Re: WARNING in try_charge

From: Michal Hocko <mhocko@kernel.org>
To: syzbot <syzbot+bab151e82a4e973fa325@syzkaller.appspotmail.com>
Cc: cgroups@vger.kernel.org, dvyukov@google.com, hannes@cmpxchg.org,
	linux-kernel@vger.kernel.org, linux-mm@kvack.org,
	penguin-kernel@I-love.SAKURA.ne.jp,
	syzkaller-bugs@googlegroups.com, vdavydov.dev@gmail.com
Subject: Re: WARNING in try_charge
Date: Mon, 6 Aug 2018 20:55:54 +0200	[thread overview]
Message-ID: <20180806185554.GG10003@dhcp22.suse.cz> (raw)
In-Reply-To: <0000000000002ec4580572c85e46@google.com>

The debugging patch was wrong but I guess I see it finally.
It's a race

: [   72.901666] Memory cgroup out of memory: Kill process 6584 (syz-executor1) score 550000 or sacrifice child
: [   72.917037] Killed process 6584 (syz-executor1) total-vm:37704kB, anon-rss:2140kB, file-rss:0kB, shmem-rss:0kB
: [   72.927256] task=syz-executor5 pid=6581 charge bypass
: [   72.928046] oom_reaper: reaped process 6584 (syz-executor1), now anon-rss:0kB, file-rss:0kB, shmem-rss:0kB
: [   72.932818] task=syz-executor6 pid=6576 invoked memcg oom killer. oom_victim=1
: [   72.942790] task=syz-executor5 pid=6581 charge for nr_pages=1
: [   72.949769] syz-executor6 invoked oom-killer: gfp_mask=0x6040c0(GFP_KERNEL|__GFP_COMP), nodemask=(null), order=0, oom_score_adj=0
: [   72.955606] task=syz-executor5 pid=6581 charge bypass
: [   72.967394] syz-executor6 cpuset=/ mems_allowed=0
: [   72.973175] task=syz-executor5 pid=6581 charge for nr_pages=1
: [...]
: [   73.534865] Task in /ile0 killed as a result of limit of /ile0
: [   73.540865] memory: usage 76kB, limit 0kB, failcnt 260
: [   73.546142] memory+swap: usage 0kB, limit 9007199254740988kB, failcnt 0
: [   73.552898] kmem: usage 0kB, limit 9007199254740988kB, failcnt 0
: [   73.559051] Memory cgroup stats for /ile0: cache:0KB rss:0KB rss_huge:0KB shmem:0KB mapped_file:0KB dirty:0KB writeback:0KB swap:0KB inactive_anon:0KB active_anon:0KB inactive_file:0KB active_file:0KB unevictable:0KB
: [   73.578533] Tasks state (memory values in pages):
: [   73.583404] [  pid  ]   uid  tgid total_vm      rss pgtables_bytes swapents oom_score_adj name
: [   73.592277] [   6569]     0  6562     9427        1    53248        0             0 syz-executor0
: [   73.601299] [   6576]     0  6576     9426        0    61440        0             0 syz-executor6
: [   73.610333] [   6578]     0  6578     9426      534    61440        0             0 syz-executor4
: [   73.619381] [   6579]     0  6579     9426        0    57344        0             0 syz-executor5
: [   73.628414] [   6582]     0  6582     9426        0    61440        0             0 syz-executor7
: [   73.637441] [   6584]     0  6584     9426        0    57344        0             0 syz-executor1
: [   73.646464] Memory cgroup out of memory: Kill process 6578 (syz-executor4) score 549000 or sacrifice child
: [   73.656295] task=syz-executor6 pid=6576 is oom victim now

This should be 6578 but we at least know that we are running in 6576
context so the we are setting the state from a remote context which
itself has been killed already

: [   73.661841] Killed process 6578 (syz-executor4) total-vm:37704kB, anon-rss:2136kB, file-rss:0kB, shmem-rss:0kB
: [   73.672035] task=syz-executor6 pid=6576 charge bypass
: [   73.672801] oom_reaper: reaped process 6578 (syz-executor4), now anon-rss:0kB, file-rss:0kB, shmem-rss:0kB
: [   73.678829] task=syz-executor4 pid=6578 invoked memcg oom killer. oom_victim=1

and here the victim finally reached the oom path finally.

: [   73.687453] task=syz-executor6 pid=6576 charge for nr_pages=1
: [   73.694534] ------------[ cut here ]------------
: [   73.700424] task=syz-executor6 pid=6576 charge bypass
: [   73.705175] Memory cgroup charge failed because of no reclaimable memory! This looks like a misconfiguration or a kernel bug.
: [   73.705321] WARNING: CPU: 1 PID: 6578 at mm/memcontrol.c:1707 try_charge+0xafa/0x1710

But there is nobody killable. So the oom kill happened _after_ our force
charge path. Therefore we should do the following regardless whether we
make tis warn or pr_$foo

#syz test: git://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git 116b181bb646afedd770985de20a68721bdb2648

diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index 4603ad75c9a9..1b6eed1bc404 100644
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c
@@ -1703,7 +1703,8 @@ static enum oom_status mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int
 		return OOM_ASYNC;
 	}
 
-	if (mem_cgroup_out_of_memory(memcg, mask, order))
+	if (mem_cgroup_out_of_memory(memcg, mask, order) ||
+			tsk_is_oom_victim(current))
 		return OOM_SUCCESS;
 
 	WARN(1,"Memory cgroup charge failed because of no reclaimable memory! "
-- 
Michal Hocko
SUSE Labs
