agora inbox for [email protected]
help / color / mirror / Atom feed[PATCH v29 10/11] Add regression tests for Incremental View Maintenance
137+ messages / 2 participants
[nested] [flat]
* [PATCH v29 10/11] Add regression tests for Incremental View Maintenance
@ 2021-03-10 02:11 Takuma Hoshiai <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Takuma Hoshiai @ 2021-03-10 02:11 UTC (permalink / raw)
---
src/bin/pg_dump/pg_dump.h | 1 +
.../regress/expected/incremental_matview.out | 1030 +++++++++++++++++
src/test/regress/parallel_schedule | 2 +-
src/test/regress/sql/incremental_matview.sql | 533 +++++++++
4 files changed, 1565 insertions(+), 1 deletion(-)
create mode 100644 src/test/regress/expected/incremental_matview.out
create mode 100644 src/test/regress/sql/incremental_matview.sql
diff --git a/src/bin/pg_dump/pg_dump.h b/src/bin/pg_dump/pg_dump.h
index 3705891d25..71ea246abd 100644
--- a/src/bin/pg_dump/pg_dump.h
+++ b/src/bin/pg_dump/pg_dump.h
@@ -323,6 +323,7 @@ typedef struct _tableInfo
int numParents; /* number of (immediate) parent tables */
struct _tableInfo **parents; /* TableInfos of immediate parents */
+
bool isivm; /* is incrementally maintainable materialized view? */
/*
diff --git a/src/test/regress/expected/incremental_matview.out b/src/test/regress/expected/incremental_matview.out
new file mode 100644
index 0000000000..8946d09f5d
--- /dev/null
+++ b/src/test/regress/expected/incremental_matview.out
@@ -0,0 +1,1030 @@
+-- create a table to use as a basis for views and materialized views in various combinations
+CREATE TABLE mv_base_a (i int, j int);
+INSERT INTO mv_base_a VALUES
+ (1,10),
+ (2,20),
+ (3,30),
+ (4,40),
+ (5,50);
+CREATE TABLE mv_base_b (i int, k int);
+INSERT INTO mv_base_b VALUES
+ (1,101),
+ (2,102),
+ (3,103),
+ (4,104);
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_1 AS SELECT i,j,k FROM mv_base_a a INNER JOIN mv_base_b b USING(i) WITH NO DATA;
+SELECT * FROM mv_ivm_1 ORDER BY 1,2,3;
+ERROR: materialized view "mv_ivm_1" has not been populated
+HINT: Use the REFRESH MATERIALIZED VIEW command.
+REFRESH MATERIALIZED VIEW mv_ivm_1;
+NOTICE: could not create an index on materialized view "mv_ivm_1" automatically
+DETAIL: This target list does not have all the primary key columns, or this view does not contain GROUP BY or DISTINCT clause.
+HINT: Create an index on the materialized view for efficient incremental maintenance.
+SELECT * FROM mv_ivm_1 ORDER BY 1,2,3;
+ i | j | k
+---+----+-----
+ 1 | 10 | 101
+ 2 | 20 | 102
+ 3 | 30 | 103
+ 4 | 40 | 104
+(4 rows)
+
+-- REFRESH WITH NO DATA
+BEGIN;
+CREATE FUNCTION dummy_ivm_trigger_func() RETURNS TRIGGER AS $$
+ BEGIN
+ RETURN NULL;
+ END
+$$ language plpgsql;
+CREATE CONSTRAINT TRIGGER dummy_ivm_trigger AFTER INSERT
+ON mv_base_a FROM mv_ivm_1 FOR EACH ROW
+EXECUTE PROCEDURE dummy_ivm_trigger_func();
+SELECT COUNT(*)
+FROM pg_depend pd INNER JOIN pg_trigger pt ON pd.objid = pt.oid
+WHERE pd.classid = 'pg_trigger'::regclass AND pd.refobjid = 'mv_ivm_1'::regclass;
+ count
+-------
+ 17
+(1 row)
+
+REFRESH MATERIALIZED VIEW mv_ivm_1 WITH NO DATA;
+SELECT COUNT(*)
+FROM pg_depend pd INNER JOIN pg_trigger pt ON pd.objid = pt.oid
+WHERE pd.classid = 'pg_trigger'::regclass AND pd.refobjid = 'mv_ivm_1'::regclass;
+ count
+-------
+ 1
+(1 row)
+
+ROLLBACK;
+-- immediate maintenance
+BEGIN;
+INSERT INTO mv_base_b VALUES(5,105);
+SELECT * FROM mv_ivm_1 ORDER BY 1,2,3;
+ i | j | k
+---+----+-----
+ 1 | 10 | 101
+ 2 | 20 | 102
+ 3 | 30 | 103
+ 4 | 40 | 104
+ 5 | 50 | 105
+(5 rows)
+
+UPDATE mv_base_a SET j = 0 WHERE i = 1;
+SELECT * FROM mv_ivm_1 ORDER BY 1,2,3;
+ i | j | k
+---+----+-----
+ 1 | 0 | 101
+ 2 | 20 | 102
+ 3 | 30 | 103
+ 4 | 40 | 104
+ 5 | 50 | 105
+(5 rows)
+
+DELETE FROM mv_base_b WHERE (i,k) = (5,105);
+SELECT * FROM mv_ivm_1 ORDER BY 1,2,3;
+ i | j | k
+---+----+-----
+ 1 | 0 | 101
+ 2 | 20 | 102
+ 3 | 30 | 103
+ 4 | 40 | 104
+(4 rows)
+
+ROLLBACK;
+SELECT * FROM mv_ivm_1 ORDER BY 1,2,3;
+ i | j | k
+---+----+-----
+ 1 | 10 | 101
+ 2 | 20 | 102
+ 3 | 30 | 103
+ 4 | 40 | 104
+(4 rows)
+
+-- rename of IVM columns
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_rename AS SELECT DISTINCT * FROM mv_base_a;
+NOTICE: created index "mv_ivm_rename_index" on materialized view "mv_ivm_rename"
+ALTER MATERIALIZED VIEW mv_ivm_rename RENAME COLUMN __ivm_count__ TO xxx;
+ERROR: IVM column can not be renamed
+DROP MATERIALIZED VIEW mv_ivm_rename;
+-- unique index on IVM columns
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_unique AS SELECT DISTINCT * FROM mv_base_a;
+NOTICE: created index "mv_ivm_unique_index" on materialized view "mv_ivm_unique"
+CREATE UNIQUE INDEX ON mv_ivm_unique(__ivm_count__);
+ERROR: unique index creation on IVM columns is not supported
+CREATE UNIQUE INDEX ON mv_ivm_unique((__ivm_count__));
+ERROR: unique index creation on IVM columns is not supported
+CREATE UNIQUE INDEX ON mv_ivm_unique((__ivm_count__ + 1));
+ERROR: unique index creation on IVM columns is not supported
+DROP MATERIALIZED VIEW mv_ivm_unique;
+-- TRUNCATE a base table in join views
+BEGIN;
+TRUNCATE mv_base_a;
+SELECT * FROM mv_ivm_1;
+ i | j | k
+---+---+---
+(0 rows)
+
+ROLLBACK;
+BEGIN;
+TRUNCATE mv_base_b;
+SELECT * FROM mv_ivm_1;
+ i | j | k
+---+---+---
+(0 rows)
+
+ROLLBACK;
+-- some query syntax
+BEGIN;
+CREATE FUNCTION ivm_func() RETURNS int LANGUAGE 'sql'
+ AS 'SELECT 1' IMMUTABLE;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_func AS SELECT * FROM ivm_func();
+NOTICE: could not create an index on materialized view "mv_ivm_func" automatically
+DETAIL: This target list does not have all the primary key columns, or this view does not contain GROUP BY or DISTINCT clause.
+HINT: Create an index on the materialized view for efficient incremental maintenance.
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_no_tbl AS SELECT 1;
+NOTICE: could not create an index on materialized view "mv_ivm_no_tbl" automatically
+DETAIL: This target list does not have all the primary key columns, or this view does not contain GROUP BY or DISTINCT clause.
+HINT: Create an index on the materialized view for efficient incremental maintenance.
+ROLLBACK;
+-- result of materialized view have DISTINCT clause or the duplicate result.
+BEGIN;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_duplicate AS SELECT j FROM mv_base_a;
+NOTICE: could not create an index on materialized view "mv_ivm_duplicate" automatically
+DETAIL: This target list does not have all the primary key columns, or this view does not contain GROUP BY or DISTINCT clause.
+HINT: Create an index on the materialized view for efficient incremental maintenance.
