pg.ddx.io  pgaudit/pgaudit GitHub issues and pull requests  
help / color / mirror / Atom feed
Re: [pgaudit/pgaudit] PR #322: Fix use-after-free reading object name/type after event trigger.
10+ messages / 2 participants
[nested] [flat]

* Re: [pgaudit/pgaudit] PR #322: Fix use-after-free reading object name/type after event trigger.
@ 2026-06-26 01:17  "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
  8 siblings, 0 replies; 10+ messages in thread

From: dwsteele (@dwsteele) @ 2026-06-26 01:17 UTC (permalink / raw)
  To: pgaudit/pgaudit <noreply+pgaudit-pgaudit@github.com>

@dwsteele assigned @dwsteele.

^ permalink  raw  [nested|flat] 10+ messages in thread

* [pgaudit/pgaudit] PR #322: Fix use-after-free reading object name/type after event trigger.
@ 2026-06-26 01:17  "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
  0 siblings, 0 replies; 10+ messages in thread

From: dwsteele (@dwsteele) @ 2026-06-26 01:17 UTC (permalink / raw)
  To: pgaudit/pgaudit <noreply+pgaudit-pgaudit@github.com>

The ddl_command_end and sql_drop event triggers store each object's name and type in a temporary memory context that is freed before the trigger returns. The audit event was only marked as logged when log_audit_event() actually emitted a record, so when the last command collected by the trigger belonged to a class that is not being logged, for example a GRANT inside CREATE SCHEMA while only DDL is logged, the event was left marked "not logged". The ProcessUtility hook then logged the statement a second time after the trigger returned, reading the object name and type from the already-freed context. Under AddressSanitizer this surfaced as a heap-buffer-overflow in append_valid_csv().

Mark the audit event as logged after each collected row is processed so the statement is not logged again with freed fields.

