agora inbox for [email protected]  

help / color / mirror / Atom feed

[PATCH v32 09/11] Add support for min/max aggregates for IVM
371+ messages / 2 participants
[nested] [flat]

* [PATCH v32 09/11] Add support for min/max aggregates for IVM
@ 2023-05-31 11:58  Yugo Nagata <[email protected]>
  0 siblings, 0 replies; 371+ messages in thread

From: Yugo Nagata @ 2023-05-31 11:58 UTC (permalink / raw)

Supporting min and max is more complicated than count, sum, or avg.

For an example of min, when tuples are inserted, the current min value
in the view and the min value in the inseteted tuples are compared,
then the smaller one is used as the latest min value. On the other
hand, when tuples are deleted, if the current min value in the view
equals to the min in the deleted tuples, we need re-computation the
latest min value from base tables. Otherwise, the current value in
the view remains.
---
 src/backend/commands/createas.c |  45 +++
 src/backend/commands/matview.c  | 644 +++++++++++++++++++++++++++++++-
 2 files changed, 680 insertions(+), 9 deletions(-)

diff --git a/src/backend/commands/createas.c b/src/backend/commands/createas.c
index ecec93ec1c..d3cf3df5a7 100644
--- a/src/backend/commands/createas.c
+++ b/src/backend/commands/createas.c
@@ -1311,6 +1311,51 @@ check_aggregate_supports_ivm(Oid aggfnoid)
 		case F_AVG_FLOAT8:
 		case F_AVG_INTERVAL:
 
+		/* min */
+		case F_MIN_ANYARRAY:
+		case F_MIN_INT8:
+		case F_MIN_INT4:
+		case F_MIN_INT2:
+		case F_MIN_OID:
+		case F_MIN_FLOAT4:
+		case F_MIN_FLOAT8:
+		case F_MIN_DATE:
+		case F_MIN_TIME:
+		case F_MIN_TIMETZ:
+		case F_MIN_MONEY:
+		case F_MIN_TIMESTAMP:
+		case F_MIN_TIMESTAMPTZ:
+		case F_MIN_INTERVAL:
+		case F_MIN_TEXT:
+		case F_MIN_NUMERIC:
+		case F_MIN_BPCHAR:
+		case F_MIN_TID:
+		case F_MIN_ANYENUM:
+		case F_MIN_INET:
+		case F_MIN_PG_LSN:
+
+		/* max */
+		case F_MAX_ANYARRAY:
+		case F_MAX_INT8:
+		case F_MAX_INT4:
+		case F_MAX_INT2:
+		case F_MAX_OID:
+		case F_MAX_FLOAT4:
+		case F_MAX_FLOAT8:
+		case F_MAX_DATE:
+		case F_MAX_TIME:
+		case F_MAX_TIMETZ:
+		case F_MAX_MONEY:
+		case F_MAX_TIMESTAMP:
+		case F_MAX_TIMESTAMPTZ:
+		case F_MAX_INTERVAL:
+		case F_MAX_TEXT:
+		case F_MAX_NUMERIC:
+		case F_MAX_BPCHAR:
+		case F_MAX_TID:
+		case F_MAX_ANYENUM:
+		case F_MAX_INET:
+		case F_MAX_PG_LSN:
 			return true;
 
 		default:
diff --git a/src/backend/commands/matview.c b/src/backend/commands/matview.c
index b7f6c3831b..42ff0ed98f 100644
--- a/src/backend/commands/matview.c
+++ b/src/backend/commands/matview.c
@@ -69,6 +69,34 @@ typedef struct
 
 #define MV_INIT_QUERYHASHSIZE	16
 
+/* MV query type codes */
+#define MV_PLAN_RECALC			1
+#define MV_PLAN_SET_VALUE		2
+
+/*
+ * MI_QueryKey
+ *
+ * The key identifying a prepared SPI plan in our query hashtable
+ */
+typedef struct MV_QueryKey
+{
+	Oid			matview_id;	/* OID of materialized view */
+	int32		query_type;	/* query type ID, see MV_PLAN_XXX above */
+} MV_QueryKey;
+
+/*
+ * MV_QueryHashEntry
+ *
+ * Hash entry for cached plans used to maintain materialized views.
+ */
+typedef struct MV_QueryHashEntry
+{
+	MV_QueryKey key;
+	SPIPlanPtr	plan;
+	SearchPathMatcher *search_path;        /* search_path used for parsing
+											 * and planning */
+} MV_QueryHashEntry;
+
 /*
  * MV_TriggerHashEntry
  *
@@ -105,6 +133,7 @@ typedef struct MV_TriggerTable
 	TupleTableSlot *slot;		/* for checking visibility in the pre-state table */
 } MV_TriggerTable;
 
