]> git.ipfire.org Git - thirdparty/postgresql.git/commit
Fix race in ReplicationSlotRelease() for ephemeral slots
authorFujii Masao <fujii@postgresql.org>
Wed, 3 Jun 2026 09:47:31 +0000 (18:47 +0900)
committerFujii Masao <fujii@postgresql.org>
Wed, 3 Jun 2026 09:47:46 +0000 (18:47 +0900)
commita4eb59d40f278471d5b83a25cef6e56bc997e696
tree7ad45d97b45058b5029c1016d7b158781db2eb02
parent259b627d5ccb7407881b0f283c0c7fa462a4ce3c
Fix race in ReplicationSlotRelease() for ephemeral slots

When releasing an ephemeral replication slot, ReplicationSlotRelease()
drops the slot via ReplicationSlotDropAcquired().

However, after dropping the slot, ReplicationSlotRelease() continued
to use its local "slot" pointer, which still referenced the dropped
slot's former shared-memory entry. It could then update fields such as
effective_xmin in that entry.

Once an ephemeral slot has been dropped (via ReplicationSlotDropAcquired()),
its slot array entry can be reused immediately by another backend
creating a new slot. As a result, those updates could corrupt
the state of an unrelated replication slot.

Fix by skipping those shared-memory updates for phemeral slots and
performing them only for non-ephemeral slots, whose shared-memory
entries remain valid after release.

Backpatch to all supported versions.

Author: Zhijie Hou <houzj.fnst@fujitsu.com>
Reviewed-by: Masao Fujii <masao.fujii@gmail.com>
Reviewed-by: Srinath Reddy Sadipiralla <srinath2133@gmail.com>
Reviewed-by: Xuneng Zhou <xunengzhou@gmail.com>
Discussion: https://postgr.es/m/TY4PR01MB177184FF9EE916F577E1F554194082@TY4PR01MB17718.jpnprd01.prod.outlook.com
Backpatch-through: 14
src/backend/replication/slot.c