Re: [PATCH] usbip: give back URBs for unsent unlink requests during cleanup

[Anirudh Rayabharam <mail@anirudhrb.com>]

On Fri, Aug 06, 2021 at 06:47:54PM +0200, Greg Kroah-Hartman wrote:
> On Fri, Aug 06, 2021 at 10:10:14PM +0530, Anirudh Rayabharam wrote:
> > In vhci_device_unlink_cleanup(), the URBs for unsent unlink requests are
> > not given back. This sometimes causes usb_kill_urb to wait indefinitely
> > for that urb to be given back. syzbot has reported a hung task issue [1]
> > for this.
> > 
> > To fix this, give back the urbs corresponding to unsent unlink requests
> > (unlink_tx list) similar to how urbs corresponding to unanswered unlink
> > requests (unlink_rx list) are given back. Since the code is almost the
> > same, extract it into a new function and call it for both unlink_rx and
> > unlink_tx lists.
> > 
> > [1]: https://syzkaller.appspot.com/bug?id=08f12df95ae7da69814e64eb5515d5a85ed06b76
> > 
> > Reported-by: syzbot+74d6ef051d3d2eacf428@syzkaller.appspotmail.com
> > Tested-by: syzbot+74d6ef051d3d2eacf428@syzkaller.appspotmail.com
> > Signed-off-by: Anirudh Rayabharam <mail@anirudhrb.com>
> > ---
> >  drivers/usb/usbip/vhci_hcd.c | 47 ++++++++++++++++++++++++++----------
> >  1 file changed, 34 insertions(+), 13 deletions(-)
> > 
> > diff --git a/drivers/usb/usbip/vhci_hcd.c b/drivers/usb/usbip/vhci_hcd.c
> > index 4ba6bcdaa8e9..45f98aa12895 100644
> > --- a/drivers/usb/usbip/vhci_hcd.c
> > +++ b/drivers/usb/usbip/vhci_hcd.c
> > @@ -945,7 +945,8 @@ static int vhci_urb_dequeue(struct usb_hcd *hcd, struct urb *urb, int status)
> >  	return 0;
> >  }
> >  
> > -static void vhci_device_unlink_cleanup(struct vhci_device *vdev)
> > +static void __vhci_cleanup_unlink_list(struct vhci_device *vdev,
> > +		struct list_head *unlink_list)
> >  {
> >  	struct vhci_hcd *vhci_hcd = vdev_to_vhci_hcd(vdev);
> >  	struct usb_hcd *hcd = vhci_hcd_to_hcd(vhci_hcd);
> > @@ -953,23 +954,25 @@ static void vhci_device_unlink_cleanup(struct vhci_device *vdev)
> >  	struct vhci_unlink *unlink, *tmp;
> >  	unsigned long flags;
> >  
> > +	if (unlink_list != &vdev->unlink_tx
> > +			&& unlink_list != &vdev->unlink_rx) {
> > +		pr_err("Invalid list passed to __vhci_cleanup_unlink_list\n");
> > +		BUG();
> 
> Do not allow the system to crash, that is not ok.
> 
> > +		return;
> 
> This call makes no sense as you just rebooted the machine :(
> 
> Handle errors properly and recover from them and move on.  A single tiny
> driver should not take down the whole system.

The execution can reach only if there is a developer error and they passed
some random list in `unlink_list`. So, BUG() here crashes the kernel and
draws attention to this fact. Is WARN() a better option here? There is
no way to recover from this and continue with the rest of the function.
Either we WARN()/BUG() or we return silently.

Thanks!

	- Anirudh.