULL;
- const char *just_columns = NULL;
-
- if (!objects_listed)
- {
- appendPQExpBufferStr(&catalog_query,
- "WITH listed_objects (object_oid, column_list) "
- "AS (\n VALUES (");
- objects_listed = true;
- }
- else
- appendPQExpBufferStr(&catalog_query, ",\n (");
-
- if (objfilter & (OBJFILTER_SCHEMA | OBJFILTER_SCHEMA_EXCLUDE))
- {
- appendStringLiteralConn(&catalog_query, cell->val, conn);
- appendPQExpBufferStr(&catalog_query, "::pg_catalog.regnamespace, ");
- }
-
- if (objfilter & OBJFILTER_TABLE)
- {
- /*
- * Split relation and column names given by the user, this is used
- * to feed the CTE with values on which are performed pre-run
- * validity checks as well. For now these happen only on the
- * relation name.
- */
- splitTableColumnsSpec(cell->val, PQclientEncoding(conn),
- &just_table, &just_columns);
-
- appendStringLiteralConn(&catalog_query, just_table, conn);
- appendPQExpBufferStr(&catalog_query, "::pg_catalog.regclass, ");
- }
-
- if (just_columns && just_columns[0] != '\0')
- appendStringLiteralConn(&catalog_query, just_columns, conn);
- else
- appendPQExpBufferStr(&catalog_query, "NULL");
-
- appendPQExpBufferStr(&catalog_query, "::pg_catalog.text)");
-
- pg_free(just_table);
- }
-
- /* Finish formatting the CTE */
- if (objects_listed)
- appendPQExpBufferStr(&catalog_query, "\n)\n");
-
- appendPQExpBufferStr(&catalog_query, "SELECT c.relname, ns.nspname");
-
- if (objects_listed)
- appendPQExpBufferStr(&catalog_query, ", listed_objects.column_list");
-
- appendPQExpBufferStr(&catalog_query,
- " FROM pg_catalog.pg_class c\n"
- " JOIN pg_catalog.pg_namespace ns"
- " ON c.relnamespace OPERATOR(pg_catalog.=) ns.oid\n"
- " CROSS JOIN LATERAL (SELECT c.relkind IN ("
- CppAsString2(RELKIND_PARTITIONED_TABLE) ", "
- CppAsString2(RELKIND_PARTITIONED_INDEX) ")) as p (inherited)\n"
- " LEFT JOIN pg_catalog.pg_class t"
- " ON c.reltoastrelid OPERATOR(pg_catalog.=) t.oid\n");
-
- /*
- * Used to match the tables or schemas listed by the user, completing the
- * JOIN clause.
- */
- if (objects_listed)
- {
- appendPQExpBufferStr(&catalog_query, " LEFT JOIN listed_objects"
- " ON listed_objects.object_oid"
- " OPERATOR(pg_catalog.=) ");
-
- if (objfilter & OBJFILTER_TABLE)
- appendPQExpBufferStr(&catalog_query, "c.oid\n");
- else
- appendPQExpBufferStr(&catalog_query, "ns.oid\n");
- }
-
- /*
- * Exclude temporary tables, beginning the WHERE clause.
- */
- appendPQExpBufferStr(&catalog_query,
- " WHERE c.relpersistence OPERATOR(pg_catalog.!=) "
- CppAsString2(RELPERSISTENCE_TEMP) "\n");
-
- /*
- * Used to match the tables or schemas listed by the user, for the WHERE
- * clause.
- */
- if (objects_listed)
- {
- if (objfilter & OBJFILTER_SCHEMA_EXCLUDE)
- appendPQExpBufferStr(&catalog_query,
- " AND listed_objects.object_oid IS NULL\n");
- else
- appendPQExpBufferStr(&catalog_query,
- " AND listed_objects.object_oid IS NOT NULL\n");
- }
-
- /*
- * If no tables were listed, filter for the relevant relation types. If
- * tables were given via --table, don't bother filtering by relation type.
- * Instead, let the server decide whether a given relation can be
- * processed in which case the user will know about it.
- */
- if ((objfilter & OBJFILTER_TABLE) == 0)
- {
- /*
- * vacuumdb should generally follow the behavior of the underlying
- * VACUUM and ANALYZE commands. If analyze_only is true, process
- * regular tables, materialized views, and partitioned tables, just
- * like ANALYZE (with no specific target tables) does. Otherwise,
- * process only regular tables and materialized views, since VACUUM
- * skips partitioned tables when no target tables are specified.
- */
- if (vacopts->analyze_only)
- appendPQExpBufferStr(&catalog_query,
- " AND c.relkind OPERATOR(pg_catalog.=) ANY (array["
- CppAsString2(RELKIND_RELATION) ", "
- CppAsString2(RELKIND_MATVIEW) ", "
- CppAsString2(RELKIND_PARTITIONED_TABLE) "])\n");
- else
- appendPQExpBufferStr(&catalog_query,
- " AND c.relkind OPERATOR(pg_catalog.=) ANY (array["
- CppAsString2(RELKIND_RELATION) ", "
- CppAsString2(RELKIND_MATVIEW) "])\n");
-
- }
-
- /*
- * For --min-xid-age and --min-mxid-age, the age of the relation is the
- * greatest of the ages of the main relation and its associated TOAST
- * table. The commands generated by vacuumdb will also process the TOAST
- * table for the relation if necessary, so it does not need to be
- * considered separately.
- */
- if (vacopts->min_xid_age != 0)
- {
- appendPQExpBuffer(&catalog_query,
- " AND GREATEST(pg_catalog.age(c.relfrozenxid),"
- " pg_catalog.age(t.relfrozenxid)) "
- " OPERATOR(pg_catalog.>=) '%d'::pg_catalog.int4\n"
- " AND c.relfrozenxid OPERATOR(pg_catalog.!=)"
- " '0'::pg_catalog.xid\n",
- vacopts->min_xid_age);
- }
-
- if (vacopts->min_mxid_age != 0)
- {
- appendPQExpBuffer(&catalog_query,
- " AND GREATEST(pg_catalog.mxid_age(c.relminmxid),"
- " pg_catalog.mxid_age(t.relminmxid)) OPERATOR(pg_catalog.>=)"
- " '%d'::pg_catalog.int4\n"
- " AND c.relminmxid OPERATOR(pg_catalog.!=)"
- " '0'::pg_catalog.xid\n",
- vacopts->min_mxid_age);
- }
-
- if (vacopts->missing_stats_only)
- {
- appendPQExpBufferStr(&catalog_query, " AND (\n");
-
- /* regular stats */
- appendPQExpBufferStr(&catalog_query,
- " EXISTS (SELECT NULL FROM pg_catalog.pg_attribute a\n"
- " WHERE a.attrelid OPERATOR(pg_catalog.=) c.oid\n"
- " AND a.attnum OPERATOR(pg_catalog.>) 0::pg_catalog.int2\n"
- " AND NOT a.attisdropped\n"
- " AND a.attstattarget IS DISTINCT FROM 0::pg_catalog.int2\n"
- " AND a.attgenerated OPERATOR(pg_catalog.<>) "
- CppAsString2(ATTRIBUTE_GENERATED_VIRTUAL) "\n"
- " AND NOT EXISTS (SELECT NULL FROM pg_catalog.pg_statistic s\n"
- " WHERE s.starelid OPERATOR(pg_catalog.=) a.attrelid\n"
- " AND s.staattnum OPERATOR(pg_catalog.=) a.attnum\n"
- " AND s.stainherit OPERATOR(pg_catalog.=) p.inherited))\n");
-
