mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
* [PATCH] media: em28xx: clear dev->v4l2 on init failure and fini to avoid double put
@ 2026-11-22 19:00 Arbab Haider
  0 siblings, 0 replies; only message in thread
From: Arbab Haider @ 2026-11-22 19:00 UTC (permalink / raw)
  To: Mauro Carvalho Chehab, Hans Verkuil
  Cc: linux-media, linux-kernel, syzbot+fb0d2df09b56117605fc, stable,
	Arbab Haider

Commit 288254383674 ("media: em28xx: use v4l2_device release callback")
switched em28xx_v4l2 teardown to v4l2_device reference counting, where
dev->v4l2 is only cleared inside em28xx_free_v4l2() once the refcount
reaches zero.

If em28xx_v4l2_init() fails after video_register_device() (for example,
when v4l2_mc_create_media_graph() returns -EINVAL) while userspace has
opened the video node, the unregister_dev error path calls
v4l2_device_put(&v4l2->v4l2_dev) to drop the initial reference, but
em28xx_free_v4l2() does not run yet because the open file handle still
holds a reference, leaving dev->v4l2 non-NULL.

When the USB device is subsequently disconnected, em28xx_v4l2_fini()
sees dev->v4l2 != NULL and calls v4l2_device_put(&v4l2->v4l2_dev) a
second time for the initial reference, freeing struct em28xx_v4l2 while
the video device node is still referenced and causing a
slab-use-after-free in media_device_unregister() /
__media_entity_remove_links().

Fix this by:
- Calling v4l2_device_disconnect() and clearing dev->v4l2 = NULL under
  dev->lock in em28xx_v4l2_init()'s unregister_dev path before dropping
  the initial v4l2_device reference.
- Checking and clearing dev->v4l2 under dev->lock in em28xx_v4l2_fini()
  so the initial reference is dropped at most once.
- Reading dev->v4l2 under dev->lock and checking for !v4l2 in
  em28xx_v4l2_close() so a close racing with teardown cannot dereference
  a NULL pointer.

Tested locally in QEMU (x86_64, KASAN+UBSAN) using dummy_hcd and
raw-gadget emulating an EM2820 device with failed media graph creation
and concurrent /dev/video0 open/disconnect.

Reported-by: syzbot+fb0d2df09b56117605fc@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=fb0d2df09b56117605fc
Fixes: 288254383674 ("media: em28xx: use v4l2_device release callback")
Cc: stable@vger.kernel.org
Signed-off-by: Arbab Haider <arbabhaider649@gmail.com>
---
 drivers/media/usb/em28xx/em28xx-video.c | 23 ++++++++++++++++-------
 1 file changed, 16 insertions(+), 7 deletions(-)

diff --git a/drivers/media/usb/em28xx/em28xx-video.c b/drivers/media/usb/em28xx/em28xx-video.c
index 84424284d..50998aa1c 100644
--- a/drivers/media/usb/em28xx/em28xx-video.c
+++ b/drivers/media/usb/em28xx/em28xx-video.c
@@ -2333,7 +2333,6 @@ static void em28xx_free_v4l2(struct v4l2_device *v4l2_dev)
 
 	v4l2_ctrl_handler_free(&v4l2->ctrl_handler);
 	v4l2_device_unregister(v4l2_dev);
-	dev->v4l2 = NULL;
 	kfree(v4l2);
 	kref_put(&dev->ref, em28xx_free_device);
 }
@@ -2414,7 +2413,7 @@ static int em28xx_v4l2_open(struct file *filp)
  */
 static int em28xx_v4l2_fini(struct em28xx *dev)
 {
-	struct em28xx_v4l2 *v4l2 = dev->v4l2;
+	struct em28xx_v4l2 *v4l2;
 
 	if (dev->is_audio_only) {
 		/* Shouldn't initialize IR for this interface */
@@ -2426,13 +2425,16 @@ static int em28xx_v4l2_fini(struct em28xx *dev)
 		return 0;
 	}
 
-	if (!v4l2)
+	mutex_lock(&dev->lock);
+
+	v4l2 = dev->v4l2;
+	if (!v4l2) {
+		mutex_unlock(&dev->lock);
 		return 0;
+	}
 
 	dev_info(&dev->intf->dev, "Closing video extension\n");
 
-	mutex_lock(&dev->lock);
-
 	v4l2_device_disconnect(&v4l2->v4l2_dev);
 
 	em28xx_uninit_usb_xfer(dev, EM28XX_ANALOG_MODE);
@@ -2455,6 +2457,8 @@ static int em28xx_v4l2_fini(struct em28xx *dev)
 		vb2_video_unregister_device(&v4l2->vdev);
 	}
 
+	dev->v4l2 = NULL;
+
 	mutex_unlock(&dev->lock);
 
 	/*
@@ -2502,18 +2506,19 @@ static int em28xx_v4l2_close(struct file *filp)
 {
 	struct video_device   *vdev = video_devdata(filp);
 	struct em28xx         *dev  = video_drvdata(filp);
-	struct em28xx_v4l2    *v4l2 = dev->v4l2;
+	struct em28xx_v4l2    *v4l2;
 	struct usb_device *udev = interface_to_usbdev(dev->intf);
 	int              err;
 	bool last_user;
 
 	mutex_lock(&dev->lock);
+	v4l2 = dev->v4l2;
 	last_user = v4l2_fh_is_singular_file(filp);
 	_vb2_fop_release(filp, vdev->queue->lock);
 
 	if (last_user) {
 		/* No sense to try to write to the device */
-		if (dev->disconnected)
+		if (dev->disconnected || !v4l2)
 			goto exit;
 
 		/* Save some power by putting tuner to sleep */
@@ -3110,6 +3115,8 @@ static int em28xx_v4l2_init(struct em28xx *dev)
 	return 0;
 
 unregister_dev:
+	v4l2_device_disconnect(&v4l2->v4l2_dev);
+
 	if (video_is_registered(&v4l2->radio_dev)) {
 		dev_info(&dev->intf->dev,
 			 "V4L2 device %s deregistered\n",
@@ -3129,6 +3136,8 @@ static int em28xx_v4l2_init(struct em28xx *dev)
 		vb2_video_unregister_device(&v4l2->vdev);
 	}
 
+	dev->v4l2 = NULL;
+
 	mutex_unlock(&dev->lock);
 
 	/*
-- 
2.53.0


^ permalink raw reply	[flat|nested] only message in thread

only message in thread, other threads:[~2026-11-22 19:01 UTC | newest]

Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2026-11-22 19:00 [PATCH] media: em28xx: clear dev->v4l2 on init failure and fini to avoid double put Arbab Haider

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®