Received-SPF: pass (google.com: domain of linux-kernel+bounces-72324-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.199.223 as permitted sender) client-ip=147.75.199.223;
Precedence: bulk
MIME-Version: 1.0
References: <000000000000cb40790611bbcffd@google.com> <98b81fbf-4905-4b45-a8c0-33e7399ef84a@collabora.com>
In-Reply-To: <98b81fbf-4905-4b45-a8c0-33e7399ef84a@collabora.com>
From: Tomasz Figa <tfiga@chromium.org>
Date: Tue, 20 Feb 2024 13:49:30 +0900
Message-ID: <CAAFQd5Cs-eeHsCZ6VWrzChTJqd4FSqBS4z6ncd8tPQPNMdN68Q@mail.gmail.com>
Subject: Re: [syzbot] [usb?] [media?] possible deadlock in vb2_video_unregister_device
To: Benjamin Gaignard <benjamin.gaignard@collabora.com>
Cc: syzbot <syzbot+3b1d4b3d5f7a358bf9a9@syzkaller.appspotmail.com>, hdanton@sina.com, 
	hverkuil-cisco@xs4all.nl, hverkuil@xs4all.nl, 
	laurent.pinchart@ideasonboard.com, linux-kernel@vger.kernel.org, 
	linux-media@vger.kernel.org, linux-usb@vger.kernel.org, 
	m.szyprowski@samsung.com, mchehab@kernel.org, syzkaller-bugs@googlegroups.com
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Tue, Feb 20, 2024 at 12:49=E2=80=AFAM Benjamin Gaignard
<benjamin.gaignard@collabora.com> wrote:
>
>
> Le 19/02/2024 =C3=A0 14:10, syzbot a =C3=A9crit :
> > syzbot has bisected this issue to:
> >
> > commit c838530d230bc638d79b78737fc4488ffc28c1ee
> > Author: Benjamin Gaignard <benjamin.gaignard@collabora.com>
> > Date:   Thu Nov 9 16:34:59 2023 +0000
> >
> >      media: media videobuf2: Be more flexible on the number of queue st=
ored buffers
> >
> > bisection log:  https://syzkaller.appspot.com/x/bisect.txt?x=3D166dc872=
180000
> > start commit:   b401b621758e Linux 6.8-rc5
> > git tree:       upstream
> > final oops:     https://syzkaller.appspot.com/x/report.txt?x=3D156dc872=
180000
> > console output: https://syzkaller.appspot.com/x/log.txt?x=3D116dc872180=
000
> > kernel config:  https://syzkaller.appspot.com/x/.config?x=3Deff9f3183d0=
a20dd
> > dashboard link: https://syzkaller.appspot.com/bug?extid=3D3b1d4b3d5f7a3=
58bf9a9
> > syz repro:      https://syzkaller.appspot.com/x/repro.syz?x=3D13ffaae81=
80000
> > C reproducer:   https://syzkaller.appspot.com/x/repro.c?x=3D13ef909c180=
000
> >
> > Reported-by: syzbot+3b1d4b3d5f7a358bf9a9@syzkaller.appspotmail.com
> > Fixes: c838530d230b ("media: media videobuf2: Be more flexible on the n=
umber of queue stored buffers")
> >
> > For information about bisection process see: https://goo.gl/tpsmEJ#bise=
ction
>
> Hans,
> I think the issue occur because of this part of the commit:
> @@ -1264,7 +1264,7 @@ void vb2_video_unregister_device(struct video_devic=
e *vdev)
>           */
>          get_device(&vdev->dev);
>          video_unregister_device(vdev);
> -       if (vdev->queue && vdev->queue->owner) {
> +       if (vdev->queue) {
>                  struct mutex *lock =3D vdev->queue->lock ?
>                          vdev->queue->lock : vdev->lock;
>
> but I wonder if the correction shouldn't be to remove usbtv->vb2q_lock mu=
tex in usbtv_video_free().
>
> Any opinion ?

That is probably what triggered the failure detected by syzbot, but I
don't think we've ever had a guarantee that owner is NULL in that
code.

Looking at the uvc driver [1], it doesn't seem to need any special
locking there (after all the vb2 code acquires them).  (It also
doesn't have the custom clean-up code that the usbtv driver has in
usbtv_stop(), but that's another story). So maybe all we need to fix
it is removing the mutex_lock/unlock() calls?

[1] https://elixir.bootlin.com/linux/v6.8-rc4/source/drivers/media/usb/uvc/=
uvc_driver.c#L1906

Best,
Tomasz