Re: [PATCH] usb: core: fix NULL dereference in usb_ifnum_to_if() during device removal

[Jie Deng]

在 2025/11/13 22:50, Alan Stern 写道:
> On Thu, Nov 13, 2025 at 07:44:11PM +0800, Jie Deng wrote:
> > During USB device hot-unplug, in the time window between when
> > usb_disconnect() calls usb_disable_device() to set
> > dev->actconfig->interface[i] to NULL but before dev->actconfig
> > is set to NULL.At this point, outside the kernel, usb_ifnum_to_if()
> > is called through usb_set_interface(), and usb_ifnum_to_if() continues
> > to access interface[i]->altsetting[i], triggering a null pointer.
> > [ 9518.891254][ 1] [ T4650] Call trace:
> > [ 9518.894817][ 1] [ T4650]  usb_ifnum_to_if+0x34/0x50
> > [ 9518.899681][ 1] [ T4650]  usb_set_interface+0x108/0x3c8
> > [ 9518.904898][ 1] [ T4650]  uvc_video_stop_streaming+0x3c/0x90 [uvcvideo]
> > [ 9518.911500][ 1] [ T4650]  uvc_stop_streaming+0x24/0x90 [uvcvideo]
> > [ 9518.917583][ 1] [ T4650]  __vb2_queue_cancel+0x44/0x458 [videobuf2_common]
> > [ 9518.924444][ 1] [ T4650]  vb2_core_streamoff+0x20/0xb8 [videobuf2_common]
> > [ 9518.931221][ 1] [ T4650]  vb2_streamoff+0x18/0x60 [videobuf2_v4l2]
> > [ 9518.937390][ 1] [ T4650]  uvc_queue_streamoff+0x30/0x50 [uvcvideo]
> > [ 9518.943557][ 1] [ T4650]  uvc_ioctl_streamoff+0x40/0x68 [uvcvideo]
> > [ 9518.949724][ 1] [ T4650]  v4l_streamoff+0x20/0x28
> > [ 9518.954415][ 1] [ T4650]  __video_do_ioctl+0x17c/0x3e0
> > [ 9518.959540][ 1] [ T4650]  video_usercopy+0x1d8/0x558
> > [ 9518.964490][ 1] [ T4650]  video_ioctl2+0x14/0x1c
> > [ 9518.969094][ 1] [ T4650]  v4l2_ioctl+0x3c/0x58
> > [ 9518.973526][ 1] [ T4650]  do_vfs_ioctl+0x374/0x7b0
> > [ 9518.978304][ 1] [ T4650]  ksys_ioctl+0x78/0xa8
> > [ 9518.982734][ 1] [ T4650]  sys_ioctl+0xc/0x18
> > [ 9518.986991][ 1] [ T4650]  __sys_trace_return+0x0/0x4
> > [ 9518.991943][ 1] [ T4650] Code: eb04005f 54000100 f9400040 91002042 (f9400003)
> > [ 9518.999153][ 1] [ T4650] ---[ end trace f7c7d3236806d9a4 ]---
> This looks like a bug in the uvc driver.  usb_disable_device() unbinds
> ther interface's driver before it sets dev->actconfig->interface[i] to
> NULL, and the uvc driver shouldn't call usb_set_interface() after it has
> been unbound.
Thank you for your correction.
> > To resolve this issue, a null pointer check for config->interface[i]
> > can be added in the usb_ifnum_to_if() function.
> That won't fix the real bug.  You need to change the uvc driver.

I will fix this issue from the uvc driver and submit a patch.

> Alan Stern