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_distinct AS SELECT DISTINCT j FROM mv_base_a;
+NOTICE: created index "mv_ivm_distinct_index" on materialized view "mv_ivm_distinct"
+INSERT INTO mv_base_a VALUES(6,20);
+SELECT * FROM mv_ivm_duplicate ORDER BY 1;
+ j
+----
+ 10
+ 20
+ 20
+ 30
+ 40
+ 50
+(6 rows)
+
+SELECT * FROM mv_ivm_distinct ORDER BY 1;
+ j
+----
+ 10
+ 20
+ 30
+ 40
+ 50
+(5 rows)
+
+DELETE FROM mv_base_a WHERE (i,j) = (2,20);
+SELECT * FROM mv_ivm_duplicate ORDER BY 1;
+ j
+----
+ 10
+ 20
+ 30
+ 40
+ 50
+(5 rows)
+
+SELECT * FROM mv_ivm_distinct ORDER BY 1;
+ j
+----
+ 10
+ 20
+ 30
+ 40
+ 50
+(5 rows)
+
+ROLLBACK;
+-- support SUM(), COUNT() and AVG() aggregate functions
+BEGIN;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_agg AS SELECT i, SUM(j), COUNT(i), AVG(j) FROM mv_base_a GROUP BY i;
+NOTICE: created index "mv_ivm_agg_index" on materialized view "mv_ivm_agg"
+SELECT * FROM mv_ivm_agg ORDER BY 1,2,3,4;
+ i | sum | count | avg
+---+-----+-------+---------------------
+ 1 | 10 | 1 | 10.0000000000000000
+ 2 | 20 | 1 | 20.0000000000000000
+ 3 | 30 | 1 | 30.0000000000000000
+ 4 | 40 | 1 | 40.0000000000000000
+ 5 | 50 | 1 | 50.0000000000000000
+(5 rows)
+
+INSERT INTO mv_base_a VALUES(2,100);
+SELECT * FROM mv_ivm_agg ORDER BY 1,2,3,4;
+ i | sum | count | avg
+---+-----+-------+---------------------
+ 1 | 10 | 1 | 10.0000000000000000
+ 2 | 120 | 2 | 60.0000000000000000
+ 3 | 30 | 1 | 30.0000000000000000
+ 4 | 40 | 1 | 40.0000000000000000
+ 5 | 50 | 1 | 50.0000000000000000
+(5 rows)
+
+UPDATE mv_base_a SET j = 200 WHERE (i,j) = (2,100);
+SELECT * FROM mv_ivm_agg ORDER BY 1,2,3,4;
+ i | sum | count | avg
+---+-----+-------+----------------------
+ 1 | 10 | 1 | 10.0000000000000000
+ 2 | 220 | 2 | 110.0000000000000000
+ 3 | 30 | 1 | 30.0000000000000000
+ 4 | 40 | 1 | 40.0000000000000000
+ 5 | 50 | 1 | 50.0000000000000000
+(5 rows)
+
+DELETE FROM mv_base_a WHERE (i,j) = (2,200);
+SELECT * FROM mv_ivm_agg ORDER BY 1,2,3,4;
+ i | sum | count | avg
+---+-----+-------+---------------------
+ 1 | 10 | 1 | 10.0000000000000000
+ 2 | 20 | 1 | 20.0000000000000000
+ 3 | 30 | 1 | 30.0000000000000000
+ 4 | 40 | 1 | 40.0000000000000000
+ 5 | 50 | 1 | 50.0000000000000000
+(5 rows)
+
+ROLLBACK;
+-- support COUNT(*) aggregate function
+BEGIN;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_agg AS SELECT i, SUM(j), COUNT(*) FROM mv_base_a GROUP BY i;
+NOTICE: created index "mv_ivm_agg_index" on materialized view "mv_ivm_agg"
+SELECT * FROM mv_ivm_agg ORDER BY 1,2,3;
+ i | sum | count
+---+-----+-------
+ 1 | 10 | 1
+ 2 | 20 | 1
+ 3 | 30 | 1
+ 4 | 40 | 1
+ 5 | 50 | 1
+(5 rows)
+
+INSERT INTO mv_base_a VALUES(2,100);
+SELECT * FROM mv_ivm_agg ORDER BY 1,2,3;
+ i | sum | count
+---+-----+-------
+ 1 | 10 | 1
+ 2 | 120 | 2
+ 3 | 30 | 1
+ 4 | 40 | 1
+ 5 | 50 | 1
+(5 rows)
+
+ROLLBACK;
+-- TRUNCATE a base table in aggregate views
+BEGIN;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_agg AS SELECT i, SUM(j), COUNT(*) FROM mv_base_a GROUP BY i;
+NOTICE: created index "mv_ivm_agg_index" on materialized view "mv_ivm_agg"
+TRUNCATE mv_base_a;
+SELECT sum, count FROM mv_ivm_agg;
+ sum | count
+-----+-------
+(0 rows)
+
+SELECT i, SUM(j), COUNT(*) FROM mv_base_a GROUP BY i;
+ i | sum | count
+---+-----+-------
+(0 rows)
+
+ROLLBACK;
+-- support aggregate functions without GROUP clause
+BEGIN;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_group AS SELECT SUM(j), COUNT(j), AVG(j) FROM mv_base_a;
+NOTICE: could not create an index on materialized view "mv_ivm_group" automatically
+DETAIL: This target list does not have all the primary key columns, or this view does not contain GROUP BY or DISTINCT clause.
+HINT: Create an index on the materialized view for efficient incremental maintenance.
+SELECT * FROM mv_ivm_group ORDER BY 1;
+ sum | count | avg
+-----+-------+---------------------
+ 150 | 5 | 30.0000000000000000
+(1 row)
+
+INSERT INTO mv_base_a VALUES(6,60);
+SELECT * FROM mv_ivm_group ORDER BY 1;
+ sum | count | avg
+-----+-------+---------------------
+ 210 | 6 | 35.0000000000000000
+(1 row)
+
+DELETE FROM mv_base_a;
+SELECT * FROM mv_ivm_group ORDER BY 1;
+ sum | count | avg
+-----+-------+-----
+ | 0 |
+(1 row)
+
+ROLLBACK;
+-- TRUNCATE a base table in aggregate views without GROUP clause
+BEGIN;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_group AS SELECT SUM(j), COUNT(j), AVG(j) FROM mv_base_a;
+NOTICE: could not create an index on materialized view "mv_ivm_group" automatically
+DETAIL: This target list does not have all the primary key columns, or this view does not contain GROUP BY or DISTINCT clause.
+HINT: Create an index on the materialized view for efficient incremental maintenance.
+TRUNCATE mv_base_a;
+SELECT sum, count, avg FROM mv_ivm_group;
+ sum | count | avg
+-----+-------+-----
+ | 0 |
+(1 row)
+
+SELECT SUM(j), COUNT(j), AVG(j) FROM mv_base_a;
+ sum | count | avg
+-----+-------+-----
+ | 0 |
+(1 row)
+
+ROLLBACK;
+-- resolved issue: When use AVG() function and values is indivisible, result of AVG() is incorrect.
+BEGIN;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_avg_bug AS SELECT i, SUM(j), COUNT(j), AVG(j) FROM mv_base_A GROUP BY i;
+NOTICE: created index "mv_ivm_avg_bug_index" on materialized view "mv_ivm_avg_bug"
+SELECT * FROM mv_ivm_avg_bug ORDER BY 1,2,3;
+ i | sum | count | avg
+---+-----+-------+---------------------
+ 1 | 10 | 1 | 10.0000000000000000
+ 2 | 20 | 1 | 20.0000000000000000
+ 3 | 30 | 1 | 30.0000000000000000
+ 4 | 40 | 1 | 40.0000000000000000
+ 5 | 50 | 1 | 50.0000000000000000
+(5 rows)
+
+INSERT INTO mv_base_a VALUES
+ (1,0),
+ (1,0),
+ (2,30),
+ (2,30);
+SELECT * FROM mv_ivm_avg_bug ORDER BY 1,2,3;
+ i | sum | count | avg
+---+-----+-------+---------------------
+ 1 | 10 | 3 | 3.3333333333333333
+ 2 | 80 | 3 | 26.6666666666666667
+ 3 | 30 | 1 | 30.0000000000000000
+ 4 | 40 | 1 | 40.0000000000000000
+ 5 | 50 | 1 | 50.0000000000000000
+(5 rows)
+
+DELETE FROM mv_base_a WHERE (i,j) = (1,0);
+DELETE FROM mv_base_a WHERE (i,j) = (2,30);
+SELECT * FROM mv_ivm_avg_bug ORDER BY 1,2,3;
+ i | sum | count | avg
+---+-----+-------+---------------------
+ 1 | 10 | 1 | 10.0000000000000000
+ 2 | 20 | 1 | 20.0000000000000000
+ 3 | 30 | 1 | 30.0000000000000000
+ 4 | 40 | 1 | 40.0000000000000000
+ 5 | 50 | 1 | 50.0000000000000000
+(5 rows)
+
+ROLLBACK;
+-- support MIN(), MAX() aggregate functions
+BEGIN;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_min_max AS SELECT i, MIN(j), MAX(j) FROM mv_base_a GROUP BY i;
+NOTICE: created index "mv_ivm_min_max_index" on materialized view "mv_ivm_min_max"
+SELECT * FROM mv_ivm_min_max ORDER BY 1,2,3;
+ i | min | max
+---+-----+-----
+ 1 | 10 | 10
+ 2 | 20 | 20
+ 3 | 30 | 30
+ 4 | 40 | 40
+ 5 | 50 | 50
+(5 rows)
+
+INSERT INTO mv_base_a VALUES
+ (1,11), (1,12),
+ (2,21), (2,22),
+ (3,31), (3,32),
+ (4,41), (4,42),
+ (5,51), (5,52);
+SELECT * FROM mv_ivm_min_max ORDER BY 1,2,3;
+ i | min | max
+---+-----+-----
+ 1 | 10 | 12
+ 2 | 20 | 22
+ 3 | 30 | 32
+ 4 | 40 | 42
+ 5 | 50 | 52
+(5 rows)
+
+DELETE FROM mv_base_a WHERE (i,j) IN ((1,10), (2,21), (3,32));
+SELECT * FROM mv_ivm_min_max ORDER BY 1,2,3;
+ i | min | max
+---+-----+-----
+ 1 | 11 | 12
+ 2 | 20 | 22
+ 3 | 30 | 31
+ 4 | 40 | 42
+ 5 | 50 | 52
+(5 rows)
+
+ROLLBACK;
+-- support MIN(), MAX() aggregate functions without GROUP clause
+BEGIN;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_min_max AS SELECT MIN(j), MAX(j) FROM mv_base_a;
+NOTICE: could not create an index on materialized view "mv_ivm_min_max" automatically
+DETAIL: This target list does not have all the primary key columns, or this view does not contain GROUP BY or DISTINCT clause.
+HINT: Create an index on the materialized view for efficient incremental maintenance.
+SELECT * FROM mv_ivm_min_max;
+ min | max
+-----+-----
+ 10 | 50
+(1 row)
+
+INSERT INTO mv_base_a VALUES
+ (0,0), (6,60), (7,70);
+SELECT * FROM mv_ivm_min_max;
+ min | max
+-----+-----
+ 0 | 70
+(1 row)
+
+DELETE FROM mv_base_a WHERE (i,j) IN ((0,0), (7,70));
+SELECT * FROM mv_ivm_min_max;
+ min | max
+-----+-----
+ 10 | 60
+(1 row)
+
+DELETE FROM mv_base_a;
+SELECT * FROM mv_ivm_min_max;
+ min | max
+-----+-----
+ |
+(1 row)
+
+ROLLBACK;
+-- Test MIN/MAX after search_path change
+BEGIN;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_min AS SELECT MIN(j) FROM mv_base_a;
+NOTICE: could not create an index on materialized view "mv_ivm_min" automatically
+DETAIL: This target list does not have all the primary key columns, or this view does not contain GROUP BY or DISTINCT clause.
+HINT: Create an index on the materialized view for efficient incremental maintenance.
+SELECT * FROM mv_ivm_min;
+ min
+-----
+ 10
+(1 row)
+
+CREATE SCHEMA myschema;
+GRANT ALL ON SCHEMA myschema TO public;
+CREATE TABLE myschema.mv_base_a (j int);
+INSERT INTO myschema.mv_base_a VALUES (1);
+DELETE FROM mv_base_a WHERE (i,j) = (1,10);
+SELECT * FROM mv_ivm_min;
+ min
+-----
+ 20
+(1 row)
+
+SET search_path TO myschema,public,pg_catalog;
+DELETE FROM public.mv_base_a WHERE (i,j) = (2,20);
+SELECT * FROM mv_ivm_min;
+ min
+-----
+ 30
+(1 row)
+
+ROLLBACK;
+-- aggregate views with column names specified
+BEGIN;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_agg(a) AS SELECT i, SUM(j) FROM mv_base_a GROUP BY i;
+NOTICE: created index "mv_ivm_agg_index" on materialized view "mv_ivm_agg"
+INSERT INTO mv_base_a VALUES (1,100), (2,200), (3,300);
+UPDATE mv_base_a SET j = 2000 WHERE (i,j) = (2,20);
+DELETE FROM mv_base_a WHERE (i,j) = (3,30);
+SELECT * FROM mv_ivm_agg ORDER BY 1,2;
+ a | sum
+---+------
+ 1 | 110
+ 2 | 2200
+ 3 | 300
+ 4 | 40
+ 5 | 50
+(5 rows)
+
+ROLLBACK;
+BEGIN;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_agg(a,b) AS SELECT i, SUM(j) FROM mv_base_a GROUP BY i;
+NOTICE: created index "mv_ivm_agg_index" on materialized view "mv_ivm_agg"
+INSERT INTO mv_base_a VALUES (1,100), (2,200), (3,300);
+UPDATE mv_base_a SET j = 2000 WHERE (i,j) = (2,20);
+DELETE FROM mv_base_a WHERE (i,j) = (3,30);
+SELECT * FROM mv_ivm_agg ORDER BY 1,2;
+ a | b
+---+------
+ 1 | 110
+ 2 | 2200
+ 3 | 300
+ 4 | 40
+ 5 | 50
+(5 rows)
+
+ROLLBACK;
+BEGIN;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_agg(a,b,c) AS SELECT i, SUM(j) FROM mv_base_a GROUP BY i;
+ERROR: too many column names were specified
+ROLLBACK;
+-- support self join view and multiple change on the same table
+BEGIN;
+CREATE TABLE base_t (i int, v int);
+INSERT INTO base_t VALUES (1, 10), (2, 20), (3, 30);
+CREATE INCREMENTAL MATERIALIZED VIEW mv_self(v1, v2) AS
+ SELECT t1.v, t2.v FROM base_t AS t1 JOIN base_t AS t2 ON t1.i = t2.i;
+NOTICE: could not create an index on materialized view "mv_self" automatically
+DETAIL: This target list does not have all the primary key columns, or this view does not contain GROUP BY or DISTINCT clause.
