linux-kernel - Re: [RFC PATCH] mm/vmalloc: fix vmalloc which may return null if called with __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: <CAGsJ_4wW-Z+H=vZ44AJZEhJA1QXgoBn=mN+JUYJxNCUyAadfog@mail.gmail.com>
Date: Thu, 9 May 2024 18:12:58 +1200
From: Barry Song <21cnbao@...il.com>
To: Christoph Hellwig <hch@...radead.org>
Cc: hailong.liu@...o.com, akpm@...ux-foundation.org, urezki@...il.com, 
	lstoakes@...il.com, linux-mm@...ck.org, linux-kernel@...r.kernel.org, 
	xiang@...nel.org, chao@...nel.org, Oven <liyangouwen1@...o.com>
Subject: Re: [RFC PATCH] mm/vmalloc: fix vmalloc which may return null if
 called with __GFP_NOFAIL

On Thu, May 9, 2024 at 4:52 PM Christoph Hellwig <hch@...radead.org> wrote:
>
> On Thu, May 09, 2024 at 02:20:03PM +1200, Barry Song wrote:
> > reverting the fix intended to address the OOM-killer issue in commit
> > dd544141b9eb.
> > Should we indeed permit the NOFAIL flag for large kvmalloc allocations?
>
> What is large?  When you don't allow actually use cases people will
> just reimplement it poorly.  E.g. we'd probably have to add back the
> XFS kmem_ wrappers.

Xiang gave his number 24KiB in erofs, but probably we still have some
"naughty" users allocating much more than 24KiB with NOFAIL in other
subsystems. We should never return NULL for NOFAIL.  So, in any case,
we require Hailong's patch in some form.

However, commit dd544141b9eb ("vmalloc: back off when the current task is
OOM-killed") is also tackling an issue. If we can find a way to preserve its
benefits even in the NOFAIL scenario, it would be preferable, though it
seems improbable. Thus, I'm considering if we could at least include a
WARN_ON_ONCE((gfp & NOFAIL) && size > LARGE) to aid in debugging
potential issues we might encounter. Furthermore, compelling a large
allocation with NOFAIL also appears pointless, as it could lead to
unpredictable long latency.

But I don't know what the proper "LARGE" value is.

Thanks
Barry