Re: general protection fault in usb_set_interface

[Alan Stern]

On Tue, 17 Sep 2019, syzbot wrote:

> Hello,
> 
> syzbot has tested the proposed patch but the reproducer still triggered  
> crash:
> possible deadlock in vidioc_querycap
> 
> ============================================
> WARNING: possible recursive locking detected
> 5.3.0-rc7+ #0 Not tainted
> --------------------------------------------
> v4l_id/3016 is trying to acquire lock:
> 0000000069c3004e (&usbvision->v4l2_lock){+.+.}, at:  
> vidioc_querycap+0x62/0x3b0 drivers/media/usb/usbvision/usbvision-video.c:456
> 
> but task is already holding lock:
> 0000000069c3004e (&usbvision->v4l2_lock){+.+.}, at:  
> __video_do_ioctl+0x3ba/0xba0 drivers/media/v4l2-core/v4l2-ioctl.c:2846

Heh.  That's what comes of trying to patch a driver when you aren't an 
expert on it already.

Okay, the lock is already held at this point so we don't need to 
acquire it.

Alan Stern

#syz test: https://github.com/google/kasan.git f0df5c1b

 drivers/media/usb/usbvision/usbvision-video.c |    6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

Index: usb-devel/drivers/media/usb/usbvision/usbvision-video.c
===================================================================
--- usb-devel.orig/drivers/media/usb/usbvision/usbvision-video.c
+++ usb-devel/drivers/media/usb/usbvision/usbvision-video.c
@@ -453,6 +453,9 @@ static int vidioc_querycap(struct file *
 {
 	struct usb_usbvision *usbvision = video_drvdata(file);
 
+	if (!usbvision->dev)
+		return -ENODEV;
+
 	strscpy(vc->driver, "USBVision", sizeof(vc->driver));
 	strscpy(vc->card,
 		usbvision_device_data[usbvision->dev_model].model_string,
@@ -1111,7 +1114,8 @@ static int usbvision_radio_close(struct
 	mutex_lock(&usbvision->v4l2_lock);
 	/* Set packet size to 0 */
 	usbvision->iface_alt = 0;
-	usb_set_interface(usbvision->dev, usbvision->iface,
+	if (usbvision->dev)
+		usb_set_interface(usbvision->dev, usbvision->iface,
 				    usbvision->iface_alt);
 
 	usbvision_audio_off(usbvision);