- /* extended stats */
- appendPQExpBufferStr(&catalog_query,
- " OR EXISTS (SELECT NULL FROM pg_catalog.pg_statistic_ext e\n"
- " WHERE e.stxrelid OPERATOR(pg_catalog.=) c.oid\n"
- " AND e.stxstattarget IS DISTINCT FROM 0::pg_catalog.int2\n"
- " AND NOT EXISTS (SELECT NULL FROM pg_catalog.pg_statistic_ext_data d\n"
- " WHERE d.stxoid OPERATOR(pg_catalog.=) e.oid\n"
- " AND d.stxdinherit OPERATOR(pg_catalog.=) p.inherited))\n");
-
- /* expression indexes */
- appendPQExpBufferStr(&catalog_query,
- " OR EXISTS (SELECT NULL FROM pg_catalog.pg_attribute a\n"
- " JOIN pg_catalog.pg_index i"
- " ON i.indexrelid OPERATOR(pg_catalog.=) a.attrelid\n"
- " WHERE i.indrelid OPERATOR(pg_catalog.=) c.oid\n"
- " AND i.indkey[a.attnum OPERATOR(pg_catalog.-) 1::pg_catalog.int2]"
- " OPERATOR(pg_catalog.=) 0::pg_catalog.int2\n"
- " AND a.attnum OPERATOR(pg_catalog.>) 0::pg_catalog.int2\n"
- " AND NOT a.attisdropped\n"
- " AND a.attstattarget IS DISTINCT FROM 0::pg_catalog.int2\n"
- " AND NOT EXISTS (SELECT NULL FROM pg_catalog.pg_statistic s\n"
- " WHERE s.starelid OPERATOR(pg_catalog.=) a.attrelid\n"
- " AND s.staattnum OPERATOR(pg_catalog.=) a.attnum\n"
- " AND s.stainherit OPERATOR(pg_catalog.=) p.inherited))\n");
-
- /* inheritance and regular stats */
- appendPQExpBufferStr(&catalog_query,
- " OR EXISTS (SELECT NULL FROM pg_catalog.pg_attribute a\n"
- " WHERE a.attrelid OPERATOR(pg_catalog.=) c.oid\n"
- " AND a.attnum OPERATOR(pg_catalog.>) 0::pg_catalog.int2\n"
- " AND NOT a.attisdropped\n"
- " AND a.attstattarget IS DISTINCT FROM 0::pg_catalog.int2\n"
- " AND a.attgenerated OPERATOR(pg_catalog.<>) "
- CppAsString2(ATTRIBUTE_GENERATED_VIRTUAL) "\n"
- " AND c.relhassubclass\n"
- " AND NOT p.inherited\n"
- " AND EXISTS (SELECT NULL FROM pg_catalog.pg_inherits h\n"
- " WHERE h.inhparent OPERATOR(pg_catalog.=) c.oid)\n"
- " AND NOT EXISTS (SELECT NULL FROM pg_catalog.pg_statistic s\n"
- " WHERE s.starelid OPERATOR(pg_catalog.=) a.attrelid\n"
- " AND s.staattnum OPERATOR(pg_catalog.=) a.attnum\n"
- " AND s.stainherit))\n");
-
- /* inheritance and extended stats */
- appendPQExpBufferStr(&catalog_query,
- " OR EXISTS (SELECT NULL FROM pg_catalog.pg_statistic_ext e\n"
- " WHERE e.stxrelid OPERATOR(pg_catalog.=) c.oid\n"
- " AND e.stxstattarget IS DISTINCT FROM 0::pg_catalog.int2\n"
- " AND c.relhassubclass\n"
- " AND NOT p.inherited\n"
- " AND EXISTS (SELECT NULL FROM pg_catalog.pg_inherits h\n"
- " WHERE h.inhparent OPERATOR(pg_catalog.=) c.oid)\n"
- " AND NOT EXISTS (SELECT NULL FROM pg_catalog.pg_statistic_ext_data d\n"
- " WHERE d.stxoid OPERATOR(pg_catalog.=) e.oid\n"
- " AND d.stxdinherit))\n");
-
- appendPQExpBufferStr(&catalog_query, " )\n");
- }
-
- /*
- * Execute the catalog query. We use the default search_path for this
- * query for consistency with table lookups done elsewhere by the user.
- */
- appendPQExpBufferStr(&catalog_query, " ORDER BY c.relpages DESC;");
- executeCommand(conn, "RESET search_path;", echo);
- res = executeQuery(conn, catalog_query.data, echo);
- termPQExpBuffer(&catalog_query);
- PQclear(executeQuery(conn, ALWAYS_SECURE_SEARCH_PATH_SQL, echo));
-
- /*
- * Build qualified identifiers for each table, including the column list
- * if given.
- */
- initPQExpBuffer(&buf);
- for (int i = 0; i < PQntuples(res); i++)
- {
- appendPQExpBufferStr(&buf,
- fmtQualifiedIdEnc(PQgetvalue(res, i, 1),
- PQgetvalue(res, i, 0),
- PQclientEncoding(conn)));
-
- if (objects_listed && !PQgetisnull(res, i, 2))
- appendPQExpBufferStr(&buf, PQgetvalue(res, i, 2));
-
- simple_string_list_append(found_objs, buf.data);
- resetPQExpBuffer(&buf);
- }
- termPQExpBuffer(&buf);
- PQclear(res);
-
- return found_objs;
-}
-
-/*
- * Vacuum/analyze all connectable databases.
- *
- * In analyze-in-stages mode, we process all databases in one stage before
- * moving on to the next stage. That ensure minimal stats are available
- * quickly everywhere before generating more detailed ones.
- */
-static void
-vacuum_all_databases(ConnParams *cparams,
- vacuumingOptions *vacopts,
- bool analyze_in_stages,
- SimpleStringList *objects,
- int concurrentCons,
- const char *progname, bool echo, bool quiet)
-{
- PGconn *conn;
- PGresult *result;
- int stage;
- int i;
-
- conn = connectMaintenanceDatabase(cparams, progname, echo);
- result = executeQuery(conn,
- "SELECT datname FROM pg_database WHERE datallowconn AND datconnlimit <> -2 ORDER BY 1;",
- echo);
- PQfinish(conn);
-
- if (analyze_in_stages)
- {
- SimpleStringList **found_objs = NULL;
-
- if (vacopts->missing_stats_only)
- found_objs = palloc0(PQntuples(result) * sizeof(SimpleStringList *));
-
- /*
- * When analyzing all databases in stages, we analyze them all in the
- * fastest stage first, so that initial statistics become available
- * for all of them as soon as possible.
- *
- * This means we establish several times as many connections, but
- * that's a secondary consideration.
- */
- for (stage = 0; stage < ANALYZE_NUM_STAGES; stage++)
- {
- for (i = 0; i < PQntuples(result); i++)
- {
- cparams->override_dbname = PQgetvalue(result, i, 0);
-
- vacuum_one_database(cparams, vacopts,
- stage,
- objects,
- vacopts->missing_stats_only ? &found_objs[i] : NULL,
- concurrentCons,
- progname, echo, quiet);
- }
- }
- }
- else
- {
- for (i = 0; i < PQntuples(result); i++)
- {
- cparams->override_dbname = PQgetvalue(result, i, 0);
-
- vacuum_one_database(cparams, vacopts,
- ANALYZE_NO_STAGE,
- objects, NULL,
- concurrentCons,
- progname, echo, quiet);
- }
- }
-
- PQclear(result);
-}
-
-/*
- * Construct a vacuum/analyze command to run based on the given options, in the
- * given string buffer, which may contain previous garbage.
- *
- * The table name used must be already properly quoted. The command generated
- * depends on the server version involved and it is semicolon-terminated.
