linux-kernel - Re: [PATCH cgroup/for-6.19] cgroup: Fix sleeping from invalid context warning on PREEMPT

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] [day] [month] [year] [list]

Message-ID: <20251106180616.tqGlAVGX@linutronix.de>
Date: Thu, 6 Nov 2025 19:06:16 +0100
From: Sebastian Andrzej Siewior <bigeasy@...utronix.de>
To: Tejun Heo <tj@...nel.org>
Cc: Calvin Owens <calvin@...nvd.org>, linux-kernel@...r.kernel.org,
	Dan Schatzberg <dschatzberg@...a.com>,
	Peter Zijlstra <peterz@...radead.org>
Subject: Re: [PATCH cgroup/for-6.19] cgroup: Fix sleeping from invalid
 context warning on PREEMPT_RT

On 2025-11-06 07:55:02 [-1000], Tejun Heo wrote:
> Hello,
Hi,

> On Thu, Nov 06, 2025 at 06:46:14PM +0100, Sebastian Andrzej Siewior wrote:
> > On 2025-11-06 07:37:16 [-1000], Tejun Heo wrote:
> > > Will switch to IRQ_WORK_LAZY_INIT.
> > 
> > Quick question: Since it is not important at all, would it work to have
> > it in task's RCU callback, __put_task_struct()?
> 
> It doesn't have to run right away but it better run in some definite time
> frame because at this point the task is not visible from userspace otherwise
> (it doesn't have a pid) but are still pinning the cgroup, so we're in this
> limbo state where reading cgroup.procs should return empty (there may be a
> bug here right now. I think the code will try to deref NULL pid pointer) but
> the cgroup is not empty. This window is not really broken in itself because
> cgroup empty state is tracked and notified separately. However, task_struct
> can be pinned and can linger for indefinite amount of time after being dead,
> and that would become an actual problem.
> 
> So, to add a bit of qualitifier, while it's okay to run it with some amount
> of delay that's not very significant to human perception, we definitely
> don't want to allow delaying it indefinitely.

Okay. This is some arguing that would justify the additional extension
of task_struct :)
Understood.

> Thanks.

Sebastian