+static HTAB *mv_query_cache = NULL;
 static HTAB *mv_trigger_info = NULL;
 
 static bool in_delta_calculation = false;
@@ -165,6 +194,9 @@ static void append_set_clause_for_sum(const char *resname, StringInfo buf_old,
 static void append_set_clause_for_avg(const char *resname, StringInfo buf_old,
 						  StringInfo buf_new, StringInfo aggs_list,
 						  const char *aggtype);
+static void append_set_clause_for_minmax(const char *resname, StringInfo buf_old,
+							 StringInfo buf_new, StringInfo aggs_list,
+							 bool is_min);
 static char *get_operation_string(IvmOp op, const char *col, const char *arg1, const char *arg2,
 					 const char* count_col, const char *castType);
 static char *get_null_condition_string(IvmOp op, const char *arg1, const char *arg2,
@@ -173,17 +205,30 @@ static void apply_old_delta(const char *matviewname, const char *deltaname_old,
 				List *keys);
 static void apply_old_delta_with_count(const char *matviewname, const char *deltaname_old,
 				List *keys, StringInfo aggs_list, StringInfo aggs_set,
-				const char *count_colname);
+				List *minmax_list, List *is_min_list,
+				const char *count_colname,
+				SPITupleTable **tuptable_recalc, uint64 *num_recalc);
 static void apply_new_delta(const char *matviewname, const char *deltaname_new,
 				StringInfo target_list);
 static void apply_new_delta_with_count(const char *matviewname, const char* deltaname_new,
 				List *keys, StringInfo target_list, StringInfo aggs_set,
 				const char* count_colname);
 static char *get_matching_condition_string(List *keys);
+static char *get_returning_string(List *minmax_list, List *is_min_list, List *keys);
+static char *get_minmax_recalc_condition_string(List *minmax_list, List *is_min_list);
+static char *get_select_for_recalc_string(List *keys);
+static void recalc_and_set_values(SPITupleTable *tuptable_recalc, int64 num_tuples,
+					  List *namelist, List *keys, Relation matviewRel);
+static SPIPlanPtr get_plan_for_recalc(Oid matviewOid, List *namelist, List *keys, Oid *keyTypes);
+static SPIPlanPtr get_plan_for_set_values(Oid matviewOid, char *matviewname, List *namelist,
+						Oid *valTypes);
 static void generate_equal(StringInfo querybuf, Oid opttype,
 			   const char *leftop, const char *rightop);
 
 static void mv_InitHashTables(void);
+static SPIPlanPtr mv_FetchPreparedPlan(MV_QueryKey *key);
+static void mv_HashPreparedPlan(MV_QueryKey *key, SPIPlanPtr plan);
+static void mv_BuildQueryKey(MV_QueryKey *key, Oid matview_id, int32 query_type);
 static void clean_up_IVM_hash_entry(MV_TriggerHashEntry *entry, bool is_abort);
 
 /*
@@ -2124,6 +2169,8 @@ apply_delta(Oid matviewOid, Tuplestorestate *old_tuplestores, Tuplestorestate *n
 	ListCell	*lc;
 	int			i;
 	List	   *keys = NIL;
+	List	   *minmax_list = NIL;
+	List	   *is_min_list = NIL;
 
 
 	/*
@@ -2205,6 +2252,17 @@ apply_delta(Oid matviewOid, Tuplestorestate *old_tuplestores, Tuplestorestate *n
 				append_set_clause_for_avg(resname, aggs_set_old, aggs_set_new, aggs_list_buf,
 										  format_type_be(aggref->aggtype));
 
+			/* min/max */
+			else if (!strcmp(aggname, "min") || !strcmp(aggname, "max"))
+			{
+				bool	is_min = (!strcmp(aggname, "min"));
+
+				append_set_clause_for_minmax(resname, aggs_set_old, aggs_set_new, aggs_list_buf, is_min);
+
+				/* make a resname list of min and max aggregates */
+				minmax_list = lappend(minmax_list, resname);
+				is_min_list = lappend_int(is_min_list, is_min);
+			}
 			else
 				elog(ERROR, "unsupported aggregate function: %s", aggname);
 		}
@@ -2234,6 +2292,8 @@ apply_delta(Oid matviewOid, Tuplestorestate *old_tuplestores, Tuplestorestate *n
 	if (old_tuplestores && tuplestore_tuple_count(old_tuplestores) > 0)
 	{
 		EphemeralNamedRelation enr = palloc(sizeof(EphemeralNamedRelationData));
+		SPITupleTable  *tuptable_recalc = NULL;
+		uint64			num_recalc;
 		int				rc;
 
