Received-SPF: pass (google.com: domain of linux-kernel+bounces-22472-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.48.161 as permitted sender) client-ip=147.75.48.161;
Date: Wed, 10 Jan 2024 17:03:20 +0100
From: Oleg Nesterov <oleg@redhat.com>
To: Linus Torvalds <torvalds@linux-foundation.org>
Cc: syzbot <syzbot+c6d438f2d77f96cae7c2@syzkaller.appspotmail.com>,
	"Eric W. Biederman" <ebiederm@xmission.com>,
	linux-kernel@vger.kernel.org, luto@kernel.org,
	michael.christie@oracle.com, mst@redhat.com, peterz@infradead.org,
	syzkaller-bugs@googlegroups.com, tglx@linutronix.de
Subject: Re: [syzbot] [kernel?] WARNING in signal_wake_up_state
Message-ID: <20240110160319.GA21002@redhat.com>
References: <000000000000a41b82060e875721@google.com>
 <CAHk-=wgM=MmqrQC-qgXoSehW=itHaqOUiBfN8jRBGAHn1=D0tg@mail.gmail.com>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <CAHk-=wgM=MmqrQC-qgXoSehW=itHaqOUiBfN8jRBGAHn1=D0tg@mail.gmail.com>
User-Agent: Mutt/1.5.24 (2015-08-30)

On 01/09, Linus Torvalds wrote:
>
> Oleg/Eric, can you make any sense of this?
>
> On Tue, 9 Jan 2024 at 10:18, syzbot
> <syzbot+c6d438f2d77f96cae7c2@syzkaller.appspotmail.com> wrote:
> >
> > The issue was bisected to:
> >
> > commit f9010dbdce911ee1f1af1398a24b1f9f992e0080
>
> Hmm. This smells more like a "that triggers the problem" than a cause.
>
> Because the warning itself is
>
> > WARNING: CPU: 1 PID: 5069 at kernel/signal.c:771 signal_wake_up_state+0xfa/0x120 kernel/signal.c:771
>
> That's
>
>         lockdep_assert_held(&t->sighand->siglock);

I have a fever, possibly I am totally confused, but this commit added

+               /* Don't require de_thread to wait for the vhost_worker */
+               if ((t->flags & (PF_IO_WORKER | PF_USER_WORKER)) != PF_USER_WORKER)
+                       count++;

into zap_other_threads().

So it seems the caller can do unshare_sighand() before vhost thread exits and
actually unshare ->sighand because oldsighand->count > 1.

This is already very wrong (plus it seems this breaks the signal->notify_count
logic). IIRC I even tried to argue with this change... not sure.

And this can explain the warning, this task can start the coredump after exec
and hit vhost_worker with the old sighand != current->sighand.

Oleg.