linux-kernel - Re: [PATCH v2] mm, oom: Tolerate processes sharing mm with different view of oom_score

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: <20190201091433.GH11599@dhcp22.suse.cz>
Date:   Fri, 1 Feb 2019 10:14:33 +0100
From:   Michal Hocko <mhocko@...nel.org>
To:     Tetsuo Handa <penguin-kernel@...ove.sakura.ne.jp>
Cc:     Andrew Morton <akpm@...ux-foundation.org>,
        Johannes Weiner <hannes@...xchg.org>,
        David Rientjes <rientjes@...gle.com>, linux-mm@...ck.org,
        Yong-Taek Lee <ytk.lee@...sung.com>,
        Paul McKenney <paulmck@...ux.vnet.ibm.com>,
        Linus Torvalds <torvalds@...ux-foundation.org>,
        LKML <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH v2] mm, oom: Tolerate processes sharing mm with different
 view of oom_score_adj.

On Fri 01-02-19 05:59:55, Tetsuo Handa wrote:
> On 2019/01/31 16:11, Michal Hocko wrote:
> > On Thu 31-01-19 07:49:35, Tetsuo Handa wrote:
> >> This patch reverts both commit 44a70adec910d692 ("mm, oom_adj: make sure
> >> processes sharing mm have same view of oom_score_adj") and commit
> >> 97fd49c2355ffded ("mm, oom: kill all tasks sharing the mm") in order to
> >> close a race and reduce the latency at __set_oom_adj(), and reduces the
> >> warning at __oom_kill_process() in order to minimize the latency.
> >>
> >> Commit 36324a990cf578b5 ("oom: clear TIF_MEMDIE after oom_reaper managed
> >> to unmap the address space") introduced the worst case mentioned in
> >> 44a70adec910d692. But since the OOM killer skips mm with MMF_OOM_SKIP set,
> >> only administrators can trigger the worst case.
> >>
> >> Since 44a70adec910d692 did not take latency into account, we can "hold RCU
> >> for minutes and trigger RCU stall warnings" by calling printk() on many
> >> thousands of thread groups. Also, current code becomes a DoS attack vector
> >> which will allow "stalling for more than one month in unkillable state"
> >> simply printk()ing same messages when many thousands of thread groups
> >> tried to iterate __set_oom_adj() on each other.
> >>
> >> I also noticed that 44a70adec910d692 is racy [1], and trying to fix the
> >> race will require a global lock which is too costly for rare events. And
> >> Michal Hocko is thinking to change the oom_score_adj implementation to per
> >> mm_struct (with shadowed score stored in per task_struct in order to
> >> support vfork() => __set_oom_adj() => execve() sequence) so that we don't
> >> need the global lock.
> >>
> >> If the worst case in 44a70adec910d692 happened, it is an administrator's
> >> request. Therefore, before changing the oom_score_adj implementation,
> >> let's eliminate the DoS attack vector first.
> > 
> > This is really ridiculous. I have already nacked the previous version
> > and provided two ways around. The simplest one is to drop the printk.
> > The second one is to move oom_score_adj to the mm struct. Could you
> > explain why do you still push for this?
> 
> Dropping printk() does not close the race.

But it does remove the source of a long operation from the RCU context.
If you are not willing to post such a trivial patch I will do so.

> You must propose an alternative patch if you dislike this patch.

I will eventually get there.
-- 
Michal Hocko
SUSE Labs