On 10/7/20 9:48 AM, Shuah Khan wrote:
> On 10/7/20 8:28 AM, Andrey Konovalov wrote:
>> On Wed, Oct 7, 2020 at 3:56 PM Shuah Khan <skhan@linuxfoundation.org> 
>> wrote:
>>>

[snip]

>>>
>>> Hi Andrey,
>>>
>>> I have been unable to reproduce the problem with the reproducer
>>> so far. You mentioned it happens quite often.
>>>
>>> - matched config with yours
>>> - load vhci_hcd module and run the reproducer
>>
>> Hm, if you matched the config, then the module should be built-in?
>>
> 
> Right. I did notice that your config has built-in. This shouldn't
> matter, I have a kernel built with it static. I will try it to
> see if it makes a difference.
> 
>>>
>>> I do see the messages during shutdown - stop threads etc.
>>>
>>> What am I missing?
>>
>> This appears to be a race that requires precise timings. I failed to
>> reproduce it with the C reproducer, but I managed to reproduce it with
>> the syzkaller repro program:
>>
>> https://syzkaller.appspot.com/x/repro.syz?x=16cbaa7d900000
>>
>> To do that you need to build syzkaller, and copy ./bin/syz-execprog
>> and ./bin/syz-executor into your testing environment, and then do:
>>
>> ./syz-execprog -sandbox=none -repeat=0 -procs=6 ./repro.prog
>>
> 

Andrey,

I am unable to reproduce the problem. I even tweaked the reproducer
to launch 10 procs and 100 threads.

Can you test the following patch with your setup

Here it is - also attached.

=======================================================================
diff --git a/drivers/usb/usbip/vhci_hcd.c b/drivers/usb/usbip/vhci_hcd.c
index 66cde5e5f796..3fdcb93b667d 100644
--- a/drivers/usb/usbip/vhci_hcd.c
+++ b/drivers/usb/usbip/vhci_hcd.c
@@ -997,14 +997,12 @@ static void vhci_device_unlink_cleanup(struct 
vhci_device *vdev)
  	spin_unlock_irqrestore(&vhci->lock, flags);
  }

-/*
- * The important thing is that only one context begins cleanup.
- * This is why error handling and cleanup become simple.
- * We do not want to consider race condition as possible.
- */
  static void vhci_shutdown_connection(struct usbip_device *ud)
  {
  	struct vhci_device *vdev = container_of(ud, struct vhci_device, ud);
+	struct vhci_hcd *vhci_hcd = vdev_to_vhci_hcd(vdev);
+	struct vhci *vhci = vhci_hcd->vhci;
+	unsigned long flags;

  	/* need this? see stub_dev.c */
  	if (ud->tcp_socket) {
@@ -1012,6 +1010,10 @@ static void vhci_shutdown_connection(struct 
usbip_device *ud)
  		kernel_sock_shutdown(ud->tcp_socket, SHUT_RDWR);
  	}

+	/* avoid races during shutdown */
+	spin_lock_irqsave(&vhci->lock, flags);
+	spin_lock(&vdev->priv_lock);
+
  	/* kill threads related to this sdev */
  	if (vdev->ud.tcp_rx) {
  		kthread_stop_put(vdev->ud.tcp_rx);
@@ -1031,6 +1033,10 @@ static void vhci_shutdown_connection(struct 
usbip_device *ud)
  	}
  	pr_info("release socket\n");

+	/* unlock - vhci_device_unlink_cleanup() holds the locks */
+	spin_unlock(&vdev->priv_lock);
+	spin_unlock_irqrestore(&vhci->lock, flags);
+
  	vhci_device_unlink_cleanup(vdev);

  	/*
=======================================================================

thanks,
-- Shuah