From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1753689AbcFTMoJ (ORCPT ); Mon, 20 Jun 2016 08:44:09 -0400 Received: from mail-wm0-f68.google.com ([74.125.82.68]:34343 "EHLO mail-wm0-f68.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752526AbcFTMn4 (ORCPT ); Mon, 20 Jun 2016 08:43:56 -0400 From: Michal Hocko To: Cc: Tetsuo Handa , David Rientjes , Oleg Nesterov , Vladimir Davydov , Andrew Morton , LKML Subject: [PATCH 0/10 -v5] Handle oom bypass more gracefully Date: Mon, 20 Jun 2016 14:43:38 +0200 Message-Id: <1466426628-15074-1-git-send-email-mhocko@kernel.org> X-Mailer: git-send-email 2.8.1 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Hi, this is the v5 version of the patchse. Previous version was posted http://lkml.kernel.org/r/1465473137-22531-1-git-send-email-mhocko@kernel.org There was one issue fixed in patch 7. Other than that we have discussed nommu situation after patch 7 which might reintroduce a theoretical race condition when the oom victim passes exit_mm->exit_mm_victim after oom_kill_process checks task_will_free_mem and before mark_oom_victim is called which was reviously fixed by 83363b917a2982dd ("oom: make sure that TIF_MEMDIE is set under task_lock"). The issue was never seen in the real life AFAIK and it would only affect nommu. On the other hand the patch fixes a class of potential lockups when we mark oom victim on a thread while other threads will stay alive. So I think it is a reasonable trade off. Tetsuo still thinks that a timeout based solution is better way to address all the issues: " To me, timeout based one is sufficient for handling any traps that hit nommu kernels after the OOM killer is invoked. Anyway, I don't like this series because this series ignores theoretical cases. I can't make progress as long as you repeat "does it really matter/occur". Please go ahead without Reviewed-by: or Acked-by: from me. " I have added Acked-by from Oleg. What remains is to make mm reapable even when it is shared with kthreads. I have already started working on that and hopefully will have something ready soon but that should go independently on this series. After that we will have MMU basically covered and lockup free. I consider this a noticeable improvement especially because changes to achieve that are much less intrusive than I expected. Nommu still suffers from potential lockups theoretically but a complete lack of reports suggests that this is not the case in the real life. If we ever see any reports we should implement the reaper for nommu as well. Shouldn't be too hard... The original description: ------------------------- The following 10 patches should put some order to very rare cases of mm shared between processes and make the paths which bypass the oom killer oom reapable and therefore much more reliable finally. Even though mm shared outside of thread group is rare (either vforked tasks for a short period, use_mm by kernel threads or exotic thread model of clone(CLONE_VM) without CLONE_SIGHAND) it is better to cover them. Not only it makes the current oom killer logic quite hard to follow and reason about it can lead to weird corner cases. E.g. it is possible to select an oom victim which shares the mm with unkillable process or bypass the oom killer even when other processes sharing the mm are still alive and other weird cases. Patch 1 drops bogus task_lock and mm check from oom_{score_}adj_write. This can be considered a bug fix with a low impact as nobody has noticed for years. Patch 2 drops sighand lock because it is not needed anymore as pointed by Oleg. Patch 3 is a clean up of oom_score_adj handling and a preparatory work for later patches. Patch 4 enforces oom_adj_score to be consistent between processes sharing the mm to behave consistently with the regular thread groups. This can be considered a user visible behavior change because one thread group updating oom_score_adj will affect others which share the same mm via clone(CLONE_VM). I argue that this should be acceptable because we already have the same behavior for threads in the same thread group and sharing the mm without signal struct is just a different model of threading. This is probably the most controversial part of the series, I would like to find some consensus here. There were some suggestions to hook some counter/oom_score_adj into the mm_struct but I feel that this is not necessary right now and we can rely on proc handler + oom_kill_process to DTRT. I can be convinced otherwise but I strongly think that whatever we do the userspace has to have a way to see the current oom priority as consistently as possible. Patch 5 makes sure that no vforked task is selected if it is sharing the mm with oom unkillable task. Patch 6 ensures that all user tasks sharing the mm are killed which in turn makes sure that all oom victims are oom reapable. Patch 7 guarantees that task_will_free_mem will always imply reapable bypass of the oom killer. Patch 8 is new in this version and it addresses an issue pointed out by 0-day OOM report where an oom victim was reaped several times. Patch 9 puts an upper bound on how many times oom_reaper tries to reap a task and hides it from the oom killer to move on when no progress can be made. This will give an upper bound to how long an oom_reapable task can block the oom killer from selecting another victim if the oom_reaper is not able to reap the victim. Patch 10 tries to plug the (hopefully) last hole when we can still lock up when the oom victim is shared with oom unkillable tasks (kthreads and global init). We just try to be best effort in that case and rather fallback to kill something else than risk a lockup. The patchset is based on the current mmotm tree (mmotm-2016-06-15-16-18). I have pushed the patchset to my git tree for an easier review git://git.kernel.org/pub/scm/linux/kernel/git/mhocko/mm.git to branch attempts/process-share-mm-oom-sanitization Michal Hocko (10): proc, oom: drop bogus task_lock and mm check proc, oom: drop bogus sighand lock proc, oom_adj: extract oom_score_adj setting into a helper mm, oom_adj: make sure processes sharing mm have same view of oom_score_adj mm, oom: skip vforked tasks from being selected mm, oom: kill all tasks sharing the mm mm, oom: fortify task_will_free_mem mm, oom: task_will_free_mem should skip oom_reaped tasks mm, oom_reaper: do not attempt to reap a task more than twice mm, oom: hide mm which is shared with kthread or global init fs/proc/base.c | 185 +++++++++++++++++++++++----------------------- include/linux/mm.h | 2 + include/linux/oom.h | 26 +------ include/linux/sched.h | 27 +++++++ mm/memcontrol.c | 4 +- mm/oom_kill.c | 198 ++++++++++++++++++++++++++++++++++---------------- 6 files changed, 266 insertions(+), 176 deletions(-)