 		/* convert tuplestores to ENR, and register for SPI */
@@ -2252,10 +2312,18 @@ apply_delta(Oid matviewOid, Tuplestorestate *old_tuplestores, Tuplestorestate *n
 			/* apply old delta and get rows to be recalculated */
 			apply_old_delta_with_count(matviewname, OLD_DELTA_ENRNAME,
 									   keys, aggs_list_buf, aggs_set_old,
-									   count_colname);
+									   minmax_list, is_min_list,
+									   count_colname, &tuptable_recalc, &num_recalc);
 		else
 			apply_old_delta(matviewname, OLD_DELTA_ENRNAME, keys);
 
+		/*
+		 * If we have min or max, we might have to recalculate aggregate values from base tables
+		 * on some tuples. TIDs and keys such tuples are returned as a result of the above query.
+		 */
+		if (minmax_list && tuptable_recalc)
+			recalc_and_set_values(tuptable_recalc, num_recalc, minmax_list, keys, matviewRel);
+
 	}
 	/* For tuple insertion */
 	if (new_tuplestores && tuplestore_tuple_count(new_tuplestores) > 0)
@@ -2447,6 +2515,70 @@ append_set_clause_for_avg(const char *resname, StringInfo buf_old,
 	);
 }
 
+/*
+ * append_set_clause_for_minmax
+ *
+ * Append SET clause string for min or max aggregation to given buffers.
+ * Also, append resnames required for calculating the aggregate value.
+ * is_min is true if this is min, false if not.
+ */
+static void
+append_set_clause_for_minmax(const char *resname, StringInfo buf_old,
+							 StringInfo buf_new, StringInfo aggs_list,
+							 bool is_min)
+{
+	char *count_col = IVM_colname("count", resname);
+
+	/* For tuple deletion */
+	if (buf_old)
+	{
+		/*
+		 * If the new value doesn't became NULL then use the value remaining
+		 * in the view although this will be recomputated afterwords.
+		 */
+		appendStringInfo(buf_old,
+			", %s = CASE WHEN %s THEN NULL ELSE %s END",
+			quote_qualified_identifier(NULL, resname),
+			get_null_condition_string(IVM_SUB, "mv", "t", count_col),
+			quote_qualified_identifier("mv", resname)
+		);
+		/* count = mv.count - t.count */
+		appendStringInfo(buf_old,
+			", %s = %s",
+			quote_qualified_identifier(NULL, count_col),
+			get_operation_string(IVM_SUB, count_col, "mv", "t", NULL, NULL)
+		);
+	}
+	/* For tuple insertion */
+	if (buf_new)
+	{
+		/*
+		 * min = LEAST(mv.min, diff.min)
+		 * max = GREATEST(mv.max, diff.max)
+		 */
+		appendStringInfo(buf_new,
+			", %s = CASE WHEN %s THEN NULL ELSE %s(%s,%s) END",
+			quote_qualified_identifier(NULL, resname),
+			get_null_condition_string(IVM_ADD, "mv", "diff", count_col),
+
+			is_min ? "LEAST" : "GREATEST",
+			quote_qualified_identifier("mv", resname),
+			quote_qualified_identifier("diff", resname)
+		);
+		/* count = mv.count + diff.count */
+		appendStringInfo(buf_new,
+			", %s = %s",
+			quote_qualified_identifier(NULL, count_col),
+			get_operation_string(IVM_ADD, count_col, "mv", "diff", NULL, NULL)
+		);
+	}
+
+	appendStringInfo(aggs_list, ", %s, %s",
+		quote_qualified_identifier("diff", resname),
+		quote_qualified_identifier("diff", IVM_colname("count", resname))
+	);
+}
+
 /*
  * get_operation_string
  *
@@ -2549,19 +2681,44 @@ get_null_condition_string(IvmOp op, const char *arg1, const char *arg2,
  * list to identify a tuple in the view. If the view has aggregates, this
  * requires strings representing resnames of aggregates and SET clause for
  * updating aggregate values.
+ *
+ * If the view has min or max aggregate, this requires a list of resnames of
+ * min/max aggregates and a list of boolean which represents which entries in
+ * minmax_list is min. These are necessary to check if we need to recalculate
+ * min or max aggregate values. In this case, this query returns TID and keys
+ * of tuples which need to be recalculated.  This result and the number of rows
+ * are stored in tuptables and num_recalc repectedly.
+ *
  */
 static void
 apply_old_delta_with_count(const char *matviewname, const char *deltaname_old,
 				List *keys, StringInfo aggs_list, StringInfo aggs_set,
-				const char *count_colname)
+				List *minmax_list, List *is_min_list,
+				const char *count_colname,
+				SPITupleTable **tuptable_recalc, uint64 *num_recalc)
 {
 	StringInfoData	querybuf;
 	char   *match_cond;
+	char   *updt_returning = "";
+	char   *select_for_recalc = "SELECT";
 	bool	agg_without_groupby = (list_length(keys) == 0);
 
