linux-kernel - Re: [PATCH 1/1] mm, oom_adj: don't loop through tasks in __set_oom

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: <20200820132631.GK5033@dhcp22.suse.cz>
Date:   Thu, 20 Aug 2020 15:26:31 +0200
From:   Michal Hocko <mhocko@...e.com>
To:     "Eric W. Biederman" <ebiederm@...ssion.com>
Cc:     Suren Baghdasaryan <surenb@...gle.com>,
        christian.brauner@...ntu.com, mingo@...nel.org,
        peterz@...radead.org, tglx@...utronix.de, esyr@...hat.com,
        christian@...lner.me, areber@...hat.com, shakeelb@...gle.com,
        cyphar@...har.com, oleg@...hat.com, adobriyan@...il.com,
        akpm@...ux-foundation.org, gladkov.alexey@...il.com,
        walken@...gle.com, daniel.m.jordan@...cle.com, avagin@...il.com,
        bernd.edlinger@...mail.de, john.johansen@...onical.com,
        laoar.shao@...il.com, timmurray@...gle.com, minchan@...nel.org,
        kernel-team@...roid.com, linux-kernel@...r.kernel.org,
        linux-fsdevel@...r.kernel.org, linux-mm@...ck.org
Subject: Re: [PATCH 1/1] mm, oom_adj: don't loop through tasks in
 __set_oom_adj when not necessary

On Thu 20-08-20 07:54:44, Eric W. Biederman wrote:
> ebiederm@...ssion.com (Eric W. Biederman) writes:
> 
> 2> Michal Hocko <mhocko@...e.com> writes:
> >
> >> On Thu 20-08-20 07:34:41, Eric W. Biederman wrote:
> >>> Suren Baghdasaryan <surenb@...gle.com> writes:
> >>> 
> >>> > Currently __set_oom_adj loops through all processes in the system to
> >>> > keep oom_score_adj and oom_score_adj_min in sync between processes
> >>> > sharing their mm. This is done for any task with more that one mm_users,
> >>> > which includes processes with multiple threads (sharing mm and signals).
> >>> > However for such processes the loop is unnecessary because their signal
> >>> > structure is shared as well.
> >>> > Android updates oom_score_adj whenever a tasks changes its role
> >>> > (background/foreground/...) or binds to/unbinds from a service, making
> >>> > it more/less important. Such operation can happen frequently.
> >>> > We noticed that updates to oom_score_adj became more expensive and after
> >>> > further investigation found out that the patch mentioned in "Fixes"
> >>> > introduced a regression. Using Pixel 4 with a typical Android workload,
> >>> > write time to oom_score_adj increased from ~3.57us to ~362us. Moreover
> >>> > this regression linearly depends on the number of multi-threaded
> >>> > processes running on the system.
> >>> > Mark the mm with a new MMF_PROC_SHARED flag bit when task is created with
> >>> > CLONE_VM and !CLONE_SIGHAND. Change __set_oom_adj to use MMF_PROC_SHARED
> >>> > instead of mm_users to decide whether oom_score_adj update should be
> >>> > synchronized between multiple processes. To prevent races between clone()
> >>> > and __set_oom_adj(), when oom_score_adj of the process being cloned might
> >>> > be modified from userspace, we use oom_adj_mutex. Its scope is changed to
> >>> > global and it is renamed into oom_adj_lock for naming consistency with
> >>> > oom_lock. Since the combination of CLONE_VM and !CLONE_SIGHAND is rarely
> >>> > used the additional mutex lock in that path of the clone() syscall should
> >>> > not affect its overall performance. Clearing the MMF_PROC_SHARED flag
> >>> > (when the last process sharing the mm exits) is left out of this patch to
> >>> > keep it simple and because it is believed that this threading model is
> >>> > rare. Should there ever be a need for optimizing that case as well, it
> >>> > can be done by hooking into the exit path, likely following the
> >>> > mm_update_next_owner pattern.
> >>> > With the combination of CLONE_VM and !CLONE_SIGHAND being quite rare, the
> >>> > regression is gone after the change is applied.
> >>> 
> >>> So I am confused.
> >>> 
> >>> Is there any reason why we don't simply move signal->oom_score_adj to
> >>> mm->oom_score_adj and call it a day?
> >>
> >> Yes. Please read through 44a70adec910 ("mm, oom_adj: make sure processes
> >> sharing mm have same view of oom_score_adj")
> >
> > That explains why the scores are synchronized.
> >
> > It doesn't explain why we don't do the much simpler thing and move
> > oom_score_adj from signal_struct to mm_struct. Which is my question.
> >
> > Why not put the score where we need it to ensure that the oom score
> > is always synchronized?  AKA on the mm_struct, not the signal_struct.
> 
> Apologies.  That 44a70adec910 does describe that some people have seen
> vfork users set oom_score.  No details unfortunately.
> 
> I will skip the part where posix calls this undefined behavior.  It
> breaks userspace to change.
> 
> It still seems like the code should be able to buffer oom_adj during
> vfork, and only move the value onto mm_struct during exec.

If you can handle vfork by other means then I am all for it. There were
no patches in that regard proposed yet. Maybe it will turn out simpler
then the heavy lifting we have to do in the oom specific code.
-- 
Michal Hocko
SUSE Labs