- */
-static void
-prepare_vacuum_command(PQExpBuffer sql, int serverVersion,
- vacuumingOptions *vacopts, const char *table)
-{
- const char *paren = " (";
- const char *comma = ", ";
- const char *sep = paren;
-
- resetPQExpBuffer(sql);
-
- if (vacopts->analyze_only)
- {
- appendPQExpBufferStr(sql, "ANALYZE");
-
- /* parenthesized grammar of ANALYZE is supported since v11 */
- if (serverVersion >= 110000)
- {
- if (vacopts->skip_locked)
- {
- /* SKIP_LOCKED is supported since v12 */
- Assert(serverVersion >= 120000);
- appendPQExpBuffer(sql, "%sSKIP_LOCKED", sep);
- sep = comma;
- }
- if (vacopts->verbose)
- {
- appendPQExpBuffer(sql, "%sVERBOSE", sep);
- sep = comma;
- }
- if (vacopts->buffer_usage_limit)
- {
- Assert(serverVersion >= 160000);
- appendPQExpBuffer(sql, "%sBUFFER_USAGE_LIMIT '%s'", sep,
- vacopts->buffer_usage_limit);
- sep = comma;
- }
- if (sep != paren)
- appendPQExpBufferChar(sql, ')');
- }
- else
- {
- if (vacopts->verbose)
- appendPQExpBufferStr(sql, " VERBOSE");
- }
- }
- else
- {
- appendPQExpBufferStr(sql, "VACUUM");
-
- /* parenthesized grammar of VACUUM is supported since v9.0 */
- if (serverVersion >= 90000)
- {
- if (vacopts->disable_page_skipping)
- {
- /* DISABLE_PAGE_SKIPPING is supported since v9.6 */
- Assert(serverVersion >= 90600);
- appendPQExpBuffer(sql, "%sDISABLE_PAGE_SKIPPING", sep);
- sep = comma;
- }
- if (vacopts->no_index_cleanup)
- {
- /* "INDEX_CLEANUP FALSE" has been supported since v12 */
- Assert(serverVersion >= 120000);
- Assert(!vacopts->force_index_cleanup);
- appendPQExpBuffer(sql, "%sINDEX_CLEANUP FALSE", sep);
- sep = comma;
- }
- if (vacopts->force_index_cleanup)
- {
- /* "INDEX_CLEANUP TRUE" has been supported since v12 */
- Assert(serverVersion >= 120000);
- Assert(!vacopts->no_index_cleanup);
- appendPQExpBuffer(sql, "%sINDEX_CLEANUP TRUE", sep);
- sep = comma;
- }
- if (!vacopts->do_truncate)
- {
- /* TRUNCATE is supported since v12 */
- Assert(serverVersion >= 120000);
- appendPQExpBuffer(sql, "%sTRUNCATE FALSE", sep);
- sep = comma;
- }
- if (!vacopts->process_main)
- {
- /* PROCESS_MAIN is supported since v16 */
- Assert(serverVersion >= 160000);
- appendPQExpBuffer(sql, "%sPROCESS_MAIN FALSE", sep);
- sep = comma;
- }
- if (!vacopts->process_toast)
- {
- /* PROCESS_TOAST is supported since v14 */
- Assert(serverVersion >= 140000);
- appendPQExpBuffer(sql, "%sPROCESS_TOAST FALSE", sep);
- sep = comma;
- }
- if (vacopts->skip_database_stats)
- {
- /* SKIP_DATABASE_STATS is supported since v16 */
- Assert(serverVersion >= 160000);
- appendPQExpBuffer(sql, "%sSKIP_DATABASE_STATS", sep);
- sep = comma;
- }
- if (vacopts->skip_locked)
- {
- /* SKIP_LOCKED is supported since v12 */
- Assert(serverVersion >= 120000);
- appendPQExpBuffer(sql, "%sSKIP_LOCKED", sep);
- sep = comma;
- }
- if (vacopts->full)
- {
- appendPQExpBuffer(sql, "%sFULL", sep);
- sep = comma;
- }
- if (vacopts->freeze)
- {
- appendPQExpBuffer(sql, "%sFREEZE", sep);
- sep = comma;
- }
- if (vacopts->verbose)
- {
- appendPQExpBuffer(sql, "%sVERBOSE", sep);
- sep = comma;
- }
- if (vacopts->and_analyze)
- {
- appendPQExpBuffer(sql, "%sANALYZE", sep);
- sep = comma;
- }
- if (vacopts->parallel_workers >= 0)
- {
- /* PARALLEL is supported since v13 */
- Assert(serverVersion >= 130000);
- appendPQExpBuffer(sql, "%sPARALLEL %d", sep,
- vacopts->parallel_workers);
- sep = comma;
- }
- if (vacopts->buffer_usage_limit)
- {
- Assert(serverVersion >= 160000);
- appendPQExpBuffer(sql, "%sBUFFER_USAGE_LIMIT '%s'", sep,
- vacopts->buffer_usage_limit);
- sep = comma;
- }
- if (sep != paren)
- appendPQExpBufferChar(sql, ')');
- }
- else
- {
- if (vacopts->full)
- appendPQExpBufferStr(sql, " FULL");
- if (vacopts->freeze)
- appendPQExpBufferStr(sql, " FREEZE");
- if (vacopts->verbose)
- appendPQExpBufferStr(sql, " VERBOSE");
- if (vacopts->and_analyze)
- appendPQExpBufferStr(sql, " ANALYZE");
- }
- }
-
- appendPQExpBuffer(sql, " %s;", table);
-}
-
-/*
- * Send a vacuum/analyze command to the server, returning after sending the
- * command.
- *
- * Any errors during command execution are reported to stderr.
- */
-static void
-run_vacuum_command(PGconn *conn, const char *sql, bool echo,
- const char *table)
-{
- bool status;
-
- if (echo)
- printf("%s\n", sql);
-
- status = PQsendQuery(conn, sql) == 1;
-
- if (!status)
- {
- if (table)
- pg_log_error("vacuuming of table \"%s\" in database \"%s\" failed: %s",
- table, PQdb(conn), PQerrorMessage(conn));
- else
- pg_log_error("vacuuming of database \"%s\" failed: %s",
- PQdb(conn), PQerrorMessage(conn));
- }
-}
static void
help(const char *progname)
diff --git a/src/bin/scripts/vacuuming.c b/src/bin/scripts/vacuuming.c
new file mode 100644
index 00000000000..9be37fcc45a
--- /dev/null
+++ b/src/bin/scripts/vacuuming.c
@@ -0,0 +1,978 @@
+/*-------------------------------------------------------------------------
+ * vacuuming.c
+ * Common routines for vacuumdb
+ *
+ * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group
+ * Portions Copyright (c) 1994, Regents of the University of California
+ *
+ * src/bin/scripts/vacuuming.c
+ *
+ *-------------------------------------------------------------------------
+ */
+
+#include "postgres_fe.h"
+
+#include <limits.h>
+
+#include "catalog/pg_attribute_d.h"
+#include "catalog/pg_class_d.h"
+#include "common/connect.h"
+#include "common/logging.h"
+#include "fe_utils/cancel.h"
+#include "fe_utils/option_utils.h"
+#include "fe_utils/parallel_slot.h"
+#include "fe_utils/query_utils.h"
+#include "fe_utils/string_utils.h"
+#include "vacuuming.h"
+
+VacObjFilter objfilter = OBJFILTER_NONE;
+
+
+/*
+ * Executes vacuum/analyze as indicated, or dies in case of failure.
+ */
+void
+vacuuming_main(ConnParams *cparams, const char *dbname,
+ const char *maintenance_db, vacuumingOptions *vacopts,
+ SimpleStringList *objects, bool analyze_in_stages,
+ int tbl_count, int concurrentCons,
+ const char *progname, bool echo, bool quiet)
+{
+ setup_cancel_handler(NULL);
+
+ /* Avoid opening extra connections. */
+ if (tbl_count && (concurrentCons > tbl_count))
+ concurrentCons = tbl_count;
+
+ if (objfilter & OBJFILTER_ALL_DBS)
+ {
+ cparams->dbname = maintenance_db;
+
+ vacuum_all_databases(cparams, vacopts,
+ analyze_in_stages,
+ objects,
+ concurrentCons,
+ progname, echo, quiet);
+ }
+ else
+ {
+ if (dbname == NULL)
+ {
+ if (getenv("PGDATABASE"))
+ dbname = getenv("PGDATABASE");
+ else if (getenv("PGUSER"))
+ dbname = getenv("PGUSER");
+ else
+ dbname = get_user_name_or_exit(progname);
+ }
+
+ cparams->dbname = dbname;
+
+ if (analyze_in_stages)
+ {
+ int stage;
+ SimpleStringList *found_objs = NULL;
+
+ for (stage = 0; stage < ANALYZE_NUM_STAGES; stage++)
+ {
+ vacuum_one_database(cparams, vacopts,
+ stage,
+ objects,
+ vacopts->missing_stats_only ? &found_objs : NULL,
+ concurrentCons,
+ progname, echo, quiet);
+ }
+ }
+ else
+ vacuum_one_database(cparams, vacopts,
+ ANALYZE_NO_STAGE,
+ objects, NULL,
+ concurrentCons,
+ progname, echo, quiet);
+ }
+}
+
+
+/*
+ * vacuum_one_database
+ *
+ * Process tables in the given database.