+	Assert(tuptable_recalc != NULL);
+	Assert(num_recalc != NULL);
+
 	/* build WHERE condition for searching tuples to be deleted */
 	match_cond = get_matching_condition_string(keys);
 
+	/*
+	 * We need a special RETURNING clause and SELECT statement for min/max to
+	 * check which tuple needs re-calculation from base tables.
+	 */
+	if (minmax_list)
+	{
+		updt_returning = get_returning_string(minmax_list, is_min_list, keys);
+		select_for_recalc = get_select_for_recalc_string(keys);
+	}
+
 	/* Search for matching tuples from the view and update or delete if found. */
 	initStringInfo(&querybuf);
 	appendStringInfo(&querybuf,
@@ -2576,10 +2733,11 @@ apply_old_delta_with_count(const char *matviewname, const char *deltaname_old,
 						"UPDATE %s AS mv SET %s = mv.%s OPERATOR(pg_catalog.-) t.%s "
 											"%s"	/* SET clauses for aggregates */
 						"FROM t WHERE mv.ctid OPERATOR(pg_catalog.=) t.ctid AND NOT for_dlt "
-					")"
-					/* delete a tuple if this is to be deleted */
-					"DELETE FROM %s AS mv USING t "
-					"WHERE mv.ctid OPERATOR(pg_catalog.=) t.ctid AND for_dlt",
+						"%s"						/* RETURNING clause for recalc infomation */
+					"), dlt AS ("			/* delete a tuple if this is to be deleted */
+						"DELETE FROM %s AS mv USING t "
+						"WHERE mv.ctid OPERATOR(pg_catalog.=) t.ctid AND for_dlt"
+					") %s",							/* SELECT returning which tuples need to be recalculated */
 					count_colname,
 					count_colname, count_colname, (agg_without_groupby ? "false" : "true"),
 					(aggs_list != NULL ? aggs_list->data : ""),
@@ -2587,10 +2745,25 @@ apply_old_delta_with_count(const char *matviewname, const char *deltaname_old,
 					match_cond,
 					matviewname, count_colname, count_colname, count_colname,
 					(aggs_set != NULL ? aggs_set->data : ""),
-					matviewname);
+					updt_returning,
+					matviewname,
+					select_for_recalc);
 
-	if (SPI_exec(querybuf.data, 0) != SPI_OK_DELETE)
+	if (SPI_exec(querybuf.data, 0) != SPI_OK_SELECT)
 		elog(ERROR, "SPI_exec failed: %s", querybuf.data);
+
+
+	/* Return tuples to be recalculated. */
+	if (minmax_list)
+	{
+		*tuptable_recalc = SPI_tuptable;
+		*num_recalc = SPI_processed;
+	}
+	else
+	{
+		*tuptable_recalc = NULL;
+		*num_recalc = 0;
+	}
 }
 
 /*
@@ -2773,6 +2946,349 @@ get_matching_condition_string(List *keys)
 	return match_cond.data;
 }
 
