ipc: create_ipc_ns: drop mqueue mount on sysctl setup failure

If setup_mq_sysctls(ns) fails after mq_init_ns(ns) succeeds, the error
path skipped releasing the internal kernel mqueue mount kept in
ns->mq_mnt. That leaves the vfsmount/superblock referenced until final
namespace teardown, i.e. a resource leak on this rare failure edge.

Unwind it by calling mntput(ns->mq_mnt) before dropping user_ns and
freeing the IPC namespace. This mirrors the normal ordering used in
free_ipc_ns().

Link: https://lkml.kernel.org/r/20251021181341.670297-1-vlad_kulikov_c@pm.me
Signed-off-by: Vlad Kulikov <vlad_kulikov_c@pm.me>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Aleksa Sarai <cyphar@cyphar.com>
Cc: Christian Brauner <brauner@kernel.org>
Cc: David Hildenbrand <david@redhat.com>
Cc: Ma Wupeng <mawupeng1@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

diff --git a/ipc/namespace.c b/ipc/namespace.c
index 59b12fcb40bdf4db9c94cbce02fbab6d783a0419..cf62d11a09b900dca6bff19acf4200ef52d428b3 100644 (file)
--- a/ipc/namespace.c
+++ b/ipc/namespace.c
@@ -75,10 +75,10 @@ static struct ipc_namespace *create_ipc_ns(struct user_namespace *user_ns,
 
        err = -ENOMEM;
        if (!setup_mq_sysctls(ns))
-               goto fail_put;
+               goto fail_mq_mount;
 
        if (!setup_ipc_sysctls(ns))
-               goto fail_mq;
+               goto fail_mq_sysctls;
 
        err = msg_init_ns(ns);
        if (err)
@@ -92,9 +92,10 @@ static struct ipc_namespace *create_ipc_ns(struct user_namespace *user_ns,
 
 fail_ipc:
        retire_ipc_sysctls(ns);
-fail_mq:
+fail_mq_sysctls:
        retire_mq_sysctls(ns);
-
+fail_mq_mount:
+       mntput(ns->mq_mnt);
 fail_put:
        put_user_ns(ns->user_ns);
        ns_common_free(ns);