+HINT: Create an index on the materialized view for efficient incremental maintenance.
+SELECT * FROM mv_self ORDER BY v1;
+ v1 | v2
+----+----
+ 10 | 10
+ 20 | 20
+ 30 | 30
+(3 rows)
+
+INSERT INTO base_t VALUES (4,40);
+DELETE FROM base_t WHERE i = 1;
+UPDATE base_t SET v = v*10 WHERE i=2;
+SELECT * FROM mv_self ORDER BY v1;
+ v1 | v2
+-----+-----
+ 30 | 30
+ 40 | 40
+ 200 | 200
+(3 rows)
+
+WITH
+ ins_t1 AS (INSERT INTO base_t VALUES (5,50) RETURNING 1),
+ ins_t2 AS (INSERT INTO base_t VALUES (6,60) RETURNING 1),
+ upd_t AS (UPDATE base_t SET v = v + 100 RETURNING 1),
+ dlt_t AS (DELETE FROM base_t WHERE i IN (4,5) RETURNING 1)
+SELECT NULL;
+ ?column?
+----------
+
+(1 row)
+
+SELECT * FROM mv_self ORDER BY v1;
+ v1 | v2
+-----+-----
+ 50 | 50
+ 60 | 60
+ 130 | 130
+ 300 | 300
+(4 rows)
+
+--- with sub-transactions
+SAVEPOINT p1;
+INSERT INTO base_t VALUES (7,70);
+RELEASE SAVEPOINT p1;
+INSERT INTO base_t VALUES (7,77);
+SELECT * FROM mv_self ORDER BY v1, v2;
+ v1 | v2
+-----+-----
+ 50 | 50
+ 60 | 60
+ 70 | 70
+ 70 | 77
+ 77 | 70
+ 77 | 77
+ 130 | 130
+ 300 | 300
+(8 rows)
+
+ROLLBACK;
+-- support simultaneous table changes
+BEGIN;
+CREATE TABLE base_r (i int, v int);
+CREATE TABLE base_s (i int, v int);
+INSERT INTO base_r VALUES (1, 10), (2, 20), (3, 30);
+INSERT INTO base_s VALUES (1, 100), (2, 200), (3, 300);
+CREATE INCREMENTAL MATERIALIZED VIEW mv(v1, v2) AS
+ SELECT r.v, s.v FROM base_r AS r JOIN base_s AS s USING(i);
+NOTICE: could not create an index on materialized view "mv" automatically
+DETAIL: This target list does not have all the primary key columns, or this view does not contain GROUP BY or DISTINCT clause.
+HINT: Create an index on the materialized view for efficient incremental maintenance.
+SELECT * FROM mv ORDER BY v1;
+ v1 | v2
+----+-----
+ 10 | 100
+ 20 | 200
+ 30 | 300
+(3 rows)
+
+WITH
+ ins_r AS (INSERT INTO base_r VALUES (1,11) RETURNING 1),
+ ins_r2 AS (INSERT INTO base_r VALUES (3,33) RETURNING 1),
+ ins_s AS (INSERT INTO base_s VALUES (2,222) RETURNING 1),
+ upd_r AS (UPDATE base_r SET v = v + 1000 WHERE i = 2 RETURNING 1),
+ dlt_s AS (DELETE FROM base_s WHERE i = 3 RETURNING 1)
+SELECT NULL;
+ ?column?
+----------
+
+(1 row)
+
+SELECT * FROM mv ORDER BY v1;
+ v1 | v2
+------+-----
+ 10 | 100
+ 11 | 100
+ 1020 | 200
+ 1020 | 222
+(4 rows)
+
+ROLLBACK;
+-- support foreign reference constraints
+BEGIN;
+CREATE TABLE ri1 (i int PRIMARY KEY);
+CREATE TABLE ri2 (i int PRIMARY KEY REFERENCES ri1(i) ON UPDATE CASCADE ON DELETE CASCADE, v int);
+INSERT INTO ri1 VALUES (1),(2),(3);
+INSERT INTO ri2 VALUES (1),(2),(3);
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ri(i1, i2) AS
+ SELECT ri1.i, ri2.i FROM ri1 JOIN ri2 USING(i);
+NOTICE: created index "mv_ri_index" on materialized view "mv_ri"
+SELECT * FROM mv_ri ORDER BY i1;
+ i1 | i2
+----+----
+ 1 | 1
+ 2 | 2
+ 3 | 3
+(3 rows)
+
+UPDATE ri1 SET i=10 where i=1;
+DELETE FROM ri1 WHERE i=2;
+SELECT * FROM mv_ri ORDER BY i2;
+ i1 | i2
+----+----
+ 3 | 3
+ 10 | 10
+(2 rows)
+
+ROLLBACK;
+-- views including NULL
+BEGIN;
+CREATE TABLE base_t (i int, v int);
+INSERT INTO base_t VALUES (1,10),(2, NULL);
+CREATE INCREMENTAL MATERIALIZED VIEW mv AS SELECT * FROM base_t;
+NOTICE: could not create an index on materialized view "mv" automatically
+DETAIL: This target list does not have all the primary key columns, or this view does not contain GROUP BY or DISTINCT clause.
+HINT: Create an index on the materialized view for efficient incremental maintenance.
+SELECT * FROM mv ORDER BY i;
+ i | v
+---+----
+ 1 | 10
+ 2 |
+(2 rows)
+
+UPDATE base_t SET v = 20 WHERE i = 2;
+SELECT * FROM mv ORDER BY i;
+ i | v
+---+----
+ 1 | 10
+ 2 | 20
+(2 rows)
+
+ROLLBACK;
+BEGIN;
+CREATE TABLE base_t (i int);
+CREATE INCREMENTAL MATERIALIZED VIEW mv AS SELECT * FROM base_t;
+NOTICE: could not create an index on materialized view "mv" automatically
+DETAIL: This target list does not have all the primary key columns, or this view does not contain GROUP BY or DISTINCT clause.
+HINT: Create an index on the materialized view for efficient incremental maintenance.
+SELECT * FROM mv ORDER BY i;
+ i
+---
+(0 rows)
+
+INSERT INTO base_t VALUES (1),(NULL);
+SELECT * FROM mv ORDER BY i;
+ i
+---
+ 1
+
+(2 rows)
+
+ROLLBACK;
+BEGIN;
+CREATE TABLE base_t (i int, v int);
+INSERT INTO base_t VALUES (NULL, 1), (NULL, 2), (1, 10), (1, 20);
+CREATE INCREMENTAL MATERIALIZED VIEW mv AS SELECT i, sum(v) FROM base_t GROUP BY i;
+NOTICE: created index "mv_index" on materialized view "mv"
+SELECT * FROM mv ORDER BY i;
+ i | sum
+---+-----
+ 1 | 30
+ | 3
+(2 rows)
+
+UPDATE base_t SET v = v * 10;
+SELECT * FROM mv ORDER BY i;
+ i | sum
+---+-----
+ 1 | 300
+ | 30
+(2 rows)
+
+ROLLBACK;
+BEGIN;
+CREATE TABLE base_t (i int, v int);
+INSERT INTO base_t VALUES (NULL, 1), (NULL, 2), (NULL, 3), (NULL, 4), (NULL, 5);
+CREATE INCREMENTAL MATERIALIZED VIEW mv AS SELECT i, min(v), max(v) FROM base_t GROUP BY i;
+NOTICE: created index "mv_index" on materialized view "mv"
+SELECT * FROM mv ORDER BY i;
+ i | min | max
+---+-----+-----
+ | 1 | 5
+(1 row)
+
+DELETE FROM base_t WHERE v = 1;
+SELECT * FROM mv ORDER BY i;
+ i | min | max
+---+-----+-----
+ | 2 | 5
+(1 row)
+
+DELETE FROM base_t WHERE v = 3;
+SELECT * FROM mv ORDER BY i;
+ i | min | max
+---+-----+-----
+ | 2 | 5
+(1 row)
+
+DELETE FROM base_t WHERE v = 5;
+SELECT * FROM mv ORDER BY i;
+ i | min | max
+---+-----+-----
+ | 2 | 4
+(1 row)
+
+ROLLBACK;
+-- IMMV containing user defined type
+BEGIN;
+CREATE TYPE mytype;
+CREATE FUNCTION mytype_in(cstring)
+ RETURNS mytype AS 'int4in'
+ LANGUAGE INTERNAL STRICT IMMUTABLE;
+NOTICE: return type mytype is only a shell
+CREATE FUNCTION mytype_out(mytype)
+ RETURNS cstring AS 'int4out'
+ LANGUAGE INTERNAL STRICT IMMUTABLE;
+NOTICE: argument type mytype is only a shell
+CREATE TYPE mytype (
+ LIKE = int4,
+ INPUT = mytype_in,
+ OUTPUT = mytype_out
+);
+CREATE FUNCTION mytype_eq(mytype, mytype)
+ RETURNS bool AS 'int4eq'
+ LANGUAGE INTERNAL STRICT IMMUTABLE;
+CREATE FUNCTION mytype_lt(mytype, mytype)
+ RETURNS bool AS 'int4lt'
+ LANGUAGE INTERNAL STRICT IMMUTABLE;
+CREATE FUNCTION mytype_cmp(mytype, mytype)
+ RETURNS integer AS 'btint4cmp'
+ LANGUAGE INTERNAL STRICT IMMUTABLE;
+CREATE OPERATOR = (
+ leftarg = mytype, rightarg = mytype,
+ procedure = mytype_eq);
+CREATE OPERATOR < (
+ leftarg = mytype, rightarg = mytype,
+ procedure = mytype_lt);
+CREATE OPERATOR CLASS mytype_ops
+ DEFAULT FOR TYPE mytype USING btree AS
+ OPERATOR 1 <,
+ OPERATOR 3 = ,
+ FUNCTION 1 mytype_cmp(mytype,mytype);
+CREATE TABLE t_mytype (x mytype);
+CREATE INCREMENTAL MATERIALIZED VIEW mv_mytype AS
+ SELECT * FROM t_mytype;
+NOTICE: could not create an index on materialized view "mv_mytype" automatically
+DETAIL: This target list does not have all the primary key columns, or this view does not contain GROUP BY or DISTINCT clause.