+/*
+ * get_returning_string
+ *
+ * Build a string for RETURNING clause of UPDATE used in apply_old_delta_with_count.
+ * This clause returns ctid and a boolean value that indicates if we need to
+ * recalculate min or max value, for each updated row.
+ */
+static char *
+get_returning_string(List *minmax_list, List *is_min_list, List *keys)
+{
+	StringInfoData returning;
+	char		*recalc_cond;
+	ListCell	*lc;
+
+	Assert(minmax_list != NIL && is_min_list != NIL);
+	recalc_cond = get_minmax_recalc_condition_string(minmax_list, is_min_list);
+
+	initStringInfo(&returning);
+
+	appendStringInfo(&returning, "RETURNING mv.ctid AS tid, (%s) AS recalc", recalc_cond);
+	foreach (lc, keys)
+	{
+		Form_pg_attribute attr = (Form_pg_attribute) lfirst(lc);
+		char *resname = NameStr(attr->attname);
+		appendStringInfo(&returning, ", %s", quote_qualified_identifier("mv", resname));
+	}
+
+	return returning.data;
+}
+
+/*
+ * get_minmax_recalc_condition_string
+ *
+ * Build a predicate string for checking if any min/max aggregate
+ * value needs to be recalculated.
+ */
+static char *
+get_minmax_recalc_condition_string(List *minmax_list, List *is_min_list)
+{
+	StringInfoData recalc_cond;
+	ListCell	*lc1, *lc2;
+
+	initStringInfo(&recalc_cond);
+
+	Assert (list_length(minmax_list) == list_length(is_min_list));
+
+	forboth (lc1, minmax_list, lc2, is_min_list)
+	{
+		char   *resname = (char *) lfirst(lc1);
+		bool	is_min = (bool) lfirst_int(lc2);
+		char   *op_str = (is_min ? ">=" : "<=");
+
+		appendStringInfo(&recalc_cond, "%s OPERATOR(pg_catalog.%s) %s",
+			quote_qualified_identifier("mv", resname),
+			op_str,
+			quote_qualified_identifier("t", resname)
+		);
+
+		if (lnext(minmax_list, lc1))
+			appendStringInfo(&recalc_cond, " OR ");
+	}
+
+	return recalc_cond.data;
+}
+
+/*
+ * get_select_for_recalc_string
+ *
+ * Build a query to return tid and keys of tuples which need
+ * recalculation. This is used as the result of the query
+ * built by apply_old_delta.
+ */
+static char *
+get_select_for_recalc_string(List *keys)
+{
+	StringInfoData qry;
+	ListCell	*lc;
+
+	initStringInfo(&qry);
+
+	appendStringInfo(&qry, "SELECT tid");
+	foreach (lc, keys)
+	{
+		Form_pg_attribute attr = (Form_pg_attribute) lfirst(lc);
+		appendStringInfo(&qry, ", %s", NameStr(attr->attname));
+	}
+
+	appendStringInfo(&qry, " FROM updt WHERE recalc");
+
+	return qry.data;
+}
+
+/*
+ * recalc_and_set_values
+ *
+ * Recalculate tuples in a materialized from base tables and update these.
+ * The tuples which needs recalculation are specified by keys, and resnames
+ * of columns to be updated are specified by namelist. TIDs and key values
+ * are given by tuples in tuptable_recalc. Its first attribute must be TID
+ * and key values must be following this.
+ */
+static void
+recalc_and_set_values(SPITupleTable *tuptable_recalc, int64 num_tuples,
+					  List *namelist, List *keys, Relation matviewRel)
+{
+	TupleDesc   tupdesc_recalc = tuptable_recalc->tupdesc;
+	Oid		   *keyTypes = NULL, *types = NULL;
+	char	   *keyNulls = NULL, *nulls = NULL;
+	Datum	   *keyVals = NULL, *vals = NULL;
+	int			num_vals = list_length(namelist);
+	int			num_keys = list_length(keys);
+	uint64      i;
+	Oid			matviewOid;
+	char	   *matviewname;
+
+	matviewOid = RelationGetRelid(matviewRel);
+	matviewname = quote_qualified_identifier(get_namespace_name(RelationGetNamespace(matviewRel)),
+											 RelationGetRelationName(matviewRel));
+
+	/* If we have keys, initialize arrays for them. */
+	if (keys)
+	{
+		keyTypes = palloc(sizeof(Oid) * num_keys);
+		keyNulls = palloc(sizeof(char) * num_keys);
+		keyVals = palloc(sizeof(Datum) * num_keys);
+		/* a tuple contains keys to be recalculated and ctid to be updated*/
+		Assert(tupdesc_recalc->natts == num_keys + 1);
+
+		/* Types of key attributes  */
+		for (i = 0; i < num_keys; i++)
+			keyTypes[i] = TupleDescAttr(tupdesc_recalc, i + 1)->atttypid;
+	}
+
+	/* allocate memory for all attribute names and tid */
+	types = palloc(sizeof(Oid) * (num_vals + 1));
+	nulls = palloc(sizeof(char) * (num_vals + 1));
+	vals = palloc(sizeof(Datum) * (num_vals + 1));
+
+	/* For each tuple which needs recalculation */
+	for (i = 0; i < num_tuples; i++)
+	{
+		int j;
+		bool isnull;
+		SPIPlanPtr plan;
+		SPITupleTable *tuptable_newvals;
+		TupleDesc   tupdesc_newvals;
+
+		/* Set group key values as parameters if needed. */
+		if (keys)
+		{
+			for (j = 0; j < num_keys; j++)
+			{
+				keyVals[j] = SPI_getbinval(tuptable_recalc->vals[i], tupdesc_recalc, j + 2, &isnull);
+				if (isnull)
+					keyNulls[j] = 'n';
+				else
+					keyNulls[j] = ' ';
+			}
+		}
+
+		/*
+		 * Get recalculated values from base tables. The result must be
+		 * only one tuple thich contains the new values for specified keys.
+		 */
+		plan = get_plan_for_recalc(matviewOid, namelist, keys, keyTypes);
+		if (SPI_execute_plan(plan, keyVals, keyNulls, false, 0) != SPI_OK_SELECT)
+			elog(ERROR, "SPI_execute_plan");
+		if (SPI_processed != 1)
+			elog(ERROR, "SPI_execute_plan returned zero or more than one rows");
+
+		tuptable_newvals = SPI_tuptable;