Reported-by: @nkPit

inline patch 1: expected/pgaudit.out, pgaudit.c, sql/pgaudit.sql (+103/-0 lines) download inline-1.patchdiff --git a/expected/pgaudit.out b/expected/pgaudit.out index 17e8de1..eafe356 100644 --- a/expected/pgaudit.out +++ b/expected/pgaudit.out @@ -2728,6 +2728,56 @@ NOTICE: AUDIT: SESSION,14,1,READ,SELECT,TABLE,public.graph_vertex,"SELECT name DROP PROPERTY GRAPH graph_test; DROP TABLE graph_vertex; +-- +-- Test that a role substatement (GRANT) collected as the last command of a DDL +-- statement does not cause the statement to be logged a second time by the +-- ProcessUtility hook, reading the object name/type after their memory context +-- has been freed, when only DDL is being logged. +SET pgaudit.log = 'none'; +CREATE EXTENSION pgaudit; +SET pgaudit.log_client = on; +SET pgaudit.log_level = 'notice'; +SET pgaudit.log_relation = off; +SET pgaudit.log = 'ddl'; +CREATE TABLE schema_grant_tbl (id int); +NOTICE: AUDIT: SESSION,15,1,DDL,CREATE TABLE,TABLE,public.schema_grant_tbl,CREATE TABLE schema_grant_tbl (id int),<not logged> +CREATE ROLE regress_schema_grant; +CREATE SCHEMA schema_grant + GRANT SELECT + ON public.schema_grant_tbl + TO regress_schema_grant; +NOTICE: AUDIT: SESSION,16,1,DDL,CREATE SCHEMA,SCHEMA,schema_grant,"CREATE SCHEMA schema_grant + GRANT SELECT + ON public.schema_grant_tbl + TO regress_schema_grant",<not logged> +-- The same scenario written with the CREATE SCHEMA ... AUTHORIZATION form. Its +-- schema elements execute as the target role, so create the table as an element +-- (owned by that role) and grant on it; the trailing GRANT substatement +-- exercises the same path. +CREATE ROLE regress_schema_auth; +CREATE SCHEMA AUTHORIZATION regress_schema_auth + CREATE TABLE schema_auth_tbl (id int) + GRANT SELECT + ON schema_auth_tbl + TO PUBLIC; +NOTICE: AUDIT: SESSION,17,1,DDL,CREATE SCHEMA,SCHEMA,regress_schema_auth,"CREATE SCHEMA AUTHORIZATION regress_schema_auth + CREATE TABLE schema_auth_tbl (id int) + GRANT SELECT + ON schema_auth_tbl + TO PUBLIC",<not logged> +NOTICE: AUDIT: SESSION,17,1,DDL,CREATE TABLE,TABLE,regress_schema_auth.schema_auth_tbl,"CREATE SCHEMA AUTHORIZATION regress_schema_auth + CREATE TABLE schema_auth_tbl (id int) + GRANT SELECT + ON schema_auth_tbl + TO PUBLIC",<not logged> +SET pgaudit.log = 'none'; +DROP SCHEMA schema_grant; +DROP SCHEMA regress_schema_auth CASCADE; +NOTICE: drop cascades to table regress_schema_auth.schema_auth_tbl +DROP TABLE schema_grant_tbl; +DROP ROLE regress_schema_grant; +DROP ROLE regress_schema_auth; +DROP EXTENSION pgaudit; -- Cleanup -- Set client_min_messages up to warning to avoid noise SET client_min_messages = 'warning'; diff --git a/pgaudit.c b/pgaudit.c index 0086883..c5cca00 100644 --- a/pgaudit.c +++ b/pgaudit.c @@ -1879,6 +1879,12 @@ pgaudit_ddl_command_end(PG_FUNCTION_ARGS) } else log_audit_event(auditEventStack); + + /* + * Mark the audit event as logged so it won't be logged again with fields + * that have been freed. + */ + auditEventStack->auditEvent.logged = true; } /* Complete the query */ @@ -1962,6 +1968,12 @@ pgaudit_sql_drop(PG_FUNCTION_ARGS) auditEventStack->auditEvent.logged = false; log_audit_event(auditEventStack); + + /* + * Mark the audit event as logged so it won't be logged again with fields + * that have been freed. + */ + auditEventStack->auditEvent.logged = true; } /* Complete the query */ diff --git a/sql/pgaudit.sql b/sql/pgaudit.sql index 9dfc571..a4f2169 100644 --- a/sql/pgaudit.sql +++ b/sql/pgaudit.sql @@ -1750,6 +1750,47 @@ SELECT name DROP PROPERTY GRAPH graph_test; DROP TABLE graph_vertex; +-- +-- Test that a role substatement (GRANT) collected as the last command of a DDL +-- statement does not cause the statement to be logged a second time by the +-- ProcessUtility hook, reading the object name/type after their memory context +-- has been freed, when only DDL is being logged. +SET pgaudit.log = 'none'; +CREATE EXTENSION pgaudit; + +SET pgaudit.log_client = on; +SET pgaudit.log_level = 'notice'; +SET pgaudit.log_relation = off; +SET pgaudit.log = 'ddl'; + +CREATE TABLE schema_grant_tbl (id int); +CREATE ROLE regress_schema_grant; + +CREATE SCHEMA schema_grant + GRANT SELECT + ON public.schema_grant_tbl + TO regress_schema_grant; + +-- The same scenario written with the CREATE SCHEMA ... AUTHORIZATION form. Its +-- schema elements execute as the target role, so create the table as an element +-- (owned by that role) and grant on it; the trailing GRANT substatement +-- exercises the same path. +CREATE ROLE regress_schema_auth; + +CREATE SCHEMA AUTHORIZATION regress_schema_auth + CREATE TABLE schema_auth_tbl (id int) + GRANT SELECT + ON schema_auth_tbl + TO PUBLIC; + +SET pgaudit.log = 'none'; +DROP SCHEMA schema_grant; +DROP SCHEMA regress_schema_auth CASCADE; +DROP TABLE schema_grant_tbl; +DROP ROLE regress_schema_grant; +DROP ROLE regress_schema_auth; +DROP EXTENSION pgaudit; + -- Cleanup -- Set client_min_messages up to warning to avoid noise SET client_min_messages = 'warning';
^ permalink raw [nested|flat] 10+ messages in thread

* Re: [pgaudit/pgaudit] PR #322: Fix use-after-free reading object name/type after event trigger.
@ 2026-06-26 01:18  "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
  8 siblings, 0 replies; 10+ messages in thread

From: dwsteele (@dwsteele) @ 2026-06-26 01:18 UTC (permalink / raw)
  To: pgaudit/pgaudit <noreply+pgaudit-pgaudit@github.com>

@dwsteele applied label "bug".

^ permalink  raw  [nested|flat] 10+ messages in thread

* Re: [pgaudit/pgaudit] PR #322: Fix use-after-free reading object name/type after event trigger.
@ 2026-07-02 19:24  "pgguru (@pgguru)" <noreply+pgguru@github.com>
  8 siblings, 0 replies; 10+ messages in thread

From: pgguru (@pgguru) @ 2026-07-02 19:24 UTC (permalink / raw)
  To: pgaudit/pgaudit <noreply+pgaudit-pgaudit@github.com>

(on expected/pgaudit.out:2745)

does `CREATE SCHEMA ... AUTHORIZATION` hit these same paths?

