linux-kernel - Re: [PATCH 1/2] Revert "vmalloc: back off when the current task is killed"

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]

Date:   Thu, 5 Oct 2017 19:36:17 +0900
From:   Tetsuo Handa <penguin-kernel@...ove.SAKURA.ne.jp>
To:     Michal Hocko <mhocko@...nel.org>,
        Johannes Weiner <hannes@...xchg.org>
Cc:     Andrew Morton <akpm@...ux-foundation.org>,
        Alan Cox <alan@...yncelyn.cymru>,
        Christoph Hellwig <hch@....de>, linux-mm@...ck.org,
        linux-kernel@...r.kernel.org, kernel-team@...com
Subject: Re: [PATCH 1/2] Revert "vmalloc: back off when the current task is
 killed"

On 2017/10/05 16:57, Michal Hocko wrote:
> On Wed 04-10-17 19:18:21, Johannes Weiner wrote:
>> On Wed, Oct 04, 2017 at 03:32:45PM -0700, Andrew Morton wrote:
> [...]
>>> You don't think they should be backported into -stables?
>>
>> Good point. For this one, it makes sense to CC stable, for 4.11 and
>> up. The second patch is more of a fortification against potential
>> future issues, and probably shouldn't go into stable.
> 
> I am not against. It is true that the memory reserves depletion fix was
> theoretical because I haven't seen any real life bug. I would argue that
> the more robust allocation failure behavior is a stable candidate as
> well, though, because the allocation can fail regardless of the vmalloc
> revert. It is less likely but still possible.
> 

I don't want this patch backported. If you want to backport,
"s/fatal_signal_pending/tsk_is_oom_victim/" is the safer way.

On 2017/10/04 17:33, Michal Hocko wrote:
> Now that we have cd04ae1e2dc8 ("mm, oom: do not rely on TIF_MEMDIE for
> memory reserves access") the risk of the memory depletion is much
> smaller so reverting the above commit should be acceptable. 

Are you aware that stable kernels do not have cd04ae1e2dc8 ?

We added fatal_signal_pending() check inside read()/write() loop
because one read()/write() request could consume 2GB of kernel memory.

What if there is a kernel module which uses vmalloc(1GB) from some
ioctl() for legitimate reason? You are going to allow such vmalloc()
calls to deplete memory reserves completely.

On 2017/10/05 8:21, Johannes Weiner wrote:
> Generally, we should leave it to the page allocator to handle memory
> reserves, not annotate random alloc_page() callsites.

I disagree. Interrupting the loop as soon as possible is preferable.

Since we don't have __GFP_KILLABLE, we had to do fatal_signal_pending()
check inside read()/write() loop. Since vmalloc() resembles read()/write()
in a sense that it can consume GB of memory, it is pointless to expect
the caller of vmalloc() to check tsk_is_oom_victim().

Again, checking tsk_is_oom_victim() inside vmalloc() loop is the better.