Re: [PATCH] usb: gadget: audio: Free requests only after callback

[Peter Chen <peter.chen@nxp.com>]

On 20-10-27 16:31:38, Jack Pham wrote:
> As per the kernel doc for usb_ep_dequeue(), it states that "this
> routine is asynchronous, that is, it may return before the completion
> routine runs". And indeed since v5.0 the dwc3 gadget driver updated
> its behavior to place dequeued requests on to a cancelled list to be
> given back later after the endpoint is stopped.
> 
> The free_ep() was incorrectly assuming that a request was ready to
> be freed after calling dequeue which results in a use-after-free
> in dwc3 when it traverses its cancelled list. Fix this by moving
> the usb_ep_free_request() call to the callback itself in case the
> ep is disabled.
> 
> Fixes: eb9fecb9e69b0 ("usb: gadget: f_uac2: split out audio core")
> Reported-and-tested-by: Ferry Toth <fntoth@gmail.com>
> Signed-off-by: Jack Pham <jackp@codeaurora.org>
> ---
>  drivers/usb/gadget/function/u_audio.c | 8 ++++++--
>  1 file changed, 6 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/usb/gadget/function/u_audio.c b/drivers/usb/gadget/function/u_audio.c
> index e6d32c536781..a3b557fad1fd 100644
> --- a/drivers/usb/gadget/function/u_audio.c
> +++ b/drivers/usb/gadget/function/u_audio.c
> @@ -89,7 +89,12 @@ static void u_audio_iso_complete(struct usb_ep *ep, struct usb_request *req)
>  	struct snd_uac_chip *uac = prm->uac;
>  
>  	/* i/f shutting down */
> -	if (!prm->ep_enabled || req->status == -ESHUTDOWN)
> +	if (!prm->ep_enabled) {
> +		usb_ep_free_request(ep, req);
> +		return;
> +	}
> +
> +	if (req->status == -ESHUTDOWN)
>  		return;
>  
>  	/*
> @@ -337,7 +342,6 @@ static inline void free_ep(struct uac_rtd_params *prm, struct usb_ep *ep)
>  	for (i = 0; i < params->req_number; i++) {
>  		if (prm->ureq[i].req) {
>  			usb_ep_dequeue(ep, prm->ureq[i].req);
> -			usb_ep_free_request(ep, prm->ureq[i].req);

Then, for normal base, eg, there is no pending request before calling
free_ep, then, where these uac request will be freed?

-- 

Thanks,
Peter Chen