On Thu, Jul 19, 2018 at 3:43 AM Michal Hocko [off-list ref] wrote:

[CC Andrew]

On Thu 19-07-18 18:06:47, Jing Xia wrote:

quoted

It was reported that a kernel crash happened in mem_cgroup_iter(),
which can be triggered if the legacy cgroup-v1 non-hierarchical
mode is used.

Unable to handle kernel paging request at virtual address 6b6b6b6b6b6b8f
......
Call trace:
  mem_cgroup_iter+0x2e0/0x6d4
  shrink_zone+0x8c/0x324
  balance_pgdat+0x450/0x640
  kswapd+0x130/0x4b8
  kthread+0xe8/0xfc
  ret_from_fork+0x10/0x20

  mem_cgroup_iter():
      ......
      if (css_tryget(css))    <-- crash here
          break;
      ......

The crashing reason is that mem_cgroup_iter() uses the memcg object
whose pointer is stored in iter->position, which has been freed before
and filled with POISON_FREE(0x6b).

And the root cause of the use-after-free issue is that
invalidate_reclaim_iterators() fails to reset the value of
iter->position to NULL when the css of the memcg is released in non-
hierarchical mode.

Well, spotted!

I suspect
Fixes: 6df38689e0e9 ("mm: memcontrol: fix possible memcg leak due to interrupted reclaim")

but maybe it goes further into past. I also suggest
Cc: stable

even though the non-hierarchical mode is strongly discouraged.

Why not set root_mem_cgroup's use_hierarchy to true by default on
init? If someone wants non-hierarchical mode, they can explicitly set
it to false.

A lack of
reports for 3 years is encouraging that not many people really use this
mode.

quoted

Signed-off-by: Jing Xia <redacted>

Acked-by: Michal Hocko <mhocko@suse.com>

Thanks!

quoted

---
 mm/memcontrol.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index e6f0d5e..8c0280b 100644
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c

@@ -850,7 +850,7 @@ static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg)
      int nid;
      int i;

-     while ((memcg = parent_mem_cgroup(memcg))) {
+     for (; memcg; memcg = parent_mem_cgroup(memcg)) {
              for_each_node(nid) {
                      mz = mem_cgroup_nodeinfo(memcg, nid);
                      for (i = 0; i <= DEF_PRIORITY; i++) {
--

1.9.1

--
Michal Hocko
SUSE Labs
--
To unsubscribe from this list: send the line "unsubscribe cgroups" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html