Re: [PATCH v1 0/3] usb: gadget: uvc: stability fixes on STREAMOFF.

[Avichal Rakesh <arakesh@google.com>]

Thank you for testing the patch, Michael!

On 10/3/23 04:09, Michael Grzeschik wrote:
> Hi
> 
> On Sat, Sep 30, 2023 at 11:48:18AM -0700, Avichal Rakesh wrote:
>> We have been seeing two main stability issues that uvc gadget driver
>> runs into when stopping streams:
>> 1. Attempting to queue usb_requests to a disabled usb_ep
>> 2. use-after-free issue for inflight usb_requests
>>
>> The three patches below fix the two issues above. Patch 1/3 fixes the
>> first issue, and Patch 2/3 and 3/3 fix the second issue.
>>
>> Avichal Rakesh (3):
>>  usb: gadget: uvc: prevent use of disabled endpoint
>>  usb: gadget: uvc: Allocate uvc_requests one at a time
>>  usb: gadget: uvc: Fix use-after-free for inflight usb_requests
>>
>> drivers/usb/gadget/function/f_uvc.c     |  11 +-
>> drivers/usb/gadget/function/f_uvc.h     |   2 +-
>> drivers/usb/gadget/function/uvc.h       |   6 +-
>> drivers/usb/gadget/function/uvc_v4l2.c  |  21 ++-
>> drivers/usb/gadget/function/uvc_video.c | 189 +++++++++++++++++-------
>> 5 files changed, 164 insertions(+), 65 deletions(-)
> 
> These patches are not applying on gregkh/usb-testing since
> Greg did take my patches first. I have already rebased them.

Ah, I didn't realize Greg had picked up your changes in his tree.
Rebased the patches in V2.

Also want to point out that 
https://lore.kernel.org/20230911002451.2860049-2-m.grzeschik@pengutronix.de/
may have introduced an issue when setting the uvc->state to 
UVC_STATE_CONNECTED in uvcg_video_enable. 

Effectively, uvc_video_enable can be called 
  1. because the host asks to halt the stream, or 
  2. if the gadget was disabled. 

Setting uvc->state to CONNECTED is fine for (1). In (2), uvc_v4l2 
sets uvc->state to DISCONNECTED before calling uvcg_video_enable. 
In this case, your change would overwrite the value to CONNECTED 
without any further checks.

> 
> In the updated version I the stack runs into the
> following error, when enabling lockdep. Could you
> try your version with lockdep enabled?
> 
> [   41.278520] configfs-gadget.vz gadget.0: uvc: reset UVC
> [   47.156261] configfs-gadget.vz gadget.0: uvc: uvc_function_set_alt(2, 0)
> [   47.169177]
> [   47.170903] ============================================
> [   47.176857] WARNING: possible recursive locking detected
> [   47.182798] 6.5.0-20230919-1+ #19 Tainted: G         C
> [   47.189323] --------------------------------------------
> [   47.195256] vzuvcd/412 is trying to acquire lock:
> [   47.200511] ffffff8009560928 (&video->req_lock){....}-{3:3}, at: uvc_video_complete+0x44/0x2e0
> [   47.210172]
> [   47.210172] but task is already holding lock:
> [   47.216687] ffffff8009560928 (&video->req_lock){....}-{3:3}, at: uvcg_video_enable+0x2d0/0x5c0
> [   47.226333]
> [   47.226333] other info that might help us debug this:
> [   47.233625]  Possible unsafe locking scenario:
> [   47.233625]
> [   47.240242]        CPU0
> [   47.242974]        ----
> [   47.245709]   lock(&video->req_lock);
> [   47.249802]   lock(&video->req_lock);
> [   47.253897]
> [   47.253897]  *** DEADLOCK ***
> [   47.253897]
> [   47.260511]  May be due to missing lock nesting notation
> [   47.260511]
> 

Thank you for catching this. I couldn't repro the deadlock, but it looks
like DWC3 controller can call the complete callback directly from 
usb_ep_dequeue for requests in its pending list. I am not sure if that 
is up to spec, considering calling usb_ep_queue explicitly forbids calling
the complete callback from within. Regardless, it is easy enough to fix
by moving the dequeue calls to before we acquire req_lock. The semantics
of the rest of the patch is unchanged.

Uploading the fixed patches shortly.

Regards,
Avi.