linux-kernel - Re: [PATCH memcg 0/1] false global OOM triggered by memcg-limited task

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <3c76e2d7-e545-ef34-b2c3-a5f63b1eff51@virtuozzo.com>
Date:   Tue, 19 Oct 2021 22:09:19 +0300
From:   Vasily Averin <vvs@...tuozzo.com>
To:     Michal Hocko <mhocko@...e.com>
Cc:     Johannes Weiner <hannes@...xchg.org>,
        Vladimir Davydov <vdavydov.dev@...il.com>,
        Andrew Morton <akpm@...ux-foundation.org>,
        Roman Gushchin <guro@...com>,
        Uladzislau Rezki <urezki@...il.com>,
        Vlastimil Babka <vbabka@...e.cz>,
        Shakeel Butt <shakeelb@...gle.com>,
        Mel Gorman <mgorman@...hsingularity.net>,
        cgroups@...r.kernel.org, linux-mm@...ck.org,
        linux-kernel@...r.kernel.org, kernel@...nvz.org
Subject: Re: [PATCH memcg 0/1] false global OOM triggered by memcg-limited
 task

On 19.10.2021 17:13, Michal Hocko wrote:
> On Tue 19-10-21 16:26:50, Vasily Averin wrote:
>> On 19.10.2021 15:04, Michal Hocko wrote:
>>> On Tue 19-10-21 13:54:42, Michal Hocko wrote:
>>>> On Tue 19-10-21 13:30:06, Vasily Averin wrote:
>>>>> On 19.10.2021 11:49, Michal Hocko wrote:
>>>>>> On Tue 19-10-21 09:30:18, Vasily Averin wrote:
>>>>>> [...]
>>>>>>> With my patch ("memcg: prohibit unconditional exceeding the limit of dying tasks") try_charge_memcg() can fail:
>>>>>>> a) due to fatal signal
>>>>>>> b) when mem_cgroup_oom -> mem_cgroup_out_of_memory -> out_of_memory() returns false (when select_bad_process() found nothing)
>>>>>>>
>>>>>>> To handle a) we can follow to your suggestion and skip excution of out_of_memory() in pagefault_out_of memory()
>>>>>>> To handle b) we can go to retry: if mem_cgroup_oom() return OOM_FAILED.
>>>>>
>>>>>> How is b) possible without current being killed? Do we allow remote
>>>>>> charging?
>>>>>
>>>>> out_of_memory for memcg_oom
>>>>>  select_bad_process
>>>>>   mem_cgroup_scan_tasks
>>>>>    oom_evaluate_task
>>>>>     oom_badness
>>>>>
>>>>>         /*
>>>>>          * Do not even consider tasks which are explicitly marked oom
>>>>>          * unkillable or have been already oom reaped or the are in
>>>>>          * the middle of vfork
>>>>>          */
>>>>>         adj = (long)p->signal->oom_score_adj;
>>>>>         if (adj == OOM_SCORE_ADJ_MIN ||
>>>>>                         test_bit(MMF_OOM_SKIP, &p->mm->flags) ||
>>>>>                         in_vfork(p)) {
>>>>>                 task_unlock(p);
>>>>>                 return LONG_MIN;
>>>>>         }
>>>>>
>>>>> This time we handle userspace page fault, so we cannot be kenrel thread,
>>>>> and cannot be in_vfork().
>>>>> However task can be marked as oom unkillable, 
>>>>> i.e. have p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN
>>>>
>>>> You are right. I am not sure there is a way out of this though. The task
>>>> can only retry for ever in this case. There is nothing actionable here.
>>>> We cannot kill the task and there is no other way to release the memory.
>>>
>>> Btw. don't we force the charge in that case?
>>
>> We should force charge for allocation from inside page fault handler,
>> to prevent endless cycle in retried page faults.
>> However we should not do it for allocations from task context,
>> to prevent memcg-limited vmalloc-eaters from to consume all host memory.
> 
> I don't see a big difference between those two. Because the #PF could
> result into the very same situation depleting all the memory by
> overcharging. A different behavior just leads to a confusion and
> unexpected behavior. E.g. in the past we only triggered memcg OOM killer
> from the #PF path and failed the charge otherwise. That is something
> different but it shows problems we haven't anticipated and had user
> visible problems. See 29ef680ae7c2 ("memcg, oom: move out_of_memory back
> to the charge path").

In this case I think we should fail this allocation.
It's better do not allow overcharge, neither in #PF not in regular allocations.

However this failure will trigger false global OOM in pagefault_out_of_memory(),
and we need to find some way to prevent it.

Thank you,
	Vasily Averin