^ permalink  raw  [nested|flat] 10+ messages in thread

* Re: [pgaudit/pgaudit] PR #322: Fix use-after-free reading object name/type after event trigger.
@ 2026-07-02 19:33  "pgguru (@pgguru)" <noreply+pgguru@github.com>
  8 siblings, 0 replies; 10+ messages in thread

From: pgguru (@pgguru) @ 2026-07-02 19:33 UTC (permalink / raw)
  To: pgaudit/pgaudit <noreply+pgaudit-pgaudit@github.com>

(on pgaudit.c:1887)

So this doesn't change anything about the context that pieces were allocated in, it only suppresses the actual later inspection of the now invalid data?

^ permalink  raw  [nested|flat] 10+ messages in thread

* Re: [pgaudit/pgaudit] PR #322: Fix use-after-free reading object name/type after event trigger.
@ 2026-07-03 03:38  "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
  8 siblings, 0 replies; 10+ messages in thread

From: dwsteele (@dwsteele) @ 2026-07-03 03:38 UTC (permalink / raw)
  To: pgaudit/pgaudit <noreply+pgaudit-pgaudit@github.com>

(on pgaudit.c:1887)

Correct, this just keeps us from touching possible stale references later. We shouldn't be logging again anyway, so this is the correct behavior.

^ permalink  raw  [nested|flat] 10+ messages in thread

* Re: [pgaudit/pgaudit] PR #322: Fix use-after-free reading object name/type after event trigger.
@ 2026-07-03 04:07  "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
  8 siblings, 0 replies; 10+ messages in thread

From: dwsteele (@dwsteele) @ 2026-07-03 04:07 UTC (permalink / raw)
  To: pgaudit/pgaudit <noreply+pgaudit-pgaudit@github.com>

(on expected/pgaudit.out:2745)

Yes and I added a test for that in bdb1591cc01a.

^ permalink  raw  [nested|flat] 10+ messages in thread

* Re: [pgaudit/pgaudit] PR #322 (approved): Fix use-after-free reading object name/type after event trigger.
@ 2026-07-20 14:40  "pgguru (@pgguru)" <noreply+pgguru@github.com>
  8 siblings, 0 replies; 10+ messages in thread

From: pgguru (@pgguru) @ 2026-07-20 14:40 UTC (permalink / raw)
  To: pgaudit/pgaudit <noreply+pgaudit-pgaudit@github.com>



^ permalink  raw  [nested|flat] 10+ messages in thread

* Re: [pgaudit/pgaudit] PR #322: Fix use-after-free reading object name/type after event trigger.
@ 2026-07-21 03:22  "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
  8 siblings, 0 replies; 10+ messages in thread

From: dwsteele (@dwsteele) @ 2026-07-21 03:22 UTC (permalink / raw)
  To: pgaudit/pgaudit <noreply+pgaudit-pgaudit@github.com>

@dwsteele closed this.

^ permalink  raw  [nested|flat] 10+ messages in thread

* Re: [pgaudit/pgaudit] PR #322: Fix use-after-free reading object name/type after event trigger.
@ 2026-07-21 03:22  "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
  8 siblings, 0 replies; 10+ messages in thread

From: dwsteele (@dwsteele) @ 2026-07-21 03:22 UTC (permalink / raw)
  To: pgaudit/pgaudit <noreply+pgaudit-pgaudit@github.com>

@dwsteele merged this (commit 9352026ab84e).

^ permalink  raw  [nested|flat] 10+ messages in thread


end of thread, other threads:[~2026-07-21 03:22 UTC | newest]

Thread overview: 10+ messages (download: mbox mbox.gz follow: Atom feed)
-- links below jump to the message on this page --
2026-06-26 01:17 [pgaudit/pgaudit] PR #322: Fix use-after-free reading object name/type after event trigger. "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
2026-06-26 01:17 ` "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
2026-06-26 01:18 ` "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
2026-07-02 19:24 ` "pgguru (@pgguru)" <noreply+pgguru@github.com>
2026-07-02 19:33 ` "pgguru (@pgguru)" <noreply+pgguru@github.com>
2026-07-03 03:38 ` "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
2026-07-03 04:07 ` "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
2026-07-20 14:40 ` Re: [pgaudit/pgaudit] PR #322 (approved): Fix use-after-free reading object name/type after event trigger. "pgguru (@pgguru)" <noreply+pgguru@github.com>
2026-07-21 03:22 ` "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
2026-07-21 03:22 ` "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>

This inbox is served by DDX for PostgreSQL; see mirroring instructions
for how to clone and mirror all data and code used for this inbox