+HINT: Create an index on the materialized view for efficient incremental maintenance.
+INSERT INTO t_mytype VALUES ('1'::mytype);
+SELECT * FROM mv_mytype;
+ x
+---
+ 1
+(1 row)
+
+ROLLBACK;
+-- outer join is not supported
+CREATE INCREMENTAL MATERIALIZED VIEW mv(a,b) AS SELECT a.i, b.i FROM mv_base_a a LEFT JOIN mv_base_b b ON a.i=b.i;
+ERROR: OUTER JOIN is not supported on incrementally maintainable materialized view
+-- CTE is not supported
+CREATE INCREMENTAL MATERIALIZED VIEW mv AS
+ WITH b AS ( SELECT * FROM mv_base_b) SELECT a.i,a.j FROM mv_base_a a, b WHERE a.i = b.i;
+ERROR: CTE is not supported on incrementally maintainable materialized view
+-- contain system column
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm01 AS SELECT i,j,xmin FROM mv_base_a;
+ERROR: system column is not supported on incrementally maintainable materialized view
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm02 AS SELECT i,j FROM mv_base_a WHERE xmin = '610';
+ERROR: system column is not supported on incrementally maintainable materialized view
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm04 AS SELECT i,j,xmin::text AS x_min FROM mv_base_a;
+ERROR: system column is not supported on incrementally maintainable materialized view
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm06 AS SELECT i,j,xidsend(xmin) AS x_min FROM mv_base_a;
+ERROR: system column is not supported on incrementally maintainable materialized view
+-- contain subquery
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm03 AS SELECT i,j FROM mv_base_a WHERE i IN (SELECT i FROM mv_base_b WHERE k < 103 );
+ERROR: subquery is not supported on incrementally maintainable materialized view
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm04 AS SELECT a.i,a.j FROM mv_base_a a, (SELECT * FROM mv_base_b) b WHERE a.i = b.i;
+ERROR: subquery is not supported on incrementally maintainable materialized view
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm05 AS SELECT i,j, (SELECT k FROM mv_base_b b WHERE a.i = b.i) FROM mv_base_a a;
+ERROR: subquery is not supported on incrementally maintainable materialized view
+-- contain ORDER BY
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm07 AS SELECT i,j,k FROM mv_base_a a INNER JOIN mv_base_b b USING(i) ORDER BY i,j,k;
+ERROR: ORDER BY clause is not supported on incrementally maintainable materialized view
+-- contain HAVING
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm08 AS SELECT i,j,k FROM mv_base_a a INNER JOIN mv_base_b b USING(i) GROUP BY i,j,k HAVING SUM(i) > 5;
+ERROR: HAVING clause is not supported on incrementally maintainable materialized view
+-- contain view or materialized view
+CREATE VIEW b_view AS SELECT i,k FROM mv_base_b;
+CREATE MATERIALIZED VIEW b_mview AS SELECT i,k FROM mv_base_b;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm07 AS SELECT a.i,a.j FROM mv_base_a a,b_view b WHERE a.i = b.i;
+ERROR: VIEW or MATERIALIZED VIEW is not supported on incrementally maintainable materialized view
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm08 AS SELECT a.i,a.j FROM mv_base_a a,b_mview b WHERE a.i = b.i;
+ERROR: VIEW or MATERIALIZED VIEW is not supported on incrementally maintainable materialized view
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm09 AS SELECT a.i,a.j FROM mv_base_a a, (SELECT i, COUNT(*) FROM mv_base_b GROUP BY i) b WHERE a.i = b.i;
+ERROR: subquery is not supported on incrementally maintainable materialized view
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm10 AS SELECT a.i,a.j FROM mv_base_a a WHERE EXISTS(SELECT 1 FROM mv_base_b b WHERE a.i = b.i) OR a.i > 5;
+ERROR: subquery is not supported on incrementally maintainable materialized view
+-- contain mutable functions
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm12 AS SELECT i,j FROM mv_base_a WHERE i = random()::int;
+ERROR: mutable function is not supported on incrementally maintainable materialized view
+HINT: functions must be marked IMMUTABLE
+-- LIMIT/OFFSET is not supported
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm13 AS SELECT i,j FROM mv_base_a LIMIT 10 OFFSET 5;
+ERROR: LIMIT/OFFSET clause is not supported on incrementally maintainable materialized view
+-- DISTINCT ON is not supported
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm14 AS SELECT DISTINCT ON(i) i, j FROM mv_base_a;
+ERROR: DISTINCT ON is not supported on incrementally maintainable materialized view
+-- TABLESAMPLE clause is not supported
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm15 AS SELECT i, j FROM mv_base_a TABLESAMPLE SYSTEM(50);
+ERROR: TABLESAMPLE clause is not supported on incrementally maintainable materialized view
+-- window functions are not supported
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm16 AS SELECT *, cume_dist() OVER (ORDER BY i) AS rank FROM mv_base_a;
+ERROR: window functions are not supported on incrementally maintainable materialized view
+-- aggregate function with some options is not supported
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm17 AS SELECT COUNT(*) FILTER(WHERE i < 3) FROM mv_base_a;
+ERROR: aggregate function with FILTER clause is not supported on incrementally maintainable materialized view
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm18 AS SELECT COUNT(DISTINCT i) FROM mv_base_a;
+ERROR: aggregate function with DISTINCT arguments is not supported on incrementally maintainable materialized view
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm19 AS SELECT array_agg(j ORDER BY i DESC) FROM mv_base_a;
+ERROR: aggregate function with ORDER clause is not supported on incrementally maintainable materialized view
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm20 AS SELECT i,SUM(j) FROM mv_base_a GROUP BY GROUPING SETS((i),());
+ERROR: GROUPING SETS, ROLLUP, or CUBE clauses is not supported on incrementally maintainable materialized view
+-- inheritance parent is not supported
+BEGIN;
+CREATE TABLE parent (i int, v int);
+CREATE TABLE child_a(options text) INHERITS(parent);
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm21 AS SELECT * FROM parent;
+ERROR: inheritance parent is not supported on incrementally maintainable materialized view
+ROLLBACK;
+-- UNION statement is not supported
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm22 AS SELECT i,j FROM mv_base_a UNION ALL SELECT i,k FROM mv_base_b;;
+ERROR: UNION/INTERSECT/EXCEPT statements are not supported on incrementally maintainable materialized view
+-- empty target list is not allowed with IVM
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm25 AS SELECT FROM mv_base_a;
+ERROR: empty target list is not supported on incrementally maintainable materialized view
+-- FOR UPDATE/SHARE is not supported
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm26 AS SELECT i,j FROM mv_base_a FOR UPDATE;
+ERROR: FOR UPDATE/SHARE clause is not supported on incrementally maintainable materialized view
+-- tartget list cannot contain ivm column that start with '__ivm'
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm28 AS SELECT i AS "__ivm_count__" FROM mv_base_a;
+ERROR: column name __ivm_count__ is not supported on incrementally maintainable materialized view
+-- expressions specified in GROUP BY must appear in the target list.
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm29 AS SELECT COUNT(i) FROM mv_base_a GROUP BY i;
+ERROR: GROUP BY expression not appearing in select list is not supported on incrementally maintainable materialized view
+-- experssions containing an aggregate is not supported
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm30 AS SELECT sum(i)*0.5 FROM mv_base_a;
+ERROR: expression containing an aggregate in it is not supported on incrementally maintainable materialized view
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm31 AS SELECT sum(i)/sum(j) FROM mv_base_a;
+ERROR: expression containing an aggregate in it is not supported on incrementally maintainable materialized view
+-- VALUES is not supported
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_only_values1 AS values(1);
+ERROR: VALUES is not supported on incrementally maintainable materialized view
+-- views containing base tables with Row Level Security
+DROP USER IF EXISTS ivm_admin;
+NOTICE: role "ivm_admin" does not exist, skipping
+DROP USER IF EXISTS ivm_user;
+NOTICE: role "ivm_user" does not exist, skipping
+CREATE USER ivm_admin;
+CREATE USER ivm_user;
+--- create a table with RLS
+SET SESSION AUTHORIZATION ivm_admin;
+CREATE TABLE rls_tbl(id int, data text, owner name);
+INSERT INTO rls_tbl VALUES
+ (1,'foo','ivm_user'),
+ (2,'bar','postgres');
+CREATE TABLE num_tbl(id int, num text);
+INSERT INTO num_tbl VALUES
+ (1,'one'),
+ (2,'two'),
+ (3,'three'),
+ (4,'four'),
+ (5,'five'),
+ (6,'six');
+--- Users can access only their own rows
+CREATE POLICY rls_tbl_policy ON rls_tbl FOR SELECT TO PUBLIC USING(owner = current_user);
+ALTER TABLE rls_tbl ENABLE ROW LEVEL SECURITY;
+GRANT ALL on rls_tbl TO PUBLIC;
+GRANT ALL on num_tbl TO PUBLIC;
+--- create a view owned by ivm_user
+SET SESSION AUTHORIZATION ivm_user;
+CREATE INCREMENTAL MATERIALIZED VIEW ivm_rls AS SELECT * FROM rls_tbl;
+NOTICE: could not create an index on materialized view "ivm_rls" automatically
+DETAIL: This target list does not have all the primary key columns, or this view does not contain GROUP BY or DISTINCT clause.
+HINT: Create an index on the materialized view for efficient incremental maintenance.
+SELECT id, data, owner FROM ivm_rls ORDER BY 1,2,3;
+ id | data | owner
+----+------+----------
+ 1 | foo | ivm_user
+(1 row)
+
+RESET SESSION AUTHORIZATION;
+--- inserts rows owned by different users
+INSERT INTO rls_tbl VALUES
+ (3,'baz','ivm_user'),
+ (4,'qux','postgres');
+SELECT id, data, owner FROM ivm_rls ORDER BY 1,2,3;
+ id | data | owner
+----+------+----------
+ 1 | foo | ivm_user
+ 3 | baz | ivm_user
+(2 rows)
+
+--- combination of diffent kinds of commands
+WITH
+ i AS (INSERT INTO rls_tbl VALUES(5,'quux','postgres'), (6,'corge','ivm_user')),
+ u AS (UPDATE rls_tbl SET owner = 'postgres' WHERE id = 1),
+ u2 AS (UPDATE rls_tbl SET owner = 'ivm_user' WHERE id = 2)
+SELECT;
+--
+(1 row)
+
+SELECT id, data, owner FROM ivm_rls ORDER BY 1,2,3;
+ id | data | owner
+----+-------+----------
+ 2 | bar | ivm_user
+ 3 | baz | ivm_user
+ 6 | corge | ivm_user
+(3 rows)
+
+---
+SET SESSION AUTHORIZATION ivm_user;
+CREATE INCREMENTAL MATERIALIZED VIEW ivm_rls2 AS SELECT * FROM rls_tbl JOIN num_tbl USING(id);
+NOTICE: could not create an index on materialized view "ivm_rls2" automatically
+DETAIL: This target list does not have all the primary key columns, or this view does not contain GROUP BY or DISTINCT clause.
