linux-kernel - Re: [PATCH] pidfs: detect refcount bugs

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: <20250509-verlieben-respekt-44058457ea59@brauner>
Date: Fri, 9 May 2025 12:34:22 +0200
From: Christian Brauner <brauner@...nel.org>
To: Mateusz Guzik <mjguzik@...il.com>
Cc: Oleg Nesterov <oleg@...hat.com>, linux-fsdevel@...r.kernel.org, 
	linux-kernel@...r.kernel.org
Subject: Re: [PATCH] pidfs: detect refcount bugs

On Tue, May 06, 2025 at 04:43:56PM +0200, Mateusz Guzik wrote:
> On Tue, May 06, 2025 at 01:55:54PM +0200, Christian Brauner wrote:
> > Now that we have pidfs_{get,register}_pid() that needs to be paired with
> > pidfs_put_pid() it's possible that someone pairs them with put_pid().
> > Thus freeing struct pid while it's still used by pidfs. Notice when that
> > happens. I'll also add a scheme to detect invalid uses of
> > pidfs_get_pid() and pidfs_put_pid() later.
> > 
> > Signed-off-by: Christian Brauner <brauner@...nel.org>
> > ---
> >  kernel/pid.c | 1 +
> >  1 file changed, 1 insertion(+)
> > 
> > diff --git a/kernel/pid.c b/kernel/pid.c
> > index 26f1e136f017..8317bcbc7cf7 100644
> > --- a/kernel/pid.c
> > +++ b/kernel/pid.c
> > @@ -100,6 +100,7 @@ void put_pid(struct pid *pid)
> >  
> >  	ns = pid->numbers[pid->level].ns;
> >  	if (refcount_dec_and_test(&pid->count)) {
> > +		WARN_ON_ONCE(pid->stashed);
> >  		kmem_cache_free(ns->pid_cachep, pid);
> >  		put_pid_ns(ns);
> >  	}
> > -- 
> > 2.47.2
> > 
> 
> With the patch as proposed you are only catching the misuse if this is
> the last ref though.
> 
> iow, the check should be hoisted above unrefing?

No, not really. If there's more than one reference then pid->stashed can
be legimitately != NULL.