linux-kernel - Re: [PATCH] mm: avoid livelock on !__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: <CAOJsxLEFPhGra8ybpWCP-V8HF_BEUXpWe-tK7MMq0sDcoRFu=w@mail.gmail.com>
Date:	Tue, 25 Oct 2011 11:08:13 +0300
From:	Pekka Enberg <penberg@...helsinki.fi>
To:	Colin Cross <ccross@...roid.com>
Cc:	linux-kernel@...r.kernel.org,
	Andrew Morton <akpm@...ux-foundation.org>,
	Mel Gorman <mgorman@...e.de>,
	KAMEZAWA Hiroyuki <kamezawa.hiroyu@...fujitsu.com>,
	Andrea Arcangeli <aarcange@...hat.com>,
	David Rientjes <rientjes@...gle.com>, linux-mm@...ck.org
Subject: Re: [PATCH] mm: avoid livelock on !__GFP_FS allocations

Hi Colin,

On Tue, Oct 25, 2011 at 9:39 AM, Colin Cross <ccross@...roid.com> wrote:
>>> Under the following conditions, __alloc_pages_slowpath can loop
>>> forever:
>>> gfp_mask & __GFP_WAIT is true
>>> gfp_mask & __GFP_FS is false
>>> reclaim and compaction make no progress
>>> order <= PAGE_ALLOC_COSTLY_ORDER
>>>
>>> These conditions happen very often during suspend and resume,
>>> when pm_restrict_gfp_mask() effectively converts all GFP_KERNEL
>>> allocations into __GFP_WAIT.

On Tue, Oct 25, 2011 at 12:40 AM, Pekka Enberg <penberg@...helsinki.fi> wrote:
>> Why does it do that? Why don't we fix the gfp mask instead?

On Tue, Oct 25, 2011 at 10:51 AM, Colin Cross <ccross@...roid.com> wrote:
> It disables __GFP_IO and __GFP_FS because the IO drivers may be suspended.

Sure but why doesn't it clear __GFP_WAIT too?

>>> The oom killer is not run because gfp_mask & __GFP_FS is false,
>>> but should_alloc_retry will always return true when order is less
>>> than PAGE_ALLOC_COSTLY_ORDER.
>>>
>>> Fix __alloc_pages_slowpath to skip retrying when oom killer is
>>> not allowed by the GFP flags, the same way it would skip if the
>>> oom killer was allowed but disabled.
>>>
>>> Signed-off-by: Colin Cross <ccross@...roid.com>
>>> ---
>>>
>>> An alternative patch would add a did_some_progress argument to
>>> __alloc_pages_may_oom, and remove the checks in
>>> __alloc_pages_slowpath that require knowledge of when
>>> __alloc_pages_may_oom chooses to run out_of_memory. If
>>> did_some_progress was still zero, it would goto nopage whether
>>> or not __alloc_pages_may_oom was actually called.
>>>
>>>  mm/page_alloc.c |    4 ++++
>>>  1 files changed, 4 insertions(+), 0 deletions(-)
>>>
>>> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
>>> index fef8dc3..dcd99b3 100644
>>> --- a/mm/page_alloc.c
>>> +++ b/mm/page_alloc.c
>>> @@ -2193,6 +2193,10 @@ rebalance:
>>>                        }
>>>
>>>                        goto restart;
>>> +               } else {
>>> +                       /* If we aren't going to try the OOM killer, give up */
>>> +                       if (!(gfp_mask & __GFP_NOFAIL))
>>> +                               goto nopage;
>>>                }
>>>        }
>>
>> I don't quite understand how __GFP_WAIT is involved here. Which path
>> is causing the infinite loop?
>
> GFP_KERNEL is __GFP_WAIT | __GFP_IO | __GFP_FS.  Once driver suspend
> has started, gfp_allowed_mask is ~(__GFP_IO | GFP_FS), so any call to
> __alloc_pages_nodemask(GFP_KERNEL, ...) gets masked to effectively
> __alloc_pages_nodemask(__GFP_WAIT, ...).
>
> The loop is in __alloc_pages_slowpath, from the rebalance label to
> should_alloc_retry.  Under the conditions I listed in the commit
> message, there is no path to the nopage label, because all the
> relevant "goto nopage" lines that would normally allow a GFP_KERNEL
> allocation to fail are inside a check for __GFP_FS.

Right. Please include that information in the changelog.

> Modifying the gfp_allowed_mask would not completely fix the issue, a
> GFP_NOIO allocation can meet the conditions outside of suspend.
> gfp_allowed_mask just makes the issue more likely, by converting
> GFP_KERNEL into GFP_NOIO.

Why would anyone want to combine __GFP_WAIT, __GFP_NOFAIL and
!__GFP_IO on purpose? What is it useful for?

As for your patch:

Acked-by: Pekka Enberg <penberg@...nel.org>

but I'd love to hear why we shouldn't also fix the suspend gfp mask to
clear __GFP_WAIT and add a WARN_ON_ONCE to your new code path.

                        Pekka
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/