+HINT: Create an index on the materialized view for efficient incremental maintenance.
+RESET SESSION AUTHORIZATION;
+WITH
+ x AS (UPDATE rls_tbl SET data = data || '_2' where id in (3,4)),
+ y AS (UPDATE num_tbl SET num = num || '_2' where id in (3,4))
+SELECT;
+--
+(1 row)
+
+SELECT * FROM ivm_rls2 ORDER BY 1,2,3;
+ id | data | owner | num
+----+-------+----------+---------
+ 2 | bar | ivm_user | two
+ 3 | baz_2 | ivm_user | three_2
+ 6 | corge | ivm_user | six
+(3 rows)
+
+-- automatic index creation
+CREATE TABLE base_a (i int primary key, j int);
+CREATE TABLE base_b (i int primary key, j int);
+--- group by: create an index
+CREATE INCREMENTAL MATERIALIZED VIEW mv_idx1 AS SELECT i, sum(j) FROM base_a GROUP BY i;
+NOTICE: created index "mv_idx1_index" on materialized view "mv_idx1"
+--- distinct: create an index
+CREATE INCREMENTAL MATERIALIZED VIEW mv_idx2 AS SELECT DISTINCT j FROM base_a;
+NOTICE: created index "mv_idx2_index" on materialized view "mv_idx2"
+--- with all pkey columns: create an index
+CREATE INCREMENTAL MATERIALIZED VIEW mv_idx3(i_a, i_b) AS SELECT a.i, b.i FROM base_a a, base_b b;
+NOTICE: created index "mv_idx3_index" on materialized view "mv_idx3"
+--- missing some pkey columns: no index
+CREATE INCREMENTAL MATERIALIZED VIEW mv_idx4 AS SELECT j FROM base_a;
+NOTICE: could not create an index on materialized view "mv_idx4" automatically
+DETAIL: This target list does not have all the primary key columns, or this view does not contain GROUP BY or DISTINCT clause.
+HINT: Create an index on the materialized view for efficient incremental maintenance.
+CREATE INCREMENTAL MATERIALIZED VIEW mv_idx5 AS SELECT a.i, b.j FROM base_a a, base_b b;
+NOTICE: could not create an index on materialized view "mv_idx5" automatically
+DETAIL: This target list does not have all the primary key columns, or this view does not contain GROUP BY or DISTINCT clause.
+HINT: Create an index on the materialized view for efficient incremental maintenance.
+-- cleanup
+DROP TABLE rls_tbl CASCADE;
+NOTICE: drop cascades to 2 other objects
+DETAIL: drop cascades to materialized view ivm_rls
+drop cascades to materialized view ivm_rls2
+DROP TABLE num_tbl CASCADE;
+DROP USER ivm_user;
+DROP USER ivm_admin;
+DROP TABLE mv_base_b CASCADE;
+NOTICE: drop cascades to 3 other objects
+DETAIL: drop cascades to materialized view mv_ivm_1
+drop cascades to view b_view
+drop cascades to materialized view b_mview
+DROP TABLE mv_base_a CASCADE;
diff --git a/src/test/regress/parallel_schedule b/src/test/regress/parallel_schedule
index 4df9d8503b..21f7247a07 100644
--- a/src/test/regress/parallel_schedule
+++ b/src/test/regress/parallel_schedule
@@ -78,7 +78,7 @@ test: brin_bloom brin_multi
# psql depends on create_am
# amutils depends on geometry, create_index_spgist, hash_index, brin
# ----------
-test: create_table_like alter_generic alter_operator misc async dbsize merge misc_functions sysviews tsrf tid tidscan tidrangescan collate.icu.utf8 incremental_sort create_role
+test: create_table_like alter_generic alter_operator misc async dbsize merge misc_functions sysviews tsrf tid tidscan tidrangescan collate.icu.utf8 incremental_sort create_role incremental_matview
# collate.*.utf8 tests cannot be run in parallel with each other
test: rules psql psql_crosstab amutils stats_ext collate.linux.utf8 collate.windows.win1252
diff --git a/src/test/regress/sql/incremental_matview.sql b/src/test/regress/sql/incremental_matview.sql
new file mode 100644
index 0000000000..82686f9324
--- /dev/null
+++ b/src/test/regress/sql/incremental_matview.sql
@@ -0,0 +1,533 @@
+-- create a table to use as a basis for views and materialized views in various combinations
+CREATE TABLE mv_base_a (i int, j int);
+INSERT INTO mv_base_a VALUES
+ (1,10),
+ (2,20),
+ (3,30),
+ (4,40),
+ (5,50);
+CREATE TABLE mv_base_b (i int, k int);
+INSERT INTO mv_base_b VALUES
+ (1,101),
+ (2,102),
+ (3,103),
+ (4,104);
+
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_1 AS SELECT i,j,k FROM mv_base_a a INNER JOIN mv_base_b b USING(i) WITH NO DATA;
+SELECT * FROM mv_ivm_1 ORDER BY 1,2,3;
+REFRESH MATERIALIZED VIEW mv_ivm_1;
+SELECT * FROM mv_ivm_1 ORDER BY 1,2,3;
+
+-- REFRESH WITH NO DATA
+BEGIN;
+CREATE FUNCTION dummy_ivm_trigger_func() RETURNS TRIGGER AS $$
+ BEGIN
+ RETURN NULL;
+ END
+$$ language plpgsql;
+
+CREATE CONSTRAINT TRIGGER dummy_ivm_trigger AFTER INSERT
+ON mv_base_a FROM mv_ivm_1 FOR EACH ROW
+EXECUTE PROCEDURE dummy_ivm_trigger_func();
+
+SELECT COUNT(*)
+FROM pg_depend pd INNER JOIN pg_trigger pt ON pd.objid = pt.oid
+WHERE pd.classid = 'pg_trigger'::regclass AND pd.refobjid = 'mv_ivm_1'::regclass;
+
+REFRESH MATERIALIZED VIEW mv_ivm_1 WITH NO DATA;
+
+SELECT COUNT(*)
+FROM pg_depend pd INNER JOIN pg_trigger pt ON pd.objid = pt.oid
+WHERE pd.classid = 'pg_trigger'::regclass AND pd.refobjid = 'mv_ivm_1'::regclass;
+ROLLBACK;
+
+-- immediate maintenance
+BEGIN;
+INSERT INTO mv_base_b VALUES(5,105);
+SELECT * FROM mv_ivm_1 ORDER BY 1,2,3;
+UPDATE mv_base_a SET j = 0 WHERE i = 1;
+SELECT * FROM mv_ivm_1 ORDER BY 1,2,3;
+DELETE FROM mv_base_b WHERE (i,k) = (5,105);
+SELECT * FROM mv_ivm_1 ORDER BY 1,2,3;
+ROLLBACK;
+SELECT * FROM mv_ivm_1 ORDER BY 1,2,3;
+
+-- rename of IVM columns
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_rename AS SELECT DISTINCT * FROM mv_base_a;
+ALTER MATERIALIZED VIEW mv_ivm_rename RENAME COLUMN __ivm_count__ TO xxx;
+DROP MATERIALIZED VIEW mv_ivm_rename;
+
+-- unique index on IVM columns
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_unique AS SELECT DISTINCT * FROM mv_base_a;
+CREATE UNIQUE INDEX ON mv_ivm_unique(__ivm_count__);
+CREATE UNIQUE INDEX ON mv_ivm_unique((__ivm_count__));
+CREATE UNIQUE INDEX ON mv_ivm_unique((__ivm_count__ + 1));
+DROP MATERIALIZED VIEW mv_ivm_unique;
+
+-- TRUNCATE a base table in join views
+BEGIN;
+TRUNCATE mv_base_a;
+SELECT * FROM mv_ivm_1;
+ROLLBACK;
+
+BEGIN;
+TRUNCATE mv_base_b;
+SELECT * FROM mv_ivm_1;
+ROLLBACK;
+
+-- some query syntax
+BEGIN;
+CREATE FUNCTION ivm_func() RETURNS int LANGUAGE 'sql'
+ AS 'SELECT 1' IMMUTABLE;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_func AS SELECT * FROM ivm_func();
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_no_tbl AS SELECT 1;
+ROLLBACK;
+
+-- result of materialized view have DISTINCT clause or the duplicate result.
+BEGIN;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_duplicate AS SELECT j FROM mv_base_a;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_distinct AS SELECT DISTINCT j FROM mv_base_a;
+INSERT INTO mv_base_a VALUES(6,20);
+SELECT * FROM mv_ivm_duplicate ORDER BY 1;
+SELECT * FROM mv_ivm_distinct ORDER BY 1;
+DELETE FROM mv_base_a WHERE (i,j) = (2,20);
+SELECT * FROM mv_ivm_duplicate ORDER BY 1;
+SELECT * FROM mv_ivm_distinct ORDER BY 1;
+ROLLBACK;
+
+-- support SUM(), COUNT() and AVG() aggregate functions
+BEGIN;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_agg AS SELECT i, SUM(j), COUNT(i), AVG(j) FROM mv_base_a GROUP BY i;
+SELECT * FROM mv_ivm_agg ORDER BY 1,2,3,4;
+INSERT INTO mv_base_a VALUES(2,100);
+SELECT * FROM mv_ivm_agg ORDER BY 1,2,3,4;
+UPDATE mv_base_a SET j = 200 WHERE (i,j) = (2,100);
+SELECT * FROM mv_ivm_agg ORDER BY 1,2,3,4;
+DELETE FROM mv_base_a WHERE (i,j) = (2,200);
+SELECT * FROM mv_ivm_agg ORDER BY 1,2,3,4;
+ROLLBACK;
+
+-- support COUNT(*) aggregate function
+BEGIN;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_agg AS SELECT i, SUM(j), COUNT(*) FROM mv_base_a GROUP BY i;
+SELECT * FROM mv_ivm_agg ORDER BY 1,2,3;
+INSERT INTO mv_base_a VALUES(2,100);
+SELECT * FROM mv_ivm_agg ORDER BY 1,2,3;
+ROLLBACK;
+
+-- TRUNCATE a base table in aggregate views
+BEGIN;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_agg AS SELECT i, SUM(j), COUNT(*) FROM mv_base_a GROUP BY i;
+TRUNCATE mv_base_a;
+SELECT sum, count FROM mv_ivm_agg;
+SELECT i, SUM(j), COUNT(*) FROM mv_base_a GROUP BY i;
+ROLLBACK;
+
+-- support aggregate functions without GROUP clause
+BEGIN;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_group AS SELECT SUM(j), COUNT(j), AVG(j) FROM mv_base_a;
+SELECT * FROM mv_ivm_group ORDER BY 1;
+INSERT INTO mv_base_a VALUES(6,60);
+SELECT * FROM mv_ivm_group ORDER BY 1;
+DELETE FROM mv_base_a;
+SELECT * FROM mv_ivm_group ORDER BY 1;
+ROLLBACK;
+
+-- TRUNCATE a base table in aggregate views without GROUP clause
+BEGIN;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_group AS SELECT SUM(j), COUNT(j), AVG(j) FROM mv_base_a;
+TRUNCATE mv_base_a;
+SELECT sum, count, avg FROM mv_ivm_group;
+SELECT SUM(j), COUNT(j), AVG(j) FROM mv_base_a;
+ROLLBACK;
+
+-- resolved issue: When use AVG() function and values is indivisible, result of AVG() is incorrect.
