linux-kernel - Re: [PATCH 2/2] mm, oom: do not enfore OOM killer for __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: <20161218163727.GC8440@dhcp22.suse.cz>
Date:   Sun, 18 Dec 2016 17:37:27 +0100
From:   Michal Hocko <mhocko@...nel.org>
To:     Tetsuo Handa <penguin-kernel@...ove.SAKURA.ne.jp>
Cc:     Johannes Weiner <hannes@...xchg.org>,
        Nils Holland <nholland@...ys.org>,
        linux-kernel@...r.kernel.org, linux-mm@...ck.org,
        Chris Mason <clm@...com>, David Sterba <dsterba@...e.cz>,
        linux-btrfs@...r.kernel.org
Subject: Re: [PATCH 2/2] mm, oom: do not enfore OOM killer for __GFP_NOFAIL
 automatically

On Sat 17-12-16 20:17:07, Tetsuo Handa wrote:
[...]
> I feel that allowing access to memory reserves based on __GFP_NOFAIL might not
> make sense. My understanding is that actual I/O operation triggered by I/O
> requests by filesystem code are processed by other threads. Even if we grant
> access to memory reserves to GFP_NOFS | __GFP_NOFAIL allocations by fs code,
> I think that it is possible that memory allocations by underlying bio code
> fails to make a further progress unless memory reserves are granted as well.

IO layer should rely on mempools to guarantee a forward progress.

> Below is a typical trace which I observe under OOM lockuped situation (though
> this trace is from an OOM stress test using XFS).
> 
> ----------------------------------------
> [ 1845.187246] MemAlloc: kworker/2:1(14498) flags=0x4208060 switches=323636 seq=48 gfp=0x2400000(GFP_NOIO) order=0 delay=430400 uninterruptible
> [ 1845.187248] kworker/2:1     D12712 14498      2 0x00000080
> [ 1845.187251] Workqueue: events_freezable_power_ disk_events_workfn
> [ 1845.187252] Call Trace:
> [ 1845.187253]  ? __schedule+0x23f/0xba0
> [ 1845.187254]  schedule+0x38/0x90
> [ 1845.187255]  schedule_timeout+0x205/0x4a0
> [ 1845.187256]  ? del_timer_sync+0xd0/0xd0
> [ 1845.187257]  schedule_timeout_uninterruptible+0x25/0x30
> [ 1845.187258]  __alloc_pages_nodemask+0x1035/0x10e0
> [ 1845.187259]  ? alloc_request_struct+0x14/0x20
> [ 1845.187261]  alloc_pages_current+0x96/0x1b0
> [ 1845.187262]  ? bio_alloc_bioset+0x20f/0x2e0
> [ 1845.187264]  bio_copy_kern+0xc4/0x180
> [ 1845.187265]  blk_rq_map_kern+0x6f/0x120
> [ 1845.187268]  __scsi_execute.isra.23+0x12f/0x160
> [ 1845.187270]  scsi_execute_req_flags+0x8f/0x100
> [ 1845.187271]  sr_check_events+0xba/0x2b0 [sr_mod]
> [ 1845.187274]  cdrom_check_events+0x13/0x30 [cdrom]
> [ 1845.187275]  sr_block_check_events+0x25/0x30 [sr_mod]
> [ 1845.187276]  disk_check_events+0x5b/0x150
> [ 1845.187277]  disk_events_workfn+0x17/0x20
> [ 1845.187278]  process_one_work+0x1fc/0x750
> [ 1845.187279]  ? process_one_work+0x167/0x750
> [ 1845.187279]  worker_thread+0x126/0x4a0
> [ 1845.187280]  kthread+0x10a/0x140
> [ 1845.187281]  ? process_one_work+0x750/0x750
> [ 1845.187282]  ? kthread_create_on_node+0x60/0x60
> [ 1845.187283]  ret_from_fork+0x2a/0x40
> ----------------------------------------
> 
> I think that this GFP_NOIO allocation request needs to consume more memory reserves
> than GFP_NOFS allocation request to make progress. 

AFAIU, this is an allocation path which doesn't block a forward progress
on a regular IO. It is merely a check whether there is a new medium in
the CDROM (aka regular polling of the device). I really fail to see any
reason why this one should get any access to memory reserves at all.

I actually do not see any reason why it should be NOIO in the first
place but I am not familiar with this code much so there might be some
reasons for that. The fact that it might stall under a heavy memory
pressure is sad but who actually cares?

> Do we want to add __GFP_NOFAIL to this GFP_NOIO allocation request
> in order to allow access to memory reserves as well as GFP_NOFS |
> __GFP_NOFAIL allocation request?

Why?

-- 
Michal Hocko
SUSE Labs