memcg, kmem: fix reference count handling on the error path
authorMichal Hocko <mhocko@suse.cz>
Mon, 8 Jul 2013 23:00:29 +0000 (16:00 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Tue, 9 Jul 2013 17:33:24 +0000 (10:33 -0700)
mem_cgroup_css_online calls mem_cgroup_put if memcg_init_kmem fails.
This is not correct because only memcg_propagate_kmem takes an
additional reference while mem_cgroup_sockets_init is allowed to fail as
well (although no current implementation fails) but it doesn't take any
reference.  This all suggests that it should be memcg_propagate_kmem
that should clean up after itself so this patch moves mem_cgroup_put
over there.

Unfortunately this is not that easy (as pointed out by Li Zefan) because
memcg_kmem_mark_dead marks the group dead (KMEM_ACCOUNTED_DEAD) if it is
marked active (KMEM_ACCOUNTED_ACTIVE) which is the case even if
memcg_propagate_kmem fails so the additional reference is dropped in
that case in kmem_cgroup_destroy which means that the reference would be
dropped two times.

The easiest way then would be to simply remove mem_cgrroup_put from
mem_cgroup_css_online and rely on kmem_cgroup_destroy doing the right
thing.

Signed-off-by: Michal Hocko <mhocko@suse.cz>
Signed-off-by: Li Zefan <lizefan@huawei.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Glauber Costa <glommer@openvz.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: <stable@vger.kernel.org> [3.8]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/memcontrol.c

index 6b73d8657d64ae45bfd0085e6d4111fa417f3337..bdeb82ca6c20fe0cfd0a725066f2812f1c3beb90 100644 (file)
@@ -6345,14 +6345,6 @@ mem_cgroup_css_online(struct cgroup *cont)
 
        error = memcg_init_kmem(memcg, &mem_cgroup_subsys);
        mutex_unlock(&memcg_create_mutex);
-       if (error) {
-               /*
-                * We call put now because our (and parent's) refcnts
-                * are already in place. mem_cgroup_put() will internally
-                * call __mem_cgroup_free, so return directly
-                */
-               mem_cgroup_put(memcg);
-       }
        return error;
 }