+BEGIN;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_avg_bug AS SELECT i, SUM(j), COUNT(j), AVG(j) FROM mv_base_A GROUP BY i;
+SELECT * FROM mv_ivm_avg_bug ORDER BY 1,2,3;
+INSERT INTO mv_base_a VALUES
+ (1,0),
+ (1,0),
+ (2,30),
+ (2,30);
+SELECT * FROM mv_ivm_avg_bug ORDER BY 1,2,3;
+DELETE FROM mv_base_a WHERE (i,j) = (1,0);
+DELETE FROM mv_base_a WHERE (i,j) = (2,30);
+SELECT * FROM mv_ivm_avg_bug ORDER BY 1,2,3;
+ROLLBACK;
+
+-- support MIN(), MAX() aggregate functions
+BEGIN;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_min_max AS SELECT i, MIN(j), MAX(j) FROM mv_base_a GROUP BY i;
+SELECT * FROM mv_ivm_min_max ORDER BY 1,2,3;
+INSERT INTO mv_base_a VALUES
+ (1,11), (1,12),
+ (2,21), (2,22),
+ (3,31), (3,32),
+ (4,41), (4,42),
+ (5,51), (5,52);
+SELECT * FROM mv_ivm_min_max ORDER BY 1,2,3;
+DELETE FROM mv_base_a WHERE (i,j) IN ((1,10), (2,21), (3,32));
+SELECT * FROM mv_ivm_min_max ORDER BY 1,2,3;
+ROLLBACK;
+
+-- support MIN(), MAX() aggregate functions without GROUP clause
+BEGIN;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_min_max AS SELECT MIN(j), MAX(j) FROM mv_base_a;
+SELECT * FROM mv_ivm_min_max;
+INSERT INTO mv_base_a VALUES
+ (0,0), (6,60), (7,70);
+SELECT * FROM mv_ivm_min_max;
+DELETE FROM mv_base_a WHERE (i,j) IN ((0,0), (7,70));
+SELECT * FROM mv_ivm_min_max;
+DELETE FROM mv_base_a;
+SELECT * FROM mv_ivm_min_max;
+ROLLBACK;
+
+-- Test MIN/MAX after search_path change
+BEGIN;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_min AS SELECT MIN(j) FROM mv_base_a;
+SELECT * FROM mv_ivm_min;
+
+CREATE SCHEMA myschema;
+GRANT ALL ON SCHEMA myschema TO public;
+CREATE TABLE myschema.mv_base_a (j int);
+INSERT INTO myschema.mv_base_a VALUES (1);
+
+DELETE FROM mv_base_a WHERE (i,j) = (1,10);
+SELECT * FROM mv_ivm_min;
+
+SET search_path TO myschema,public,pg_catalog;
+DELETE FROM public.mv_base_a WHERE (i,j) = (2,20);
+SELECT * FROM mv_ivm_min;
+ROLLBACK;
+
+-- aggregate views with column names specified
+BEGIN;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_agg(a) AS SELECT i, SUM(j) FROM mv_base_a GROUP BY i;
+INSERT INTO mv_base_a VALUES (1,100), (2,200), (3,300);
+UPDATE mv_base_a SET j = 2000 WHERE (i,j) = (2,20);
+DELETE FROM mv_base_a WHERE (i,j) = (3,30);
+SELECT * FROM mv_ivm_agg ORDER BY 1,2;
+ROLLBACK;
+BEGIN;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_agg(a,b) AS SELECT i, SUM(j) FROM mv_base_a GROUP BY i;
+INSERT INTO mv_base_a VALUES (1,100), (2,200), (3,300);
+UPDATE mv_base_a SET j = 2000 WHERE (i,j) = (2,20);
+DELETE FROM mv_base_a WHERE (i,j) = (3,30);
+SELECT * FROM mv_ivm_agg ORDER BY 1,2;
+ROLLBACK;
+BEGIN;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_agg(a,b,c) AS SELECT i, SUM(j) FROM mv_base_a GROUP BY i;
+ROLLBACK;
+
+-- support self join view and multiple change on the same table
+BEGIN;
+CREATE TABLE base_t (i int, v int);
+INSERT INTO base_t VALUES (1, 10), (2, 20), (3, 30);
+CREATE INCREMENTAL MATERIALIZED VIEW mv_self(v1, v2) AS
+ SELECT t1.v, t2.v FROM base_t AS t1 JOIN base_t AS t2 ON t1.i = t2.i;
+SELECT * FROM mv_self ORDER BY v1;
+INSERT INTO base_t VALUES (4,40);
+DELETE FROM base_t WHERE i = 1;
+UPDATE base_t SET v = v*10 WHERE i=2;
+SELECT * FROM mv_self ORDER BY v1;
+WITH
+ ins_t1 AS (INSERT INTO base_t VALUES (5,50) RETURNING 1),
+ ins_t2 AS (INSERT INTO base_t VALUES (6,60) RETURNING 1),
+ upd_t AS (UPDATE base_t SET v = v + 100 RETURNING 1),
+ dlt_t AS (DELETE FROM base_t WHERE i IN (4,5) RETURNING 1)
+SELECT NULL;
+SELECT * FROM mv_self ORDER BY v1;
+
+--- with sub-transactions
+SAVEPOINT p1;
+INSERT INTO base_t VALUES (7,70);
+RELEASE SAVEPOINT p1;
+INSERT INTO base_t VALUES (7,77);
+SELECT * FROM mv_self ORDER BY v1, v2;
+
+ROLLBACK;
+
+-- support simultaneous table changes
+BEGIN;
+CREATE TABLE base_r (i int, v int);
+CREATE TABLE base_s (i int, v int);
+INSERT INTO base_r VALUES (1, 10), (2, 20), (3, 30);
+INSERT INTO base_s VALUES (1, 100), (2, 200), (3, 300);
+CREATE INCREMENTAL MATERIALIZED VIEW mv(v1, v2) AS
+ SELECT r.v, s.v FROM base_r AS r JOIN base_s AS s USING(i);
+SELECT * FROM mv ORDER BY v1;
+WITH
+ ins_r AS (INSERT INTO base_r VALUES (1,11) RETURNING 1),
+ ins_r2 AS (INSERT INTO base_r VALUES (3,33) RETURNING 1),
+ ins_s AS (INSERT INTO base_s VALUES (2,222) RETURNING 1),
+ upd_r AS (UPDATE base_r SET v = v + 1000 WHERE i = 2 RETURNING 1),
+ dlt_s AS (DELETE FROM base_s WHERE i = 3 RETURNING 1)
+SELECT NULL;
+SELECT * FROM mv ORDER BY v1;
+ROLLBACK;
+
+-- support foreign reference constraints
+BEGIN;
+CREATE TABLE ri1 (i int PRIMARY KEY);
+CREATE TABLE ri2 (i int PRIMARY KEY REFERENCES ri1(i) ON UPDATE CASCADE ON DELETE CASCADE, v int);
+INSERT INTO ri1 VALUES (1),(2),(3);
+INSERT INTO ri2 VALUES (1),(2),(3);
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ri(i1, i2) AS
+ SELECT ri1.i, ri2.i FROM ri1 JOIN ri2 USING(i);
+SELECT * FROM mv_ri ORDER BY i1;
+UPDATE ri1 SET i=10 where i=1;
+DELETE FROM ri1 WHERE i=2;
+SELECT * FROM mv_ri ORDER BY i2;
+ROLLBACK;
+
+-- views including NULL
+BEGIN;
+CREATE TABLE base_t (i int, v int);
+INSERT INTO base_t VALUES (1,10),(2, NULL);
+CREATE INCREMENTAL MATERIALIZED VIEW mv AS SELECT * FROM base_t;
+SELECT * FROM mv ORDER BY i;
+UPDATE base_t SET v = 20 WHERE i = 2;
+SELECT * FROM mv ORDER BY i;
+ROLLBACK;
+
+BEGIN;
+CREATE TABLE base_t (i int);
+CREATE INCREMENTAL MATERIALIZED VIEW mv AS SELECT * FROM base_t;
+SELECT * FROM mv ORDER BY i;
+INSERT INTO base_t VALUES (1),(NULL);
+SELECT * FROM mv ORDER BY i;
+ROLLBACK;
+
+BEGIN;
+CREATE TABLE base_t (i int, v int);
+INSERT INTO base_t VALUES (NULL, 1), (NULL, 2), (1, 10), (1, 20);
+CREATE INCREMENTAL MATERIALIZED VIEW mv AS SELECT i, sum(v) FROM base_t GROUP BY i;
+SELECT * FROM mv ORDER BY i;
+UPDATE base_t SET v = v * 10;
+SELECT * FROM mv ORDER BY i;
+ROLLBACK;
+
+BEGIN;
+CREATE TABLE base_t (i int, v int);
+INSERT INTO base_t VALUES (NULL, 1), (NULL, 2), (NULL, 3), (NULL, 4), (NULL, 5);
+CREATE INCREMENTAL MATERIALIZED VIEW mv AS SELECT i, min(v), max(v) FROM base_t GROUP BY i;
+SELECT * FROM mv ORDER BY i;
+DELETE FROM base_t WHERE v = 1;
+SELECT * FROM mv ORDER BY i;
+DELETE FROM base_t WHERE v = 3;
+SELECT * FROM mv ORDER BY i;
+DELETE FROM base_t WHERE v = 5;
+SELECT * FROM mv ORDER BY i;
+ROLLBACK;
+
+-- IMMV containing user defined type
+BEGIN;
+
+CREATE TYPE mytype;
+CREATE FUNCTION mytype_in(cstring)
+ RETURNS mytype AS 'int4in'
+ LANGUAGE INTERNAL STRICT IMMUTABLE;
+CREATE FUNCTION mytype_out(mytype)
+ RETURNS cstring AS 'int4out'
+ LANGUAGE INTERNAL STRICT IMMUTABLE;
+CREATE TYPE mytype (
+ LIKE = int4,
+ INPUT = mytype_in,
+ OUTPUT = mytype_out
+);
+
+CREATE FUNCTION mytype_eq(mytype, mytype)
+ RETURNS bool AS 'int4eq'
+ LANGUAGE INTERNAL STRICT IMMUTABLE;
+CREATE FUNCTION mytype_lt(mytype, mytype)
+ RETURNS bool AS 'int4lt'
+ LANGUAGE INTERNAL STRICT IMMUTABLE;
+CREATE FUNCTION mytype_cmp(mytype, mytype)
+ RETURNS integer AS 'btint4cmp'
+ LANGUAGE INTERNAL STRICT IMMUTABLE;
+
+CREATE OPERATOR = (
+ leftarg = mytype, rightarg = mytype,
+ procedure = mytype_eq);
+CREATE OPERATOR < (
+ leftarg = mytype, rightarg = mytype,
+ procedure = mytype_lt);
+
+CREATE OPERATOR CLASS mytype_ops
+ DEFAULT FOR TYPE mytype USING btree AS
+ OPERATOR 1 <,
+ OPERATOR 3 = ,
+ FUNCTION 1 mytype_cmp(mytype,mytype);
+
+CREATE TABLE t_mytype (x mytype);
+CREATE INCREMENTAL MATERIALIZED VIEW mv_mytype AS
+ SELECT * FROM t_mytype;
+INSERT INTO t_mytype VALUES ('1'::mytype);
+SELECT * FROM mv_mytype;
+
+ROLLBACK;
+
+-- outer join is not supported
+CREATE INCREMENTAL MATERIALIZED VIEW mv(a,b) AS SELECT a.i, b.i FROM mv_base_a a LEFT JOIN mv_base_b b ON a.i=b.i;
+-- CTE is not supported
+CREATE INCREMENTAL MATERIALIZED VIEW mv AS
+ WITH b AS ( SELECT * FROM mv_base_b) SELECT a.i,a.j FROM mv_base_a a, b WHERE a.i = b.i;
+-- contain system column
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm01 AS SELECT i,j,xmin FROM mv_base_a;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm02 AS SELECT i,j FROM mv_base_a WHERE xmin = '610';
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm04 AS SELECT i,j,xmin::text AS x_min FROM mv_base_a;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm06 AS SELECT i,j,xidsend(xmin) AS x_min FROM mv_base_a;
+-- contain subquery
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm03 AS SELECT i,j FROM mv_base_a WHERE i IN (SELECT i FROM mv_base_b WHERE k < 103 );
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm04 AS SELECT a.i,a.j FROM mv_base_a a, (SELECT * FROM mv_base_b) b WHERE a.i = b.i;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm05 AS SELECT i,j, (SELECT k FROM mv_base_b b WHERE a.i = b.i) FROM mv_base_a a;
