[PATCH v3] media: em28xx: clear dev->v4l2 when v4l2 init fails

[Diego Fernando Mancera Gomez <diegomancera.dev@gmail.com>]

In em28xx_v4l2_init(), if an error occurs after video_register_device()
succeeds, the unregister_dev error path unregisters the video devices and
drops the initial v4l2_dev reference, but leaves dev->v4l2 set.

An em28xx_v4l2_open() call already in progress (such as from v4l_id)
that blocked on dev->lock while em28xx_v4l2_init() held it will then
find dev->v4l2 still non-NULL and succeed on an unregistered node
instead of returning -ENODEV. When the device is later disconnected,
em28xx_v4l2_fini() also sees dev->v4l2 is still set and calls
v4l2_device_put() a second time, dropping the refcount to zero and
freeing v4l2 while v4l_id still holds it, causing a slab-use-after-free
in v4l2_open().

Fix this by clearing dev->v4l2 under dev->lock in the unregister_dev
error path of em28xx_v4l2_init().

Fixes: 288254383674 ("media: em28xx: use v4l2_device release callback")
Cc: stable@vger.kernel.org
Reported-by: syzbot+1a7507a194fff09e5c44@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=1a7507a194fff09e5c44
Signed-off-by: Diego Fernando Mancera Gomez <diegomancera.dev@gmail.com>
---
Changes in v3:
- Resend: v2 reached lore with the "diff --git" header line split in
  two, so the Media CI robot could not apply it. No code changes.

Changes in v2:
- Drop the non-standard 'Assisted-by: Gemini' trailer that failed
  checkpatch.

Note: compile-tested drivers/media/usb/em28xx/em28xx-video.o with
allmodconfig (no warnings); not runtime-tested locally against the
syzbot reproducer.

 drivers/media/usb/em28xx/em28xx-video.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/drivers/media/usb/em28xx/em28xx-video.c b/drivers/media/usb/em28xx/em28xx-video.c
index 84424284d486..f78325a41124 100644
--- a/drivers/media/usb/em28xx/em28xx-video.c
+++ b/drivers/media/usb/em28xx/em28xx-video.c
@@ -3129,6 +3129,12 @@ static int em28xx_v4l2_init(struct em28xx *dev)
 		vb2_video_unregister_device(&v4l2->vdev);
 	}
 
+	/*
+	 * Clear dev->v4l2 to prevent an in-progress em28xx_v4l2_open() from
+	 * succeeding on an unregistered node (returning -ENODEV instead) and
+	 * to prevent em28xx_v4l2_fini() from tearing down the instance again.
+	 */
+	dev->v4l2 = NULL;
 	mutex_unlock(&dev->lock);
 
 	/*
-- 
2.56.1