linux-kernel - Re: [PATCH] mm: ratelimit oversized kvmalloc warnings instead of once

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: Wed, 19 Jun 2024 10:47:51 -0700
From: Shakeel Butt <shakeel.butt@...ux.dev>
To: Michal Hocko <mhocko@...e.com>
Cc: Andrew Morton <akpm@...ux-foundation.org>,
	Linus Torvalds <torvalds@...ux-foundation.org>,
	kernel-team@...a.com, linux-mm@...ck.org,
	linux-kernel@...r.kernel.org, Kyle McMartin <kyle@...radead.org>
Subject: Re: [PATCH] mm: ratelimit oversized kvmalloc warnings instead of once

On Wed, Jun 19, 2024 at 10:48:07AM +0200, Michal Hocko wrote:
> On Wed 19-06-24 10:30:46, Michal Hocko wrote:
> > On Wed 19-06-24 01:03:16, Shakeel Butt wrote:
> > > On Wed, Jun 19, 2024 at 09:19:41AM GMT, Michal Hocko wrote:
> > > > On Tue 18-06-24 14:34:21, Shakeel Butt wrote:
> > > > > At the moment oversize kvmalloc warnings are triggered once using
> > > > > WARN_ON_ONCE() macro. One issue with this approach is that it only
> > > > > detects the first abuser and then ignores the remaining abusers which
> > > > > complicates detecting all such abusers in a timely manner. The situation
> > > > > becomes worse when the repro has low probability and requires production
> > > > > traffic and thus require large set of machines to find such abusers. In
> > > > > Mera production, this warn once is slowing down the detection of these
> > > > > abusers. Simply replace WARN_ON_ONCE with WARN_RATELIMIT.
> > > > 
> > > > Long time ago, I've had a patch to do the once_per_callsite WARN. I
> > > > cannot find reference at the moment but it used stack depot to note
> > > > stacks that have already triggered. Back then there was no reponse on
> > > > the ML. Should I try to dig deep and recover it from my archives? I
> > > > think this is exactly kind of usecase where it would fit.
> > > > 
> > > 
> > > Do you mean something like warn once per unique call stack?
> > 
> > Exactly!
> > 
> > > If yes then
> > > I think that is better than the simple ratelimiting version as
> > > ratelimiting one may still miss some abusers and also may keep warning
> > > about the same abuser. Please do share your patch.
> > 
> > https://lore.kernel.org/all/20170103134424.28123-1-mhocko@kernel.org/
> 
> Btw. the code has changed a lot since 2017 when this was posted so it
> will likely need a lot of massaging to rebase. Also I am not entirely
> sure it is ok to change WARN_ONCE semantic like that anymore. Maybe we
> need an explicit variant that does this per-call-site warnings. It is a
> notable difference between library functions which can be called from
> different callpaths and those that are used from a single place. I do
> not have much time to dig deeper into this but if you want to take over
> then go ahead. I still think this is a useful WARN_ONCE or in general
> do_something_once semantic.

I think a separate variant like WARN_UNIQUE() would be better. I will
look into this.

Linus, please let me know if you have any concerns on the approach
Michal is suggesting i.e. a variant for warn once for unique call stack.

Shakeel