+ *
+ * There are two ways to specify the list of objects to process:
+ *
+ * 1) The "found_objs" parameter is a double pointer to a fully qualified list
+ * of objects to process, as returned by a previous call to
+ * vacuum_one_database().
+ *
+ * a) If both "found_objs" (the double pointer) and "*found_objs" (the
+ * once-dereferenced double pointer) are not NULL, this list takes
+ * priority, and anything specified in "objects" is ignored.
+ *
+ * b) If "found_objs" (the double pointer) is not NULL but "*found_objs"
+ * (the once-dereferenced double pointer) _is_ NULL, the "objects"
+ * parameter takes priority, and the results of the catalog query
+ * described in (2) are stored in "found_objs".
+ *
+ * c) If "found_objs" (the double pointer) is NULL, the "objects"
+ * parameter again takes priority, and the results of the catalog query
+ * are not saved.
+ *
+ * 2) The "objects" parameter is a user-specified list of objects to process.
+ * When (1b) or (1c) applies, this function performs a catalog query to
+ * retrieve a fully qualified list of objects to process, as described
+ * below.
+ *
+ * a) If "objects" is not NULL, the catalog query gathers only the objects
+ * listed in "objects".
+ *
+ * b) If "objects" is NULL, all tables in the database are gathered.
+ *
+ * Note that this function is only concerned with running exactly one stage
+ * when in analyze-in-stages mode; caller must iterate on us if necessary.
+ *
+ * If concurrentCons is > 1, multiple connections are used to vacuum tables
+ * in parallel.
+ */
+void
+vacuum_one_database(ConnParams *cparams,
+ vacuumingOptions *vacopts,
+ int stage,
+ SimpleStringList *objects,
+ SimpleStringList **found_objs,
+ int concurrentCons,
+ const char *progname, bool echo, bool quiet)
+{
+ PQExpBufferData sql;
+ PGconn *conn;
+ SimpleStringListCell *cell;
+ ParallelSlotArray *sa;
+ int ntups = 0;
+ bool failed = false;
+ const char *initcmd;
+ SimpleStringList *ret = NULL;
+ const char *stage_commands[] = {
+ "SET default_statistics_target=1; SET vacuum_cost_delay=0;",
+ "SET default_statistics_target=10; RESET vacuum_cost_delay;",
+ "RESET default_statistics_target;"
+ };
+ const char *stage_messages[] = {
+ gettext_noop("Generating minimal optimizer statistics (1 target)"),
+ gettext_noop("Generating medium optimizer statistics (10 targets)"),
+ gettext_noop("Generating default (full) optimizer statistics")
+ };
+
+ Assert(stage == ANALYZE_NO_STAGE ||
+ (stage >= 0 && stage < ANALYZE_NUM_STAGES));
+
+ conn = connectDatabase(cparams, progname, echo, false, true);
+
+ if (vacopts->disable_page_skipping && PQserverVersion(conn) < 90600)
+ {
+ PQfinish(conn);
+ pg_fatal("cannot use the \"%s\" option on server versions older than PostgreSQL %s",
+ "disable-page-skipping", "9.6");
+ }
+
+ if (vacopts->no_index_cleanup && PQserverVersion(conn) < 120000)
+ {
+ PQfinish(conn);
+ pg_fatal("cannot use the \"%s\" option on server versions older than PostgreSQL %s",
+ "no-index-cleanup", "12");
+ }
+
+ if (vacopts->force_index_cleanup && PQserverVersion(conn) < 120000)
+ {
+ PQfinish(conn);
+ pg_fatal("cannot use the \"%s\" option on server versions older than PostgreSQL %s",
+ "force-index-cleanup", "12");
+ }
+
+ if (!vacopts->do_truncate && PQserverVersion(conn) < 120000)
+ {
+ PQfinish(conn);
+ pg_fatal("cannot use the \"%s\" option on server versions older than PostgreSQL %s",
+ "no-truncate", "12");
+ }
+
+ if (!vacopts->process_main && PQserverVersion(conn) < 160000)
+ {
+ PQfinish(conn);
+ pg_fatal("cannot use the \"%s\" option on server versions older than PostgreSQL %s",
+ "no-process-main", "16");
+ }
+
+ if (!vacopts->process_toast && PQserverVersion(conn) < 140000)
+ {
+ PQfinish(conn);
+ pg_fatal("cannot use the \"%s\" option on server versions older than PostgreSQL %s",
+ "no-process-toast", "14");
+ }
+
+ if (vacopts->skip_locked && PQserverVersion(conn) < 120000)
+ {
+ PQfinish(conn);
+ pg_fatal("cannot use the \"%s\" option on server versions older than PostgreSQL %s",
+ "skip-locked", "12");
+ }
+
+ if (vacopts->min_xid_age != 0 && PQserverVersion(conn) < 90600)
+ {
+ PQfinish(conn);
+ pg_fatal("cannot use the \"%s\" option on server versions older than PostgreSQL %s",
+ "--min-xid-age", "9.6");
+ }
+
+ if (vacopts->min_mxid_age != 0 && PQserverVersion(conn) < 90600)
+ {
+ PQfinish(conn);
+ pg_fatal("cannot use the \"%s\" option on server versions older than PostgreSQL %s",
+ "--min-mxid-age", "9.6");
+ }
+
+ if (vacopts->parallel_workers >= 0 && PQserverVersion(conn) < 130000)
+ {
+ PQfinish(conn);
+ pg_fatal("cannot use the \"%s\" option on server versions older than PostgreSQL %s",
+ "--parallel", "13");
+ }
+
+ if (vacopts->buffer_usage_limit && PQserverVersion(conn) < 160000)
+ {
+ PQfinish(conn);
+ pg_fatal("cannot use the \"%s\" option on server versions older than PostgreSQL %s",
+ "--buffer-usage-limit", "16");
+ }
+
+ if (vacopts->missing_stats_only && PQserverVersion(conn) < 150000)
+ {
+ PQfinish(conn);
+ pg_fatal("cannot use the \"%s\" option on server versions older than PostgreSQL %s",
+ "--missing-stats-only", "15");
+ }
+
+ /* skip_database_stats is used automatically if server supports it */
+ vacopts->skip_database_stats = (PQserverVersion(conn) >= 160000);
+
+ if (!quiet)
+ {
+ if (stage != ANALYZE_NO_STAGE)
+ printf(_("%s: processing database \"%s\": %s\n"),
+ progname, PQdb(conn), _(stage_messages[stage]));
+ else
+ printf(_("%s: vacuuming database \"%s\"\n"),
+ progname, PQdb(conn));
+ fflush(stdout);
+ }
+
+ /*
+ * If the caller provided the results of a previous catalog query, just
+ * use that. Otherwise, run the catalog query ourselves and set the
+ * return variable if provided.
+ */
+ if (found_objs && *found_objs)
+ ret = *found_objs;
+ else
+ {
+ ret = retrieve_objects(conn, vacopts, objects, echo);
+ if (found_objs)
+ *found_objs = ret;
+ }
+
+ /*
+ * Count the number of objects in the catalog query result. If there are
+ * none, we are done.
