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

[Greg KH <gregkh@linuxfoundation.org>]

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.

I do not understand, sorry.  What do you mean by "outside the kernel"?

> kernel log:
> [ 9518.779148][ 1] [ T4650] pc : usb_ifnum_to_if+0x34/0x50
> [ 9518.784360][ 1] [ T4650] lr : usb_hcd_alloc_bandwidth+0x260/0x348

What is the [TXXX] stuff?

> [ 9518.790439][ 1] [ T4650] sp : ffffffa25a6c79d0
> [ 9518.794868][ 1] [ T4650] x29: ffffffa25a6c79d0 x28: 0000000040045613
> [ 9518.801294][ 1] [ T4650] x27: 0000000000000000 x26: 0000000000000000
> [ 9518.807720][ 1] [ T4650] x25: ffffffa2e1597408 x24: ffffffa2e1597408
> [ 9518.814146][ 1] [ T4650] x23: ffffffa2e15974f8 x22: 0000000000000000
> [ 9518.820572][ 1] [ T4650] x21: ffffffa2e9acc000 x20: ffffffa2e6712000
> [ 9518.826998][ 1] [ T4650] x19: ffffffa2e6a8a800 x18: 0000000000000000
> [ 9518.833423][ 1] [ T4650] x17: 0000007fbb91b4b0 x16: ffffffc01016a170
> [ 9518.839849][ 1] [ T4650] x15: 0000000000000000 x14: 0845c02202702800
> [ 9518.846275][ 1] [ T4650] x13: 0000000000000001 x12: 0000000000000000
> [ 9518.852700][ 1] [ T4650] x11: 0000000000000400 x10: ffffffff89e5d720
> [ 9518.859126][ 1] [ T4650] x9 : 0000000000000000 x8 : 0000000000000000
> [ 9518.865551][ 1] [ T4650] x7 : ffffffa2fff1e440 x6 : ffffffa28175c900
> [ 9518.871977][ 1] [ T4650] x5 : 0000000000000060 x4 : ffffffa2e9bc54b0
> [ 9518.878403][ 1] [ T4650] x3 : ffffffa2e9bc54a0 x2 : ffffffa2e9bc54a0
> [ 9518.884828][ 1] [ T4650] x1 : 0000000000000001 x0 : 0000000000000000
> [ 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 ]---
> 
> To resolve this issue, a null pointer check for config->interface[i]
> can be added in the usb_ifnum_to_if() function.
> 
> Signed-off-by: Jie Deng <dengjie03@kylinos.cn>
> ---
>  drivers/usb/core/usb.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/usb/core/usb.c b/drivers/usb/core/usb.c
> index e740f7852bcd..85dcda06a838 100644
> --- a/drivers/usb/core/usb.c
> +++ b/drivers/usb/core/usb.c
> @@ -355,7 +355,7 @@ struct usb_interface *usb_ifnum_to_if(const struct usb_device *dev,
>  	if (!config)
>  		return NULL;
>  	for (i = 0; i < config->desc.bNumInterfaces; i++)
> -		if (config->interface[i]->altsetting[0]
> +		if (config->interface[i] && config->interface[i]->altsetting[0]

Are you sure this is not just papering over a race?  What prevents this
from changing right after you check it?

And what is causing this to happen now?  What commit broke this to
require this change?

thanks,

greg k-h