+-- contain ORDER BY
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm07 AS SELECT i,j,k FROM mv_base_a a INNER JOIN mv_base_b b USING(i) ORDER BY i,j,k;
+-- contain HAVING
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm08 AS SELECT i,j,k FROM mv_base_a a INNER JOIN mv_base_b b USING(i) GROUP BY i,j,k HAVING SUM(i) > 5;
+
+-- contain view or materialized view
+CREATE VIEW b_view AS SELECT i,k FROM mv_base_b;
+CREATE MATERIALIZED VIEW b_mview AS SELECT i,k FROM mv_base_b;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm07 AS SELECT a.i,a.j FROM mv_base_a a,b_view b WHERE a.i = b.i;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm08 AS SELECT a.i,a.j FROM mv_base_a a,b_mview b WHERE a.i = b.i;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm09 AS SELECT a.i,a.j FROM mv_base_a a, (SELECT i, COUNT(*) FROM mv_base_b GROUP BY i) b WHERE a.i = b.i;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm10 AS SELECT a.i,a.j FROM mv_base_a a WHERE EXISTS(SELECT 1 FROM mv_base_b b WHERE a.i = b.i) OR a.i > 5;
+
+-- contain mutable functions
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm12 AS SELECT i,j FROM mv_base_a WHERE i = random()::int;
+
+-- LIMIT/OFFSET is not supported
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm13 AS SELECT i,j FROM mv_base_a LIMIT 10 OFFSET 5;
+
+-- DISTINCT ON is not supported
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm14 AS SELECT DISTINCT ON(i) i, j FROM mv_base_a;
+
+-- TABLESAMPLE clause is not supported
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm15 AS SELECT i, j FROM mv_base_a TABLESAMPLE SYSTEM(50);
+
+-- window functions are not supported
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm16 AS SELECT *, cume_dist() OVER (ORDER BY i) AS rank FROM mv_base_a;
+
+-- aggregate function with some options is not supported
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm17 AS SELECT COUNT(*) FILTER(WHERE i < 3) FROM mv_base_a;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm18 AS SELECT COUNT(DISTINCT i) FROM mv_base_a;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm19 AS SELECT array_agg(j ORDER BY i DESC) FROM mv_base_a;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm20 AS SELECT i,SUM(j) FROM mv_base_a GROUP BY GROUPING SETS((i),());
+
+-- inheritance parent is not supported
+BEGIN;
+CREATE TABLE parent (i int, v int);
+CREATE TABLE child_a(options text) INHERITS(parent);
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm21 AS SELECT * FROM parent;
+ROLLBACK;
+
+-- UNION statement is not supported
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm22 AS SELECT i,j FROM mv_base_a UNION ALL SELECT i,k FROM mv_base_b;;
+
+-- empty target list is not allowed with IVM
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm25 AS SELECT FROM mv_base_a;
+
+-- FOR UPDATE/SHARE is not supported
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm26 AS SELECT i,j FROM mv_base_a FOR UPDATE;
+
+-- tartget list cannot contain ivm column that start with '__ivm'
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm28 AS SELECT i AS "__ivm_count__" FROM mv_base_a;
+
+-- expressions specified in GROUP BY must appear in the target list.
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm29 AS SELECT COUNT(i) FROM mv_base_a GROUP BY i;
+
+-- experssions containing an aggregate is not supported
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm30 AS SELECT sum(i)*0.5 FROM mv_base_a;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm31 AS SELECT sum(i)/sum(j) FROM mv_base_a;
+
+-- VALUES is not supported
+CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm_only_values1 AS values(1);
+
+-- views containing base tables with Row Level Security
+DROP USER IF EXISTS ivm_admin;
+DROP USER IF EXISTS ivm_user;
+CREATE USER ivm_admin;
+CREATE USER ivm_user;
+
+--- create a table with RLS
+SET SESSION AUTHORIZATION ivm_admin;
+CREATE TABLE rls_tbl(id int, data text, owner name);
+INSERT INTO rls_tbl VALUES
+ (1,'foo','ivm_user'),
+ (2,'bar','postgres');
+CREATE TABLE num_tbl(id int, num text);
+INSERT INTO num_tbl VALUES
+ (1,'one'),
+ (2,'two'),
+ (3,'three'),
+ (4,'four'),
+ (5,'five'),
+ (6,'six');
+
+--- Users can access only their own rows
+CREATE POLICY rls_tbl_policy ON rls_tbl FOR SELECT TO PUBLIC USING(owner = current_user);
+ALTER TABLE rls_tbl ENABLE ROW LEVEL SECURITY;
+GRANT ALL on rls_tbl TO PUBLIC;
+GRANT ALL on num_tbl TO PUBLIC;
+
+--- create a view owned by ivm_user
+SET SESSION AUTHORIZATION ivm_user;
+
+CREATE INCREMENTAL MATERIALIZED VIEW ivm_rls AS SELECT * FROM rls_tbl;
+SELECT id, data, owner FROM ivm_rls ORDER BY 1,2,3;
+RESET SESSION AUTHORIZATION;
+
+--- inserts rows owned by different users
+INSERT INTO rls_tbl VALUES
+ (3,'baz','ivm_user'),
+ (4,'qux','postgres');
+SELECT id, data, owner FROM ivm_rls ORDER BY 1,2,3;
+
+--- combination of diffent kinds of commands
+WITH
+ i AS (INSERT INTO rls_tbl VALUES(5,'quux','postgres'), (6,'corge','ivm_user')),
+ u AS (UPDATE rls_tbl SET owner = 'postgres' WHERE id = 1),
+ u2 AS (UPDATE rls_tbl SET owner = 'ivm_user' WHERE id = 2)
+SELECT;
+SELECT id, data, owner FROM ivm_rls ORDER BY 1,2,3;
+
+---
+SET SESSION AUTHORIZATION ivm_user;
+CREATE INCREMENTAL MATERIALIZED VIEW ivm_rls2 AS SELECT * FROM rls_tbl JOIN num_tbl USING(id);
+RESET SESSION AUTHORIZATION;
+
+WITH
+ x AS (UPDATE rls_tbl SET data = data || '_2' where id in (3,4)),
+ y AS (UPDATE num_tbl SET num = num || '_2' where id in (3,4))
+SELECT;
+SELECT * FROM ivm_rls2 ORDER BY 1,2,3;
+
+-- automatic index creation
+CREATE TABLE base_a (i int primary key, j int);
+CREATE TABLE base_b (i int primary key, j int);
+
+--- group by: create an index
+CREATE INCREMENTAL MATERIALIZED VIEW mv_idx1 AS SELECT i, sum(j) FROM base_a GROUP BY i;
+
+--- distinct: create an index
+CREATE INCREMENTAL MATERIALIZED VIEW mv_idx2 AS SELECT DISTINCT j FROM base_a;
+
+--- with all pkey columns: create an index
+CREATE INCREMENTAL MATERIALIZED VIEW mv_idx3(i_a, i_b) AS SELECT a.i, b.i FROM base_a a, base_b b;
+
+--- missing some pkey columns: no index
+CREATE INCREMENTAL MATERIALIZED VIEW mv_idx4 AS SELECT j FROM base_a;
+CREATE INCREMENTAL MATERIALIZED VIEW mv_idx5 AS SELECT a.i, b.j FROM base_a a, base_b b;
+
+-- cleanup
+
+DROP TABLE rls_tbl CASCADE;
+DROP TABLE num_tbl CASCADE;
+DROP USER ivm_user;
+DROP USER ivm_admin;
+
+DROP TABLE mv_base_b CASCADE;
+DROP TABLE mv_base_a CASCADE;
--
2.25.1
--Multipart=_Mon__28_Aug_2023_11_52_52_+0900_hj6L5h176QaSGtg7
Content-Type: text/x-diff;
name="v29-0011-Add-documentations-about-Incremental-View-Mainte.patch"
Content-Disposition: attachment;
filename="v29-0011-Add-documentations-about-Incremental-View-Mainte.patch"
Content-Transfer-Encoding: 7bit
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v2] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index fb6a4914b06..1da073f0aaa 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index d3febfe58f1..075194a8af8 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1844,6 +1844,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index 6d304f32fb0..d346aefe9b6 100644
--- a/meson.build
+++ b/meson.build
@@ -2694,6 +2694,7 @@ decl_checks = [
['strlcpy', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index 563f20374ff..b41a5f9dcca 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#if HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#if HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#if HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 339268dc8ef..e7b8e023829 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -80,6 +80,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of `memset_s', and to 0 if you
don't. */
#undef HAVE_DECL_MEMSET_S
base-commit: 6831cd9e3b082d7b830c3196742dd49e3540c49b
--
2.49.0
--7u2oub7w3nl66bkx--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
* [PATCH v1] Use open file description locks for data directory lockfile
@ 2025-12-18 17:21 Dmitrii Dolgov <[email protected]>
0 siblings, 0 replies; 137+ messages in thread
From: Dmitrii Dolgov @ 2025-12-18 17:21 UTC (permalink / raw)
When starting up, postmaster checks for an existing data directory lockfile. If
this file contains current process PID, it's assumed to be stale. Turns out
there is another possibility: we might be running in a PID namespace, and there
is another postgres running inside another PID namespace using the same data
directory. The result is that we don't see another process due to namespace
isolation and start concurrently with the other.
