linux-kernel - Re: [PATCH] mm: memcontrol: zap task_struct->memcg_oom_{gfp

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: <20160311154704.GW27701@dhcp22.suse.cz>
Date:	Fri, 11 Mar 2016 16:47:04 +0100
From:	Michal Hocko <mhocko@...nel.org>
To:	Vladimir Davydov <vdavydov@...tuozzo.com>
Cc:	Andrew Morton <akpm@...ux-foundation.org>,
	Johannes Weiner <hannes@...xchg.org>, linux-mm@...ck.org,
	linux-kernel@...r.kernel.org
Subject: Re: [PATCH] mm: memcontrol: zap
 task_struct->memcg_oom_{gfp_mask,order}

On Fri 11-03-16 18:02:24, Vladimir Davydov wrote:
> On Fri, Mar 11, 2016 at 03:30:31PM +0100, Michal Hocko wrote:
[...]
> > Not really. GFP_KERNEL would allow to invoke some shrinkers which are
> > GFP_NOFS incopatible.
> 
> Can't a GFP_NOFS allocation happen when there is no shrinkable objects
> to drop so that there's no real difference between GFP_KERNEL and
> GFP_NOFS?

Yes it can and we do not handle that case even in the global case.
 
[...]
> > > We could ratelimit these messages. Slab charge failures are already
> > > reported to dmesg (see ___slab_alloc -> slab_out_of_memory) and nobody's
> > > complained so far. Are there any non-slab GFP_NOFS allocations charged
> > > to memcg?
> > 
> > I believe there might be some coming from FS via add_to_page_cache_lru.
> > Especially when their mapping gfp_mask clears __GFP_FS. I haven't
> > checked the code deeper but some of those might be called from the page
> > fault path and trigger memcg OOM. I would have to look closer.
> 
> If you think this warning is really a must have, and you don't like to
> warn about every charge failure, may be we could just print info about
> allocation that triggered OOM right in mem_cgroup_oom, like the code
> below does? I think it would be more-or-less equivalent to what we have
> now except it wouldn't require storing gfp_mask on task_struct.
> 
> diff --git a/mm/memcontrol.c b/mm/memcontrol.c
> index a217b1374c32..d8e130d14f5d 100644
> --- a/mm/memcontrol.c
> +++ b/mm/memcontrol.c
> @@ -1604,6 +1604,8 @@ static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
>  	 */
>  	css_get(&memcg->css);
>  	current->memcg_in_oom = memcg;
> +
> +	pr_warn("Process ... triggered OOM in memcg ... gfp ...\n");

Hmm, that could lead to intermixed oom reports and matching the failure
to the particular report would be slighltly harder. But I guess it would
be acceptable if it can help to shrink the task_struct in the end. There
are people (google at least) who rely on the oom reports so I would
asked them if they are OK with that. I do not see any obvious issues
with this.

>  }
>  
>  /**

-- 
Michal Hocko
SUSE Labs