Re: Possible race in cpia.c?

videodev.c holds the Big Kernel Lock while calling the drivers' open()callbacks. So, this race cannot happen.

Note that with Gerd's V4L redesign, I'm fairly sure that this race *can*happen if the driver doesn't replace this check withvideo_exclusive_open(), or protect it with a semaphore. This is becausethe V4L layer doesn't have the open() wrapper function that it used to have.


Gerd, can you make sure that your cpia patch fixes this?

Joe Burks wrote:

I noticed the following code in cpia.c:

static int cpia_open(struct video_device *dev, int flags)
{
   int i;
   struct cam_data *cam = dev->priv;

   if (!cam) {
       DBG("Internal error, cam_data not found!\n");
       return -EBUSY;
   }

   if (cam->open_count > 0) {
       DBG("Camera already open\n");
       return -EBUSY;
   }
Is it possible (although extremely unlikely) that two processes could tryand open a device at the same time and both of them believe thatcam->open_count == 0 ? If so, given the unlikely nature of thispossibility, should it be a concern?



--
Mark McClelland
mmcclell@xxxxxxxxxxx