[PATCH] af_unix: Fix wrong socket tested in the symmetric half of unix_state_lock_cmp_fn().

From: lirongqing

Date: Fri Nov 06 2026 - 15:19:32 EST


From: Li RongQing <lirongqing@xxxxxxxxx>

The second half of unix_state_lock_cmp_fn() is meant to mirror the
first one with the roles of a and b swapped, but it switches on b's
state again:

if (b->sk.sk_state == TCP_LISTEN) {
switch (b->sk.sk_state) {
case TCP_CLOSE:
case TCP_ESTABLISHED:
return 1;

The enclosing if already established that b is TCP_LISTEN, so the
switch can only reach the default label and the branch always returns
0, "no order known", instead of the intended 1.

Test a's state, as the first half does. The code is only compiled with
CONFIG_PROVE_LOCKING, so there is no runtime memory-safety impact; the
effect is that lockdep cannot validate this nesting order of
unix_state_lock().

Fixes: 98f706de445b ("af_unix: Define locking order for U_LOCK_SECOND in unix_stream_connect().")
Signed-off-by: Li RongQing <lirongqing@xxxxxxxxx>
---
net/unix/af_unix.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/net/unix/af_unix.c b/net/unix/af_unix.c
index e7a0582..501e693 100644
--- a/net/unix/af_unix.c
+++ b/net/unix/af_unix.c
@@ -163,7 +163,7 @@ static int unix_state_lock_cmp_fn(const struct lockdep_map *_a,

/* Should never happen. Just to be symmetric. */
if (b->sk.sk_state == TCP_LISTEN) {
- switch (b->sk.sk_state) {
+ switch (a->sk.sk_state) {
case TCP_CLOSE:
case TCP_ESTABLISHED:
return 1;
--
2.9.4