+		tupdesc_newvals = tuptable_newvals->tupdesc;
+
+		Assert(tupdesc_newvals->natts == num_vals);
+
+		/* Set the new values as parameters */
+		for (j = 0; j < tupdesc_newvals->natts; j++)
+		{
+			if (i == 0)
+				types[j] = TupleDescAttr(tupdesc_newvals, j)->atttypid;
+
+			vals[j] = SPI_getbinval(tuptable_newvals->vals[0], tupdesc_newvals, j + 1, &isnull);
+			if (isnull)
+				nulls[j] = 'n';
+			else
+				nulls[j] = ' ';
+		}
+		/* Set TID of the view tuple to be updated as a parameter */
+		types[j] = TIDOID;
+		vals[j] = SPI_getbinval(tuptable_recalc->vals[i], tupdesc_recalc, 1, &isnull);
+		nulls[j] = ' ';
+
+		/* Update the view tuple to the new values */
+		plan = get_plan_for_set_values(matviewOid, matviewname, namelist, types);
+		if (SPI_execute_plan(plan, vals, nulls, false, 0) != SPI_OK_UPDATE)
+			elog(ERROR, "SPI_execute_plan");
+	}
+}
+
+
+/*
+ * get_plan_for_recalc
+ *
+ * Create or fetch a plan for recalculating value in the view's target list
+ * from base tables using the definition query of materialized view specified
+ * by matviewOid. namelist is a list of resnames of values to be recalculated.
+ *
+ * keys is a list of keys to identify tuples to be recalculated if this is not
+ * empty. KeyTypes is an array of types of keys.
+ */
+static SPIPlanPtr
+get_plan_for_recalc(Oid matviewOid, List *namelist, List *keys, Oid *keyTypes)
+{
+	MV_QueryKey hash_key;
+	SPIPlanPtr	plan;
+
+	/* Fetch or prepare a saved plan for the recalculation */
+	mv_BuildQueryKey(&hash_key, matviewOid, MV_PLAN_RECALC);
+	if ((plan = mv_FetchPreparedPlan(&hash_key)) == NULL)
+	{
+		ListCell	   *lc;
+		StringInfoData	str;
+		char   *viewdef;
+
+		/* get view definition of matview */
+		viewdef = text_to_cstring((text *) DatumGetPointer(
+					DirectFunctionCall1(pg_get_viewdef, ObjectIdGetDatum(matviewOid))));
+		/* get rid of trailing semi-colon */
+		viewdef[strlen(viewdef)-1] = '\0';
+
+		/*
+		 * Build a query string for recalculating values. This is like
+		 *
+		 *  SELECT x1, x2, x3, ... FROM ( ... view definition query ...) mv
+		 *   WHERE (key1, key2, ...) = ($1, $2, ...);
+		 */
+
+		initStringInfo(&str);
+		appendStringInfo(&str, "SELECT ");
+		foreach (lc, namelist)
+		{
+			appendStringInfo(&str, "%s", (char *) lfirst(lc));
+			if (lnext(namelist, lc))
+				appendStringInfoString(&str, ", ");
+		}
+		appendStringInfo(&str, " FROM (%s) mv", viewdef);
+
+		if (keys)
+		{
+			int		i = 1;
+			char	paramname[16];
+
+			appendStringInfo(&str, " WHERE (");
+			foreach (lc, keys)
+			{
+				Form_pg_attribute attr = (Form_pg_attribute) lfirst(lc);
+				char   *resname = NameStr(attr->attname);
+				Oid		typid = attr->atttypid;
+
+				sprintf(paramname, "$%d", i);
+				appendStringInfo(&str, "(");
+				generate_equal(&str, typid, resname, paramname);
+				appendStringInfo(&str, " OR (%s IS NULL AND %s IS NULL))",
+								 resname, paramname);
+
+				if (lnext(keys, lc))
+					appendStringInfoString(&str, " AND ");
+				i++;
+			}
+			appendStringInfo(&str, ")");
+		}
+		else
+			keyTypes = NULL;
+
+		plan = SPI_prepare(str.data, list_length(keys), keyTypes);
+		if (plan == NULL)
+			elog(ERROR, "SPI_prepare returned %s for %s", SPI_result_code_string(SPI_result), str.data);
+
+		SPI_keepplan(plan);
+		mv_HashPreparedPlan(&hash_key, plan);
+	}
+
+	return plan;
+}
+
+/*
+ * get_plan_for_set_values
+ *
+ * Create or fetch a plan for applying new values calculated by
+ * get_plan_for_recalc to a materialized view specified by matviewOid.
+ * matviewname is the name of the view.  namelist is a list of resnames
+ * of attributes to be updated, and valTypes is an array of types of the
+ * values.
+ */
+static SPIPlanPtr
+get_plan_for_set_values(Oid matviewOid, char *matviewname, List *namelist,
+						Oid *valTypes)
+{
+	MV_QueryKey	key;
+	SPIPlanPtr	plan;
+
+	/* Fetch or prepare a saved plan for the real check */
+	mv_BuildQueryKey(&key, matviewOid, MV_PLAN_SET_VALUE);
+	if ((plan = mv_FetchPreparedPlan(&key)) == NULL)
+	{
+		ListCell	  *lc;
+		StringInfoData str;
+		int		i;
+
+		/*
+		 * Build a query string for applying min/max values. This is like
+		 *
+		 *  UPDATE matviewname AS mv
+		 *   SET (x1, x2, x3, x4) = ($1, $2, $3, $4)
+		 *   WHERE ctid = $5;
+		 */
+
+		initStringInfo(&str);
+		appendStringInfo(&str, "UPDATE %s AS mv SET (", matviewname);
+		foreach (lc, namelist)
+		{
+			appendStringInfo(&str, "%s", (char *) lfirst(lc));
+			if (lnext(namelist, lc))
+				appendStringInfoString(&str, ", ");
+		}
+		appendStringInfo(&str, ") = ROW(");
+
+		for (i = 1; i <= list_length(namelist); i++)
+			appendStringInfo(&str, "%s$%d", (i==1 ? "" : ", "), i);
+
+		appendStringInfo(&str, ") WHERE ctid OPERATOR(pg_catalog.=) $%d", i);
+
+		plan = SPI_prepare(str.data, list_length(namelist) + 1, valTypes);
+		if (plan == NULL)
+			elog(ERROR, "SPI_prepare returned %s for %s", SPI_result_code_string(SPI_result), str.data);
+
+		SPI_keepplan(plan);
+		mv_HashPreparedPlan(&key, plan);
+	}
+
+	return plan;
+}
+
 /*
  * generate_equals
  *
@@ -2806,6 +3322,13 @@ mv_InitHashTables(void)
 {
 	HASHCTL		ctl;
 