+ */
+ for (cell = ret ? ret->head : NULL; cell; cell = cell->next)
+ ntups++;
+
+ if (ntups == 0)
+ {
+ PQfinish(conn);
+ return;
+ }
+
+ /*
+ * Ensure concurrentCons is sane. If there are more connections than
+ * vacuumable relations, we don't need to use them all.
+ */
+ if (concurrentCons > ntups)
+ concurrentCons = ntups;
+ if (concurrentCons <= 0)
+ concurrentCons = 1;
+
+ /*
+ * All slots need to be prepared to run the appropriate analyze stage, if
+ * caller requested that mode. We have to prepare the initial connection
+ * ourselves before setting up the slots.
+ */
+ if (stage == ANALYZE_NO_STAGE)
+ initcmd = NULL;
+ else
+ {
+ initcmd = stage_commands[stage];
+ executeCommand(conn, initcmd, echo);
+ }
+
+ /*
+ * Setup the database connections. We reuse the connection we already have
+ * for the first slot. If not in parallel mode, the first slot in the
+ * array contains the connection.
+ */
+ sa = ParallelSlotsSetup(concurrentCons, cparams, progname, echo, initcmd);
+ ParallelSlotsAdoptConn(sa, conn);
+
+ initPQExpBuffer(&sql);
+
+ cell = ret->head;
+ do
+ {
+ const char *tabname = cell->val;
+ ParallelSlot *free_slot;
+
+ if (CancelRequested)
+ {
+ failed = true;
+ goto finish;
+ }
+
+ free_slot = ParallelSlotsGetIdle(sa, NULL);
+ if (!free_slot)
+ {
+ failed = true;
+ goto finish;
+ }
+
+ prepare_vacuum_command(&sql, PQserverVersion(free_slot->connection),
+ vacopts, tabname);
+
+ /*
+ * Execute the vacuum. All errors are handled in processQueryResult
+ * through ParallelSlotsGetIdle.
+ */
+ ParallelSlotSetHandler(free_slot, TableCommandResultHandler, NULL);
+ run_vacuum_command(free_slot->connection, sql.data,
+ echo, tabname);
+
+ cell = cell->next;
+ } while (cell != NULL);
+
+ if (!ParallelSlotsWaitCompletion(sa))
+ {
+ failed = true;
+ goto finish;
+ }
+
+ /* If we used SKIP_DATABASE_STATS, mop up with ONLY_DATABASE_STATS */
+ if (vacopts->skip_database_stats &&
+ stage == ANALYZE_NO_STAGE)
+ {
+ const char *cmd = "VACUUM (ONLY_DATABASE_STATS);";
+ ParallelSlot *free_slot = ParallelSlotsGetIdle(sa, NULL);
+
+ if (!free_slot)
+ {
+ failed = true;
+ goto finish;
+ }
+
+ ParallelSlotSetHandler(free_slot, TableCommandResultHandler, NULL);
+ run_vacuum_command(free_slot->connection, cmd, echo, NULL);
+
+ if (!ParallelSlotsWaitCompletion(sa))
+ failed = true;
+ }
+
+finish:
+ ParallelSlotsTerminate(sa);
+ pg_free(sa);
+
+ termPQExpBuffer(&sql);
+
+ if (failed)
+ exit(1);
+}
+
+/*
+ * Prepare the list of tables to process by querying the catalogs.
+ *
+ * Since we execute the constructed query with the default search_path (which
+ * could be unsafe), everything in this query MUST be fully qualified.
+ *
+ * First, build a WITH clause for the catalog query if any tables were
+ * specified, with a set of values made of relation names and their optional
+ * set of columns. This is used to match any provided column lists with the
+ * generated qualified identifiers and to filter for the tables provided via
+ * --table. If a listed table does not exist, the catalog query will fail.
+ */
+SimpleStringList *
+retrieve_objects(PGconn *conn, vacuumingOptions *vacopts,
+ SimpleStringList *objects, bool echo)
+{
+ PQExpBufferData buf;
+ PQExpBufferData catalog_query;
+ PGresult *res;
+ SimpleStringListCell *cell;
+ SimpleStringList *found_objs = palloc0(sizeof(SimpleStringList));
+ bool objects_listed = false;
+
+ initPQExpBuffer(&catalog_query);
+ for (cell = objects ? objects->head : NULL; cell; cell = cell->next)
+ {
+ char *just_table = NULL;
+ const char *just_columns = NULL;
+
+ if (!objects_listed)
+ {
+ appendPQExpBufferStr(&catalog_query,
+ "WITH listed_objects (object_oid, column_list) AS (\n"
+ " VALUES (");
+ objects_listed = true;
+ }
+ else
+ appendPQExpBufferStr(&catalog_query, ",\n (");
+
+ if (objfilter & (OBJFILTER_SCHEMA | OBJFILTER_SCHEMA_EXCLUDE))
+ {
+ appendStringLiteralConn(&catalog_query, cell->val, conn);
+ appendPQExpBufferStr(&catalog_query, "::pg_catalog.regnamespace, ");
+ }
+
+ if (objfilter & OBJFILTER_TABLE)
+ {
+ /*
+ * Split relation and column names given by the user, this is used
+ * to feed the CTE with values on which are performed pre-run
+ * validity checks as well. For now these happen only on the
+ * relation name.
+ */
+ splitTableColumnsSpec(cell->val, PQclientEncoding(conn),
+ &just_table, &just_columns);
+
+ appendStringLiteralConn(&catalog_query, just_table, conn);
+ appendPQExpBufferStr(&catalog_query, "::pg_catalog.regclass, ");
+ }
+
+ if (just_columns && just_columns[0] != '\0')
+ appendStringLiteralConn(&catalog_query, just_columns, conn);
+ else
+ appendPQExpBufferStr(&catalog_query, "NULL");
+
+ appendPQExpBufferStr(&catalog_query, "::pg_catalog.text)");
+
+ pg_free(just_table);
+ }
+
+ /* Finish formatting the CTE */
+ if (objects_listed)
+ appendPQExpBufferStr(&catalog_query, "\n)\n");
+
+ appendPQExpBufferStr(&catalog_query, "SELECT c.relname, ns.nspname");
+
+ if (objects_listed)
+ appendPQExpBufferStr(&catalog_query, ", listed_objects.column_list");
+
+ appendPQExpBufferStr(&catalog_query,
+ " FROM pg_catalog.pg_class c\n"
+ " JOIN pg_catalog.pg_namespace ns"
+ " ON c.relnamespace OPERATOR(pg_catalog.=) ns.oid\n"
+ " CROSS JOIN LATERAL (SELECT c.relkind IN ("
+ CppAsString2(RELKIND_PARTITIONED_TABLE) ", "
+ CppAsString2(RELKIND_PARTITIONED_INDEX) ")) as p (inherited)\n"
+ " LEFT JOIN pg_catalog.pg_class t"
+ " ON c.reltoastrelid OPERATOR(pg_catalog.=) t.oid\n");
+
+ /*
+ * Used to match the tables or schemas listed by the user, completing the
+ * JOIN clause.
+ */
+ if (objects_listed)
+ {
+ appendPQExpBufferStr(&catalog_query, " LEFT JOIN listed_objects"
+ " ON listed_objects.object_oid"
+ " OPERATOR(pg_catalog.=) ");
+
+ if (objfilter & OBJFILTER_TABLE)
+ appendPQExpBufferStr(&catalog_query, "c.oid\n");
+ else
+ appendPQExpBufferStr(&catalog_query, "ns.oid\n");
+ }
+
+ /*
+ * Exclude temporary tables, beginning the WHERE clause.
+ */
+ appendPQExpBufferStr(&catalog_query,
+ " WHERE c.relpersistence OPERATOR(pg_catalog.!=) "
+ CppAsString2(RELPERSISTENCE_TEMP) "\n");
+
+ /*
+ * Used to match the tables or schemas listed by the user, for the WHERE
+ * clause.
