On 29/11/2022 10:23, Michael Grzeschik wrote:
On Tue, Nov 29, 2022 at 05:10:24AM +0200, Laurent Pinchart wrote:
Hi Michael,
(CC'ing Dan)
Thank you for the patch.
On Mon, Nov 28, 2022 at 11:31:25AM +0100, Michael Grzeschik wrote:
When the userspace gets the setup requests for UVC_GET_CUR UVC_GET_MIN,
UVC_GET_MAX, UVC_GET_DEF it will fill out the ctrl response. This data
needs to be validated. Since the kernel also knows the limits for valid
cases, it can fixup the values in case the userspace is setting invalid
data.
Why is this a good idea ?
Why is it not? We don't want the userspace to communicate other things
to the host than what is configured in the configfs. If you only object
the explanation, then I will improve the commit message and send an
fixed v8. If you have more objections please share your doubts, thanks.
I'm also not really sure of the benefit; wouldn't this result in
userspace streaming data that's configured differently to what the host
is expecting?
static int
uvc_send_response(struct uvc_device *uvc, struct uvc_request_data
*data)
{
@@ -192,6 +253,21 @@ uvc_send_response(struct uvc_device *uvc,
struct uvc_request_data *data)
memcpy(req->buf, data->data, req->length);
+ /* validate the ctrl content and fixup */
+ if (!uvc->event_setup_out) {
+ struct uvc_streaming_control *ctrl = req->buf;
+
+ switch (uvc->streaming_request) {
+ case UVC_GET_CUR:
+ case UVC_GET_MIN:
+ case UVC_GET_MAX:
+ case UVC_GET_DEF:
+ uvc_validate_streaming_ctrl(uvc, ctrl);
+ default:
+ break;
+ }
+ }
+
What about read requests for controls that aren't for the streaming
interface?
return usb_ep_queue(cdev->gadget->ep0, req, GFP_KERNEL);
}
--
Regards,
Laurent Pinchart