+	memset(&ctl, 0, sizeof(ctl));
+	ctl.keysize = sizeof(MV_QueryKey);
+	ctl.entrysize = sizeof(MV_QueryHashEntry);
+	mv_query_cache = hash_create("MV query cache",
+								 MV_INIT_QUERYHASHSIZE,
+								 &ctl, HASH_ELEM | HASH_BLOBS);
+
 	memset(&ctl, 0, sizeof(ctl));
 	ctl.keysize = sizeof(Oid);
 	ctl.entrysize = sizeof(MV_TriggerHashEntry);
@@ -2814,6 +3337,109 @@ mv_InitHashTables(void)
 								 &ctl, HASH_ELEM | HASH_BLOBS);
 }
 
+/*
+ * mv_FetchPreparedPlan
+ */
+static SPIPlanPtr
+mv_FetchPreparedPlan(MV_QueryKey *key)
+{
+	MV_QueryHashEntry *entry;
+	SPIPlanPtr	plan;
+
+	/*
+	 * On the first call initialize the hashtable
+	 */
+	if (!mv_query_cache)
+		mv_InitHashTables();
+
+	/*
+	 * Lookup for the key
+	 */
+	entry = (MV_QueryHashEntry *) hash_search(mv_query_cache,
+											  (void *) key,
+											  HASH_FIND, NULL);
+	if (entry == NULL)
+		return NULL;
+
+	/*
+	 * Check whether the plan is still valid.  If it isn't, we don't want to
+	 * simply rely on plancache.c to regenerate it; rather we should start
+	 * from scratch and rebuild the query text too.  This is to cover cases
+	 * such as table/column renames.  We depend on the plancache machinery to
+	 * detect possible invalidations, though.
+	 *
+	 * CAUTION: this check is only trustworthy if the caller has already
+	 * locked both materialized views and base tables.
+	 *
+	 * Also, check whether the search_path is still the same as when we made it.
+	 * If it isn't, we need to rebuild the query text because the result of
+	 * pg_ivm_get_viewdef() will change.
+	 */
+	plan = entry->plan;
+	if (plan && SPI_plan_is_valid(plan) &&
+		SearchPathMatchesCurrentEnvironment(entry->search_path))
+		return plan;
+
+	/*
+	 * Otherwise we might as well flush the cached plan now, to free a little
+	 * memory space before we make a new one.
+	 */
+	if (plan)
+		SPI_freeplan(plan);
+	if (entry->search_path)
+		pfree(entry->search_path);
+
+	entry->plan = NULL;
+	entry->search_path = NULL;
+
+	return NULL;
+}
+
+/*
+ * mv_HashPreparedPlan
+ *
+ * Add another plan to our private SPI query plan hashtable.
+ */
+static void
+mv_HashPreparedPlan(MV_QueryKey *key, SPIPlanPtr plan)
+{
+	MV_QueryHashEntry *entry;
+	bool		found;
+
+	/*
+	 * On the first call initialize the hashtable
+	 */
+	if (!mv_query_cache)
+		mv_InitHashTables();
+
+	/*
+	 * Add the new plan.  We might be overwriting an entry previously found
+	 * invalid by mv_FetchPreparedPlan.
+	 */
+	entry = (MV_QueryHashEntry *) hash_search(mv_query_cache,
+											  (void *) key,
+											  HASH_ENTER, &found);
+	Assert(!found || entry->plan == NULL);
+	entry->plan = plan;
+	entry->search_path = GetSearchPathMatcher(TopMemoryContext);
+}
+
+/*
+ * mv_BuildQueryKey
+ *
+ * Construct a hashtable key for a prepared SPI plan for IVM.
+ */
+static void
+mv_BuildQueryKey(MV_QueryKey *key, Oid matview_id, int32 query_type)
+{
+	/*
+	 * We assume struct MV_QueryKey contains no padding bytes, else we'd need
+	 * to use memset to clear them.
+	 */
+	key->matview_id = matview_id;
+	key->query_type = query_type;
+}
+
 /*
  * AtAbort_IVM
  *
-- 
2.25.1