+ */
+ if (objects_listed)
+ {
+ if (objfilter & OBJFILTER_SCHEMA_EXCLUDE)
+ appendPQExpBufferStr(&catalog_query,
+ " AND listed_objects.object_oid IS NULL\n");
+ else
+ appendPQExpBufferStr(&catalog_query,
+ " AND listed_objects.object_oid IS NOT NULL\n");
+ }
+
+ /*
+ * If no tables were listed, filter for the relevant relation types. If
+ * tables were given via --table, don't bother filtering by relation type.
+ * Instead, let the server decide whether a given relation can be
+ * processed in which case the user will know about it.
+ */
+ if ((objfilter & OBJFILTER_TABLE) == 0)
+ {
+ /*
+ * vacuumdb should generally follow the behavior of the underlying
+ * VACUUM and ANALYZE commands. If analyze_only is true, process
+ * regular tables, materialized views, and partitioned tables, just
+ * like ANALYZE (with no specific target tables) does. Otherwise,
+ * process only regular tables and materialized views, since VACUUM
+ * skips partitioned tables when no target tables are specified.
+ */
+ if (vacopts->analyze_only)
+ appendPQExpBufferStr(&catalog_query,
+ " AND c.relkind OPERATOR(pg_catalog.=) ANY (array["
+ CppAsString2(RELKIND_RELATION) ", "
+ CppAsString2(RELKIND_MATVIEW) ", "
+ CppAsString2(RELKIND_PARTITIONED_TABLE) "])\n");
+ else
+ appendPQExpBufferStr(&catalog_query,
+ " AND c.relkind OPERATOR(pg_catalog.=) ANY (array["
+ CppAsString2(RELKIND_RELATION) ", "
+ CppAsString2(RELKIND_MATVIEW) "])\n");
+ }
+
+ /*
+ * For --min-xid-age and --min-mxid-age, the age of the relation is the
+ * greatest of the ages of the main relation and its associated TOAST
+ * table. The commands generated by vacuumdb will also process the TOAST
+ * table for the relation if necessary, so it does not need to be
+ * considered separately.
+ */
+ if (vacopts->min_xid_age != 0)
+ {
+ appendPQExpBuffer(&catalog_query,
+ " AND GREATEST(pg_catalog.age(c.relfrozenxid),"
+ " pg_catalog.age(t.relfrozenxid)) "
+ " OPERATOR(pg_catalog.>=) '%d'::pg_catalog.int4\n"
+ " AND c.relfrozenxid OPERATOR(pg_catalog.!=)"
+ " '0'::pg_catalog.xid\n",
+ vacopts->min_xid_age);
+ }
+
+ if (vacopts->min_mxid_age != 0)
+ {
+ appendPQExpBuffer(&catalog_query,
+ " AND GREATEST(pg_catalog.mxid_age(c.relminmxid),"
+ " pg_catalog.mxid_age(t.relminmxid)) OPERATOR(pg_catalog.>=)"
+ " '%d'::pg_catalog.int4\n"
+ " AND c.relminmxid OPERATOR(pg_catalog.!=)"
+ " '0'::pg_catalog.xid\n",
+ vacopts->min_mxid_age);
+ }
+
+ if (vacopts->missing_stats_only)
+ {
+ appendPQExpBufferStr(&catalog_query, " AND (\n");
+
+ /* regular stats */
+ appendPQExpBufferStr(&catalog_query,
+ " EXISTS (SELECT NULL FROM pg_catalog.pg_attribute a\n"
+ " WHERE a.attrelid OPERATOR(pg_catalog.=) c.oid\n"
+ " AND a.attnum OPERATOR(pg_catalog.>) 0::pg_catalog.int2\n"
+ " AND NOT a.attisdropped\n"
+ " AND a.attstattarget IS DISTINCT FROM 0::pg_catalog.int2\n"
+ " AND a.attgenerated OPERATOR(pg_catalog.<>) "
+ CppAsString2(ATTRIBUTE_GENERATED_VIRTUAL) "\n"
+ " AND NOT EXISTS (SELECT NULL FROM pg_catalog.pg_statistic s\n"
+ " WHERE s.starelid OPERATOR(pg_catalog.=) a.attrelid\n"
+ " AND s.staattnum OPERATOR(pg_catalog.=) a.attnum\n"
+ " AND s.stainherit OPERATOR(pg_catalog.=) p.inherited))\n");
+
+ /* extended stats */
+ appendPQExpBufferStr(&catalog_query,
+ " OR EXISTS (SELECT NULL FROM pg_catalog.pg_statistic_ext e\n"
+ " WHERE e.stxrelid OPERATOR(pg_catalog.=) c.oid\n"
+ " AND e.stxstattarget IS DISTINCT FROM 0::pg_catalog.int2\n"
+ " AND NOT EXISTS (SELECT NULL FROM pg_catalog.pg_statistic_ext_data d\n"
+ " WHERE d.stxoid OPERATOR(pg_catalog.=) e.oid\n"
+ " AND d.stxdinherit OPERATOR(pg_catalog.=) p.inherited))\n");
+
+ /* expression indexes */
+ appendPQExpBufferStr(&catalog_query,
+ " OR EXISTS (SELECT NULL FROM pg_catalog.pg_attribute a\n"
+ " JOIN pg_catalog.pg_index i"
+ " ON i.indexrelid OPERATOR(pg_catalog.=) a.attrelid\n"
+ " WHERE i.indrelid OPERATOR(pg_catalog.=) c.oid\n"
+ " AND i.indkey[a.attnum OPERATOR(pg_catalog.-) 1::pg_catalog.int2]"
+ " OPERATOR(pg_catalog.=) 0::pg_catalog.int2\n"
+ " AND a.attnum OPERATOR(pg_catalog.>) 0::pg_catalog.int2\n"
+ " AND NOT a.attisdropped\n"
+ " AND a.attstattarget IS DISTINCT FROM 0::pg_catalog.int2\n"
+ " AND NOT EXISTS (SELECT NULL FROM pg_catalog.pg_statistic s\n"
+ " WHERE s.starelid OPERATOR(pg_catalog.=) a.attrelid\n"
+ " AND s.staattnum OPERATOR(pg_catalog.=) a.attnum\n"
+ " AND s.stainherit OPERATOR(pg_catalog.=) p.inherited))\n");
+
+ /* inheritance and regular stats */
+ appendPQExpBufferStr(&catalog_query,
+ " OR EXISTS (SELECT NULL FROM pg_catalog.pg_attribute a\n"
+ " WHERE a.attrelid OPERATOR(pg_catalog.=) c.oid\n"
+ " AND a.attnum OPERATOR(pg_catalog.>) 0::pg_catalog.int2\n"
+ " AND NOT a.attisdropped\n"
+ " AND a.attstattarget IS DISTINCT FROM 0::pg_catalog.int2\n"
+ " AND a.attgenerated OPERATOR(pg_catalog.<>) "
+ CppAsString2(ATTRIBUTE_GENERATED_VIRTUAL) "\n"
+ " AND c.relhassubclass\n"
+ " AND NOT p.inherited\n"
+ " AND EXISTS (SELECT NULL FROM pg_catalog.pg_inherits h\n"
+ " WHERE h.inhparent OPERATOR(pg_catalog.=) c.oid)\n"
+ " AND NOT EXISTS (SELECT NULL FROM pg_catalog.pg_statistic s\n"
+ " WHERE s.starelid OPERATOR(pg_catalog.=) a.attrelid\n"
+ " AND s.staattnum OPERATOR(pg_catalog.=) a.attnum\n"
+ " AND s.stainherit))\n");
+
+ /* inheritance and extended stats */
+ appendPQExpBufferStr(&catalog_query,
+ " OR EXISTS (SELECT NULL FROM pg_catalog.pg_statistic_ext e\n"
+ " WHERE e.stxrelid OPERATOR(pg_catalog.=) c.oid\n"
+ " AND e.stxstattarget IS DISTINCT FROM 0::pg_catalog.int2\n"
+ " AND c.relhassubclass\n"
+ " AND NOT p.inherited\n"
+ " AND EXISTS (SELECT NULL FROM pg_catalog.pg_inherits h\n"
+ " WHERE h.inhparent OPERATOR(pg_catalog.=) c.oid)\n"
+ " AND NOT EXISTS (SELECT NULL FROM pg_catalog.pg_statistic_ext_data d\n"
+ " WHERE d.stxoid OPERATOR(pg_catalog.=) e.oid\n"
+ " AND d.stxdinherit))\n");
+
+ appendPQExpBufferStr(&catalog_query, " )\n");
+ }
+
+ /*
+ * Execute the catalog query. We use the default search_path for this
+ * query for consistency with table lookups done elsewhere by the user.