To prevent such situations, at startup use fcntl to get an exclusive open file
description lock for data directory lockfile. Since such locks are associated
with open file descriptors, meaning they're not affected by PID namespace
isolation. It's a "best effort" locking, intended to work with already existing
mechanism, not replace it.
This approach was discussed multiple times in the past, and usually was
rejected as the main work horse for the data directory lockfile due to:
* Portability issues. Open file description lock was a non-POSIX extension in
Linux and similar flock is from BSD standard. But looks like everybody agrees
that such locks make more sense than a typical advisory locks, and
F_OFD_SETLK made its way into POSIX.1 2024 [1].
* Issues with NFS. The current state of things here looks like this:
- NFSv3 doesn't implement open file description locks, they're converted to
advisory locks instead. Advisory locks are subject to namespace isolation,
meaning that processes in different PID namespaces will not see each other
advisory lock, and it's still possible to run multiple postgres
instances on the same data directory.
- NFSv4 uses a lease system for locking, I haven't found any mention of
conversion to advisory locks neither in the man page nor in RFC [2].
To summarize, the approach is now considered POSIX and should fix the described
problem everywhere, except NFSv3.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/functions/fcntl.html
[2]: https://www.rfc-editor.org/rfc/rfc7530
---
configure | 14 ++++
configure.ac | 3 +
meson.build | 1 +
src/backend/utils/init/miscinit.c | 107 +++++++++++++++++++++++++-----
src/include/pg_config.h.in | 4 ++
5 files changed, 111 insertions(+), 18 deletions(-)
diff --git a/configure b/configure
index 14ad0a5006f..b176ac39799 100755
--- a/configure
+++ b/configure
@@ -16177,6 +16177,20 @@ cat >>confdefs.h <<_ACEOF
_ACEOF
+# Linux open file descriptor locks
+ac_fn_c_check_decl "$LINENO" "F_OFD_SETLK" "ac_cv_have_decl_F_OFD_SETLK" "#include <fcntl.h>
+"
+if test "x$ac_cv_have_decl_F_OFD_SETLK" = xyes; then :
+ ac_have_decl=1
+else
+ ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_F_OFD_SETLK $ac_have_decl
+_ACEOF
+
+
ac_fn_c_check_func "$LINENO" "explicit_bzero" "ac_cv_func_explicit_bzero"
if test "x$ac_cv_func_explicit_bzero" = xyes; then :
$as_echo "#define HAVE_EXPLICIT_BZERO 1" >>confdefs.h
diff --git a/configure.ac b/configure.ac
index 01b3bbc1be8..d6cf1f27771 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1838,6 +1838,9 @@ AC_CHECK_DECLS([memset_s], [], [], [#define __STDC_WANT_LIB_EXT1__ 1
# This is probably only present on macOS, but may as well check always
AC_CHECK_DECLS(F_FULLFSYNC, [], [], [#include <fcntl.h>])
+# Linux open file descriptor locks
+AC_CHECK_DECLS([F_OFD_SETLK], [], [], [#include <fcntl.h>])
+
AC_REPLACE_FUNCS(m4_normalize([
explicit_bzero
getopt
diff --git a/meson.build b/meson.build
index d7c5193d4ce..ad9ecad829f 100644
--- a/meson.build
+++ b/meson.build
@@ -2667,6 +2667,7 @@ decl_checks = [
['strnlen', 'string.h'],
['strsep', 'string.h'],
['timingsafe_bcmp', 'string.h'],
+ ['F_OFD_SETLK', 'fcntl.h'],
]
# Need to check for function declarations for these functions, because
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
index fec79992c8d..78bb7df543e 100644
--- a/src/backend/utils/init/miscinit.c
+++ b/src/backend/utils/init/miscinit.c
@@ -68,6 +68,11 @@ static List *lock_files = NIL;
static Latch LocalLatchData;
+#ifdef HAVE_DECL_F_OFD_SETLK
+/* File descriptor for data directory lock file. */
+static int DataDirLockFD;
+#endif
+
/* ----------------------------------------------------------------
* ignoring system indexes support stuff
*
@@ -1117,6 +1122,45 @@ RestoreClientConnectionInfo(char *conninfo)
*-------------------------------------------------------------------------
*/
+/*
+ * Flock the data directory lockfile.
+ *
+ * Lock the data directory lockfile with an open file description lock. If the
+ * lock is already taken, it's a hard stop. It's only a best effort test, and
+ * any other errors are ignored. On succes the file descriptor is duplicated,
+ * to make sure there will be at least one open copy of it to keep the lock.
+ *
+ * filename is used only for reporting purposes.
+ */
+static void
+FlockDataDirLockFile(int fd, const char *filename)
+{
+
+#ifdef HAVE_DECL_F_OFD_SETLK
+ struct flock lock;
+
+ lock.l_type = F_WRLCK;
+ lock.l_whence = SEEK_SET;
+ lock.l_start = 0;
+ lock.l_len = 0;
+ lock.l_pid = 0;
+
+ if (fcntl(fd, F_OFD_SETLK, &lock) == -1)
+ {
+ if (errno == EAGAIN)
+ ereport(FATAL,
+ (errcode(ERRCODE_LOCK_FILE_EXISTS),
+ errmsg("cannot lock the lock file \"%s\"", filename),
+ errhint("Another server is starting.")));
+ else
+ elog(WARNING, "Failed locking file \"%s\", %m", filename);
+ }
+ else
+ DataDirLockFD = dup(fd);
+#endif
+
+}
+
/*
* proc_exit callback to remove lockfiles.
*/
@@ -1125,6 +1169,11 @@ UnlinkLockFiles(int status, Datum arg)
{
ListCell *l;
+#ifdef HAVE_DECL_F_OFD_SETLK
+ /* Close the file descriptor, which keeps the open file description lock */
+ close(DataDirLockFD);
+#endif
+
foreach(l, lock_files)
{
char *curfile = (char *) lfirst(l);
@@ -1171,22 +1220,32 @@ CreateLockFile(const char *filename, bool amPostmaster,
const char *envvar;
/*
- * If the PID in the lockfile is our own PID or our parent's or
- * grandparent's PID, then the file must be stale (probably left over from
- * a previous system boot cycle). We need to check this because of the
- * likelihood that a reboot will assign exactly the same PID as we had in
- * the previous reboot, or one that's only one or two counts larger and
- * hence the lockfile's PID now refers to an ancestor shell process. We
- * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
- * via the environment variable PG_GRANDPARENT_PID; this is so that
- * launching the postmaster via pg_ctl can be just as reliable as
- * launching it directly. There is no provision for detecting
- * further-removed ancestor processes, but if the init script is written
- * carefully then all but the immediate parent shell will be root-owned
- * processes and so the kill test will fail with EPERM. Note that we
- * cannot get a false negative this way, because an existing postmaster
- * would surely never launch a competing postmaster or pg_ctl process
- * directly.
+ * If we find an already existing lockfile containing our own PID,
+ * there are few options:
+ *
+ * - There is another process, that we don't see due to PID namespace
+ * isolation, which is already running in this data directory.
+ *
+ * To prevent two concurrent processes working with the same data
+ * directory, we first try to lock the lockfile exclusively.
+ *
+ * - The file must be stale, probably left over from a previous system boot
+ * cycle. The same if the lockfile contains our parent's or grandparent's
+ * PID.
+ *
+ * We need to check this because of the likelihood that a reboot will
+ * assign exactly the same PID as we had in the previous reboot, or one
+ * that's only one or two counts larger and hence the lockfile's PID now
+ * refers to an ancestor shell process. We allow pg_ctl to pass down its
+ * parent shell PID (our grandparent PID) via the environment variable
+ * PG_GRANDPARENT_PID; this is so that launching the postmaster via
+ * pg_ctl can be just as reliable as launching it directly. There is no
+ * provision for detecting further-removed ancestor processes, but if the
+ * init script is written carefully then all but the immediate parent
+ * shell will be root-owned processes and so the kill test will fail with
+ * EPERM. Note that we cannot get a false negative this way, because an
+ * existing postmaster would surely never launch a competing postmaster
+ * or pg_ctl process directly.
*/
my_pid = getpid();
@@ -1222,7 +1281,11 @@ CreateLockFile(const char *filename, bool amPostmaster,
*/
fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
if (fd >= 0)
- break; /* Success; exit the retry loop */
+ {
+ /* Success; lock and exit the retry loop */
+ FlockDataDirLockFile(fd, filename);
+ break;
+ }
/*
* Couldn't create the pid file. Probably it already exists.
@@ -1236,8 +1299,12 @@ CreateLockFile(const char *filename, bool amPostmaster,
/*
* Read the file to get the old owner's PID. Note race condition
* here: file might have been deleted since we tried to create it.
+ *
+ * We're going to use the same fd for flock, and want to create a write
+ * lock for the latter one. Since both fd and the lock have to be of
+ * the same type, open the file for read and write.
*/
- fd = open(filename, O_RDONLY, pg_file_create_mode);
+ fd = open(filename, O_RDWR, pg_file_create_mode);
if (fd < 0)
{
if (errno == ENOENT)
@@ -1247,6 +1314,10 @@ CreateLockFile(const char *filename, bool amPostmaster,
errmsg("could not open lock file \"%s\": %m",
filename)));
}
+
+ /* Try to lock the file. We stop here, if it's already locked. */
+ FlockDataDirLockFile(fd, filename);
+
pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
ereport(FATAL,
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 92fcc5f3063..c19a50f108e 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -83,6 +83,10 @@
don't. */
#undef HAVE_DECL_F_FULLFSYNC
+/* Define to 1 if you have the declaration of `F_OFD_SETLK', and to 0 if you
+ don't. */
+#undef HAVE_DECL_F_OFD_SETLK
+
/* Define to 1 if you have the declaration of
`LLVMCreateGDBRegistrationListener', and to 0 if you don't. */
#undef HAVE_DECL_LLVMCREATEGDBREGISTRATIONLISTENER
base-commit: 44f49511b7940adf3be4337d4feb2de38fe92297
--
2.49.0
--n6t6hie3zne7u6vd--
^ permalink raw reply [nested|flat] 137+ messages in thread
end of thread, other threads:[~2025-12-18 17:21 UTC | newest]
Thread overview: 137+ messages (download: mbox mbox.gz follow: Atom feed)
-- links below jump to the message on this page --
2021-03-10 02:11 [PATCH v29 10/11] Add regression tests for Incremental View Maintenance Takuma Hoshiai <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v1] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
2025-12-18 17:21 [PATCH v2] Use open file description locks for data directory lockfile Dmitrii Dolgov <[email protected]>
This inbox is served by agora; see mirroring instructions
for how to clone and mirror all data and code used for this inbox