On Saturday 08 August 2009 11:07:22 Eduardo Valentin wrote:
> On Fri, Aug 07, 2009 at 01:51:36PM +0200, ext Hans Verkuil wrote:
> > Hi Eduardo,
> >
> > Douglas pointed out to me that I hadn't reviewed this series yet.
> >
> > That was mostly because it's pretty good as far as I'm concerned :-)
> >
> > I do think one small thing should change:
> >
> > On Monday 27 July 2009 17:12:08 Eduardo Valentin wrote:
> > > diff --git a/linux/drivers/media/radio/si4713-i2c.c b/linux/drivers/media/radio/si4713-i2c.c
> >
> > <snip>
> >
> > > +/* write string property */
> > > +static int si4713_write_econtrol_string(struct si4713_device *sdev,
> > > + struct v4l2_ext_control *control)
> > > +{
> > > + struct v4l2_queryctrl vqc;
> > > + int len;
> > > + s32 rval = 0;
> > > +
> > > + vqc.id = control->id;
> > > + rval = si4713_queryctrl(&sdev->sd, &vqc);
> > > + if (rval < 0)
> > > + goto exit;
> > > +
> > > + switch (control->id) {
> > > + case V4L2_CID_RDS_TX_PS_NAME: {
> > > + char ps_name[MAX_RDS_PS_NAME + 1];
> > > +
> > > + len = control->size - 1;
> > > + if (len > MAX_RDS_PS_NAME)
> > > + len = MAX_RDS_PS_NAME;
> > > + rval = copy_from_user(ps_name, control->string, len);
> > > + if (rval < 0)
> > > + goto exit;
> > > + ps_name[len] = '\0';
> > > +
> > > + if (strlen(ps_name) % vqc.step) {
> > > + rval = -EINVAL;
> >
> > I think we should return -ERANGE instead. It makes more sense than -EINVAL,
> > since it is the string length that is out of bounds. -ERANGE is the expected
> > error code when the control boundary checks fail.
> >
> > I know I said EINVAL before, but after thinking about it some more I've
> > reconsidered.
>
>
> Hans, just to make sure in this point. Here I am returning EINVAL because
> the string length is not multiple of control step and not because it is greater
> than it is supposed to be. As discussed earlier, during control write, if
> the coming string value is larger than it was supposed to be, I'm shrinking it.
> See above lines. So the question is: should I return ERANGE when string length
> is not multiple of step?
The V4L2 spec says that the device driver author has a choice: either modify
the control's value to fit the min/max and/or step values, or return -ERANGE.
In this case you can decide to either reduce the string length to the first
valid length or return -ERANGE. If you reduce the length, then you must update
the control->string as well! (A copy_to_user() of the new terminating 0).
To be honest, in this case I think you should just return -ERANGE and not
attempt to modify the string. I don't really see a good use-case for that.
Cutting off a string is a rather unexpected side-effect IMHO.
Regards,
Hans
--
Hans Verkuil - video4linux developer - sponsored by TANDBERG Telecom
--
To unsubscribe from this list: send the line "unsubscribe linux-media" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
