linux-kernel - Re: [PATCH v5 4/6] usb: gadget: add mechanism to specify an explicit status stage

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20190118163106.GC7331@localhost.localdomain>
Date:   Fri, 18 Jan 2019 11:31:06 -0500
From:   Paul Elder <paul.elder@...asonboard.com>
To:     Alan Stern <stern@...land.harvard.edu>
Cc:     laurent.pinchart@...asonboard.com, kieran.bingham@...asonboard.com,
        b-liu@...com, rogerq@...com, balbi@...nel.org,
        gregkh@...uxfoundation.org, linux-usb@...r.kernel.org,
        linux-kernel@...r.kernel.org
Subject: Re: [PATCH v5 4/6] usb: gadget: add mechanism to specify an explicit
 status stage

On Wed, Jan 16, 2019 at 10:06:53AM -0500, Alan Stern wrote:
> On Wed, 16 Jan 2019, Paul Elder wrote:
> 
> > On Mon, Jan 14, 2019 at 10:24:44AM -0500, Alan Stern wrote:
> > > On Mon, 14 Jan 2019, Paul Elder wrote:
> > > 
> > > > > > > Can you check your uvc
> > > > > > > changes using dummy_hcd with the patch below?
> > > > > > 
> > > > > > I'm not sure what to make of the test results. I get the same results
> > > > > > with or without the patch. Which I guess makes sense... in dummy_queue,
> > > > > > this is getting hit when the uvc function driver tries to complete the
> > > > > > delayed status:
> > > > > > 
> > > > > > 	req = usb_request_to_dummy_request(_req);
> > > > > > 	if (!_req || !list_empty(&req->queue) || !_req->complete)
> > > > > > 		return -EINVAL;
> > > > > > 
> > > > > > So the delayed/explicit status stage is never completed, afaict.
> > > > > 
> > > > > I presume you are hitting the !list_empty(&req->queue) test, yes?  The 
> > > > > other two tests are trivial.
> > > > 
> > > > Yes, that is what's happening.
> > > > 
> > > > > Triggering the !list_empty() test means the request has already been
> > > > > submitted and not yet completed.  This probably indicates there is a
> > > > > bug in the uvc function driver code.
> > > > 
> > > > The uvc function driver works with musb, though :/
> > > > 
> > > > I compared the sequence of calls to the uvc setup, completion handler,
> > > > and status stage sending, and for some reason dummy_hcd, after an OUT
> > > > setup-completion-status sequence, calls a completion-status-completion
> > > > sequence, and then goes on the the next request. musb simply goes on to
> > > > the next request after the setup-completion-status sequence.
> > > 
> > > I don't quite understand.  There's a control-OUT transfer, the setup, 
> > > data, and status transactions all complete normally, and then what 
> > > happens?  What do you mean by "a completion-status-completion 
> > > sequence"?  A more detailed description would help.
> > > 
> > 
> > I meant the functions (procedures) in the function driver, so the setup
> > handler (uvc_function_setup), the completion handler
> > (uvc_function_ep0_complete), and the status sender (uvc_send_response),
> > although the last one actually sends the data stage for control IN.
> > So after the status is sent on the uvc gadget driver's end, its
> > completion handler is called again without the setup handler being
> > called beforehand and I cant figure out why.
> 
> Isn't this what you should expect?  Every usb_request, if it is queued
> successfully, eventually gets a completion callback.  That promise is
> made by every UDC driver; it's part of the gadget API.  So for a
> control transfer with a data stage, you expect to have:
> 
> 	Setup handler called
> 	Data-stage request submitted
> 	Data-stage request completion callback
> 	Status-stage request submitted
> 	Status-stage request completion callback
> 
> Thus, two completion callbacks but only one setup callback.

omg how did I not notice this :/

I guess I have to fix the uvc function driver so it works with that.
musb doesn't call the status stage completion callback though; not that
it does anything so it seems fine to me, but indeed the function driver
has to be ready for it if it is called.

> > > > I commented out the paranoia block in dummy_timer, and dummy_hcd still
> > > > does the extra completion, but it doesn't error out anymore. I doubt
> > > > that's the/a solution though, especially since I get:
> > > > 
> > > > [   22.616577] uvcvideo: Failed to query (129) UVC probe control : -75 (exp. 26).
> > > > [   22.624481] uvcvideo: Failed to initialize the device (-5).
> > > > 
> > > > Not sure if that's a result of dummy_hcd not supporting isochronous
> > > > transfers or not.
> > > > 
> > > > I'm not sure where to continue investigating :/
> > > 
> > > Perhaps removing the "#if 0" protecting the dev_dbg line in 
> > > dummy_queue() would provide some helpful output.
> > 
> > It did, but didn't get me much farther :/
> > 
> > > Another thing to check would be if the "implement an emulated 
> > > single-request FIFO" in dummy_queue() is causing trouble.  There's no 
> > > harm in replacing the long "if" condition with "if (0)".
> > 
> > That didn't change anything.
> > 
> > Although I did notice that the dummy_queue that calls the completion
> > handler without the preceeding setup handler says that it's in the
> > status stage (ep->status_stage == 1).
> 
> That is consistent with the events outlined above.


Thanks,

Paul