+ */
+ appendPQExpBufferStr(&catalog_query, " ORDER BY c.relpages DESC;");
+ executeCommand(conn, "RESET search_path;", echo);
+ res = executeQuery(conn, catalog_query.data, echo);
+ termPQExpBuffer(&catalog_query);
+ PQclear(executeQuery(conn, ALWAYS_SECURE_SEARCH_PATH_SQL, echo));
+
+ /*
+ * Build qualified identifiers for each table, including the column list
+ * if given.
+ */
+ initPQExpBuffer(&buf);
+ for (int i = 0; i < PQntuples(res); i++)
+ {
+ appendPQExpBufferStr(&buf,
+ fmtQualifiedIdEnc(PQgetvalue(res, i, 1),
+ PQgetvalue(res, i, 0),
+ PQclientEncoding(conn)));
+
+ if (objects_listed && !PQgetisnull(res, i, 2))
+ appendPQExpBufferStr(&buf, PQgetvalue(res, i, 2));
+
+ simple_string_list_append(found_objs, buf.data);
+ resetPQExpBuffer(&buf);
+ }
+ termPQExpBuffer(&buf);
+ PQclear(res);
+
+ return found_objs;
+}
+
+/*
+ * Vacuum/analyze all connectable databases.
+ *
+ * In analyze-in-stages mode, we process all databases in one stage before
+ * moving on to the next stage. That ensure minimal stats are available
+ * quickly everywhere before generating more detailed ones.
+ */
+void
+vacuum_all_databases(ConnParams *cparams,
+ vacuumingOptions *vacopts,
+ bool analyze_in_stages,
+ SimpleStringList *objects,
+ int concurrentCons,
+ const char *progname, bool echo, bool quiet)
+{
+ PGconn *conn;
+ PGresult *result;
+ int stage;
+ int i;
+
+ conn = connectMaintenanceDatabase(cparams, progname, echo);
+ result = executeQuery(conn,
+ "SELECT datname FROM pg_database WHERE datallowconn AND datconnlimit <> -2 ORDER BY 1;",
+ echo);
+ PQfinish(conn);
+
+ if (analyze_in_stages)
+ {
+ SimpleStringList **found_objs = NULL;
+
+ if (vacopts->missing_stats_only)
+ found_objs = palloc0(PQntuples(result) * sizeof(SimpleStringList *));
+
+ /*
+ * When analyzing all databases in stages, we analyze them all in the
+ * fastest stage first, so that initial statistics become available
+ * for all of them as soon as possible.
+ *
+ * This means we establish several times as many connections, but
+ * that's a secondary consideration.
+ */
+ for (stage = 0; stage < ANALYZE_NUM_STAGES; stage++)
+ {
+ for (i = 0; i < PQntuples(result); i++)
+ {
+ cparams->override_dbname = PQgetvalue(result, i, 0);
+
+ vacuum_one_database(cparams, vacopts,
+ stage,
+ objects,
+ vacopts->missing_stats_only ? &found_objs[i] : NULL,
+ concurrentCons,
+ progname, echo, quiet);
+ }
+ }
+ }
+ else
+ {
+ for (i = 0; i < PQntuples(result); i++)
+ {
+ cparams->override_dbname = PQgetvalue(result, i, 0);
+
+ vacuum_one_database(cparams, vacopts,
+ ANALYZE_NO_STAGE,
+ objects, NULL,
+ concurrentCons,
+ progname, echo, quiet);
+ }
+ }
+
+ PQclear(result);
+}
+
+/*
+ * Construct a vacuum/analyze command to run based on the given
+ * options, in the given string buffer, which may contain previous garbage.
+ *
+ * The table name used must be already properly quoted. The command generated
+ * depends on the server version involved and it is semicolon-terminated.
+ */
+void
+prepare_vacuum_command(PQExpBuffer sql, int serverVersion,
+ vacuumingOptions *vacopts, const char *table)
+{
+ const char *paren = " (";
+ const char *comma = ", ";
+ const char *sep = paren;
+
+ resetPQExpBuffer(sql);
+
+ if (vacopts->analyze_only)
+ {
+ appendPQExpBufferStr(sql, "ANALYZE");
+
+ /* parenthesized grammar of ANALYZE is supported since v11 */
+ if (serverVersion >= 110000)
+ {
+ if (vacopts->skip_locked)
+ {
+ /* SKIP_LOCKED is supported since v12 */
+ Assert(serverVersion >= 120000);
+ appendPQExpBuffer(sql, "%sSKIP_LOCKED", sep);
+ sep = comma;
+ }
+ if (vacopts->verbose)
+ {
+ appendPQExpBuffer(sql, "%sVERBOSE", sep);
+ sep = comma;
+ }
+ if (vacopts->buffer_usage_limit)
+ {
+ Assert(serverVersion >= 160000);
+ appendPQExpBuffer(sql, "%sBUFFER_USAGE_LIMIT '%s'", sep,
+ vacopts->buffer_usage_limit);
+ sep = comma;
+ }
+ if (sep != paren)
+ appendPQExpBufferChar(sql, ')');
+ }
+ else
+ {
+ if (vacopts->verbose)
+ appendPQExpBufferStr(sql, " VERBOSE");
+ }
+ }
+ else
+ {
+ appendPQExpBufferStr(sql, "VACUUM");
+
+ /* parenthesized grammar of VACUUM is supported since v9.0 */
+ if (serverVersion >= 90000)
+ {
+ if (vacopts->disable_page_skipping)
+ {
+ /* DISABLE_PAGE_SKIPPING is supported since v9.6 */
+ Assert(serverVersion >= 90600);
+ appendPQExpBuffer(sql, "%sDISABLE_PAGE_SKIPPING", sep);
+ sep = comma;
+ }
+ if (vacopts->no_index_cleanup)
+ {
+ /* "INDEX_CLEANUP FALSE" has been supported since v12 */
+ Assert(serverVersion >= 120000);
+ Assert(!vacopts->force_index_cleanup);
+ appendPQExpBuffer(sql, "%sINDEX_CLEANUP FALSE", sep);
+ sep = comma;
+ }
+ if (vacopts->force_index_cleanup)
+ {
+ /* "INDEX_CLEANUP TRUE" has been supported since v12 */
+ Assert(serverVersion >= 120000);
+ Assert(!vacopts->no_index_cleanup);
+ appendPQExpBuffer(sql, "%sINDEX_CLEANUP TRUE", sep);
+ sep = comma;
+ }
+ if (!vacopts->do_truncate)
+ {
+ /* TRUNCATE is supported since v12 */
+ Assert(serverVersion >= 120000);
+ appendPQExpBuffer(sql, "%sTRUNCATE FALSE", sep);
+ sep = comma;
+ }
+ if (!vacopts->process_main)
+ {
+ /* PROCESS_MAIN is supported since v16 */
+ Assert(serverVersion >= 160000);
+ appendPQExpBuffer(sql, "%sPROCESS_MAIN FALSE", sep);
+ sep = comma;
+ }
+ if (!vacopts->process_toast)
+ {
+ /* PROCESS_TOAST is supported since v14 */
+ Assert(serverVersion >= 140000);
+ appendPQExpBuffer(sql, "%sPROCESS_TOAST FALSE", sep);
+ sep = comma;
+ }
+ if (vacopts->skip_database_stats)
+ {
+ /* SKIP_DATABASE_STATS is supported since v16 */
+ Assert(serverVersion >= 160000);
+ appendPQExpBuffer(sql, "%sSKIP_DATABASE_STATS", sep);
+ sep = comma;
+ }
+ if (vacopts->skip_locked)
+ {
+ /* SKIP_LOCKED is supported since v12 */
+ Assert(serverVersion >= 120000);
+ appendPQExpBuffer(sql, "%sSKIP_LOCKED", sep);
