memcg: fix endless loop caused by mem_cgroup_iter
authorMichal Hocko <mhocko@suse.cz>
Thu, 23 Jan 2014 23:53:35 +0000 (15:53 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 24 Jan 2014 00:36:53 +0000 (16:36 -0800)
Hugh has reported an endless loop when the hardlimit reclaim sees the
same group all the time.  This might happen when the reclaim races with
the memcg removal.

shrink_zone
                                                [rmdir root]
  mem_cgroup_iter(root, NULL, reclaim)
    // prev = NULL
    rcu_read_lock()
    mem_cgroup_iter_load
      last_visited = iter->last_visited   // gets root || NULL
      css_tryget(last_visited)            // failed
      last_visited = NULL                 [1]
    memcg = root = __mem_cgroup_iter_next(root, NULL)
    mem_cgroup_iter_update
      iter->last_visited = root;
    reclaim->generation = iter->generation

 mem_cgroup_iter(root, root, reclaim)
   // prev = root
   rcu_read_lock
    mem_cgroup_iter_load
      last_visited = iter->last_visited   // gets root
      css_tryget(last_visited)            // failed
    [1]

The issue seemed to be introduced by commit 5f5781619718 ("memcg: relax
memcg iter caching") which has replaced unconditional css_get/css_put by
css_tryget/css_put for the cached iterator.

This patch fixes the issue by skipping css_tryget on the root of the
tree walk in mem_cgroup_iter_load and symmetrically doesn't release it
in mem_cgroup_iter_update.

Signed-off-by: Michal Hocko <mhocko@suse.cz>
Reported-by: Hugh Dickins <hughd@google.com>
Tested-by: Hugh Dickins <hughd@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Greg Thelen <gthelen@google.com>
Cc: <stable@vger.kernel.org> [3.10+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/memcontrol.c

index c8336e8f8df064c9a1ecdbb8055a785b3481d4f2..da07784dde875247dd7b2c47cd25eacb42780921 100644 (file)
@@ -1158,7 +1158,15 @@ mem_cgroup_iter_load(struct mem_cgroup_reclaim_iter *iter,
        if (iter->last_dead_count == *sequence) {
                smp_rmb();
                position = iter->last_visited;
-               if (position && !css_tryget(&position->css))
+
+               /*
+                * We cannot take a reference to root because we might race
+                * with root removal and returning NULL would end up in
+                * an endless loop on the iterator user level when root
+                * would be returned all the time.
+                */
+               if (position && position != root &&
+                               !css_tryget(&position->css))
                        position = NULL;
        }
        return position;
@@ -1167,9 +1175,11 @@ mem_cgroup_iter_load(struct mem_cgroup_reclaim_iter *iter,
 static void mem_cgroup_iter_update(struct mem_cgroup_reclaim_iter *iter,
                                   struct mem_cgroup *last_visited,
                                   struct mem_cgroup *new_position,
+                                  struct mem_cgroup *root,
                                   int sequence)
 {
-       if (last_visited)
+       /* root reference counting symmetric to mem_cgroup_iter_load */
+       if (last_visited && last_visited != root)
                css_put(&last_visited->css);
        /*
         * We store the sequence count from the time @last_visited was
@@ -1244,7 +1254,8 @@ struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
                memcg = __mem_cgroup_iter_next(root, last_visited);
 
                if (reclaim) {
-                       mem_cgroup_iter_update(iter, last_visited, memcg, seq);
+                       mem_cgroup_iter_update(iter, last_visited, memcg, root,
+                                       seq);
 
                        if (!memcg)
                                iter->generation++;