--Multipart=_Sun__31_Mar_2024_22_59_31_+0900_msknEviJj08_wgqO
Content-Type: text/x-diff;
 name="v32-0010-Add-regression-tests-for-Incremental-View-Mainte.patch"
Content-Disposition: attachment;
 filename="v32-0010-Add-regression-tests-for-Incremental-View-Mainte.patch"
Content-Transfer-Encoding: 7bit



^ permalink  raw  reply  [nested|flat] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ 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; 371+ messages in thread

From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)

When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.

To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.

This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:

* Portability issues. Open file description lock was a non-POSIX extension in
  Linux and similar flock is from BSD standard. But looks like everybody agrees
  that such locks make more sense than a typical advisory locks, and
  F_OFD_SETLK made its way into POSIX.1 2024 [1].

* Issues with NFS. The current state of things here looks like this:

  - NFSv3 doesn't implement open file description locks, they're converted to
    advisory locks instead. Advisory locks are subject to namespace isolation,
    meaning that processes in different PID namespaces will not see each other
    advisory lock, and it's still possible to run multiple postgres
    instances on the same data directory.

  - NFSv4 uses a lease system for locking, I haven't found any mention of
    conversion to advisory locks neither in the man page nor in RFC [2].

To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.

[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
 configure                         |  14 ++++
 configure.ac                      |   3 +
 meson.build                       |   1 +
 src/backend/utils/init/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] 371+ messages in thread