+ sep = comma;
+ }
+ if (vacopts->full)
+ {
+ appendPQExpBuffer(sql, "%sFULL", sep);
+ sep = comma;
+ }
+ if (vacopts->freeze)
+ {
+ appendPQExpBuffer(sql, "%sFREEZE", sep);
+ sep = comma;
+ }
+ if (vacopts->verbose)
+ {
+ appendPQExpBuffer(sql, "%sVERBOSE", sep);
+ sep = comma;
+ }
+ if (vacopts->and_analyze)
+ {
+ appendPQExpBuffer(sql, "%sANALYZE", sep);
+ sep = comma;
+ }
+ if (vacopts->parallel_workers >= 0)
+ {
+ /* PARALLEL is supported since v13 */
+ Assert(serverVersion >= 130000);
+ appendPQExpBuffer(sql, "%sPARALLEL %d", sep,
+ vacopts->parallel_workers);
+ sep = comma;
+ }
+ if (vacopts->buffer_usage_limit)
+ {
+ Assert(serverVersion >= 160000);
+ appendPQExpBuffer(sql, "%sBUFFER_USAGE_LIMIT '%s'", sep,
+ vacopts->buffer_usage_limit);
+ sep = comma;
+ }
+ if (sep != paren)
+ appendPQExpBufferChar(sql, ')');
+ }
+ else
+ {
+ if (vacopts->full)
+ appendPQExpBufferStr(sql, " FULL");
+ if (vacopts->freeze)
+ appendPQExpBufferStr(sql, " FREEZE");
+ if (vacopts->verbose)
+ appendPQExpBufferStr(sql, " VERBOSE");
+ if (vacopts->and_analyze)
+ appendPQExpBufferStr(sql, " ANALYZE");
+ }
+ }
+
+ appendPQExpBuffer(sql, " %s;", table);
+}
+
+/*
+ * Send a vacuum/analyze command to the server, returning after sending the
+ * command.
+ *
+ * Any errors during command execution are reported to stderr.
+ */
+void
+run_vacuum_command(PGconn *conn, const char *sql, bool echo,
+ const char *table)
+{
+ bool status;
+
+ if (echo)
+ printf("%s\n", sql);
+
+ status = PQsendQuery(conn, sql) == 1;
+
+ if (!status)
+ {
+ if (table)
+ {
+ pg_log_error("vacuuming of table \"%s\" in database \"%s\" failed: %s",
+ table, PQdb(conn), PQerrorMessage(conn));
+ }
+ else
+ {
+ pg_log_error("vacuuming of database \"%s\" failed: %s",
+ PQdb(conn), PQerrorMessage(conn));
+ }
+ }
+}
+
+/*
+ * Returns a newly malloc'd version of 'src' with escaped single quotes and
+ * backslashes.
+ */
+char *
+escape_quotes(const char *src)
+{
+ char *result = escape_single_quotes_ascii(src);
+
+ if (!result)
+ pg_fatal("out of memory");
+ return result;
+}
diff --git a/src/bin/scripts/vacuuming.h b/src/bin/scripts/vacuuming.h
new file mode 100644
index 00000000000..d3f000840fa
--- /dev/null
+++ b/src/bin/scripts/vacuuming.h
@@ -0,0 +1,95 @@
+/*-------------------------------------------------------------------------
+ *
+ * vacuuming.h
+ * Common declarations for vacuuming.c
+ *
+ * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group
+ * Portions Copyright (c) 1994, Regents of the University of California
+ *
+ * src/bin/scripts/vacuuming.h
+ *
+ *-------------------------------------------------------------------------
+ */
+#ifndef VACUUMING_H
+#define VACUUMING_H
+
+#include "common.h"
+#include "fe_utils/connect_utils.h"
+#include "fe_utils/simple_list.h"
+
+/* For analyze-in-stages mode */
+#define ANALYZE_NO_STAGE -1
+#define ANALYZE_NUM_STAGES 3
+
+/* vacuum options controlled by user flags */
+typedef struct vacuumingOptions
+{
+ bool analyze_only;
+ bool verbose;
+ bool and_analyze;
+ bool full;
+ bool freeze;
+ bool disable_page_skipping;
+ bool skip_locked;
+ int min_xid_age;
+ int min_mxid_age;
+ int parallel_workers; /* >= 0 indicates user specified the
+ * parallel degree, otherwise -1 */
+ bool no_index_cleanup;
+ bool force_index_cleanup;
+ bool do_truncate;
+ bool process_main;
+ bool process_toast;
+ bool skip_database_stats;
+ char *buffer_usage_limit;
+ bool missing_stats_only;
+} vacuumingOptions;
+
+/* object filter options */
+typedef enum
+{
+ OBJFILTER_NONE = 0, /* no filter used */
+ OBJFILTER_ALL_DBS = (1 << 0), /* -a | --all */
+ OBJFILTER_DATABASE = (1 << 1), /* -d | --dbname */
+ OBJFILTER_TABLE = (1 << 2), /* -t | --table */
+ OBJFILTER_SCHEMA = (1 << 3), /* -n | --schema */
+ OBJFILTER_SCHEMA_EXCLUDE = (1 << 4), /* -N | --exclude-schema */
+} VacObjFilter;
+
+extern VacObjFilter objfilter;
+
+extern void vacuuming_main(ConnParams *cparams, const char *dbname,
+ const char *maintenance_db, vacuumingOptions *vacopts,
+ SimpleStringList *objects, bool analyze_in_stages,
+ int tbl_count, int concurrentCons,
+ const char *progname, bool echo, bool quiet);
+
+extern SimpleStringList *retrieve_objects(PGconn *conn,
+ vacuumingOptions *vacopts,
+ SimpleStringList *objects,
+ bool echo);
+
+extern void vacuum_one_database(ConnParams *cparams,
+ vacuumingOptions *vacopts,
+ int stage,
+ SimpleStringList *objects,
+ SimpleStringList **found_objs,
+ int concurrentCons,
+ const char *progname, bool echo, bool quiet);
+
+extern void vacuum_all_databases(ConnParams *cparams,
+ vacuumingOptions *vacopts,
+ bool analyze_in_stages,
+ SimpleStringList *objects,
+ int concurrentCons,
+ const char *progname, bool echo, bool quiet);
+
+extern void prepare_vacuum_command(PQExpBuffer sql, int serverVersion,
+ vacuumingOptions *vacopts, const char *table);
+
+extern void run_vacuum_command(PGconn *conn, const char *sql, bool echo,
+ const char *table);
+
+extern char *escape_quotes(const char *src);
+
+#endif /* VACUUMING_H */
--
2.39.5
--3q6txozlgl6t37td
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
filename="v20-0002-Add-REPACK-command.patch"
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 477+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 477+ messages in thread
end of thread, other threads:[~2025-12-18 17:21 UTC | newest]
Thread overview: 477+ messages (download: mbox mbox.gz follow: Atom feed)
-- links below jump to the message on this page --
2025-08-30 12:39 [PATCH v20 1/6] Split vacuumdb to create vacuuming.c/h Álvaro Herrera <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
This inbox is served by agora; see mirroring instructions
for how to clone and mirror all data and code used for this inbox