[PATCH v2] ioctl: fix ioctl command number decoding in case of conflicts

[Patrik Jakobsson]

On Mon, Nov 30, 2015 at 4:59 PM, Gabriel Laskar <gabriel at lse.epita.fr> wrote:
> On Mon, 30 Nov 2015 16:21:53 +0100
> Patrik Jakobsson <patrik.r.jakobsson at gmail.com> wrote:
>
>> On Mon, Nov 30, 2015 at 4:13 PM, Dmitry V. Levin <ldv at altlinux.org> wrote:
>> > On Mon, Nov 30, 2015 at 02:49:11PM +0100, Patrik Jakobsson wrote:
>> >> On Thu, Sep 24, 2015 at 2:45 AM, Dmitry V. Levin <ldv at altlinux.org> wrote:
>> >> > On Wed, Sep 23, 2015 at 10:11:55AM +0200, Gabriel Laskar wrote:
>> >> >> When a command number was decoded through ioctl_decode_command_number(),
>> >> >> there was no check for conflicts with other potential ioctls numbers.
>> >> >
>> >> > Merged, thanks.
>> >>
>> >> Hi, sorry for noticing this a bit late but this change makes it
>> >> impossible to override the ioctl decoding for drm. The correct
>> >> decoding can be done by identifying the driver and I need a way to
>> >> stop the generic decoding at that point.
>> >
>> > Do you mean that you want to be able to turn off ioctl_lookup completely?
>>
>> Yes that was my idea. ioctl_lookup() doesn't know about the drm driver
>> backing the device so it cannot give us the correct ioctl.
>
> I am working on a semi-automated generator for all ioctl decoding, and
> I have the same kind of issues.
>
> Identifying the correct device is a problem that is recurring in all
> the ioctl decoding code. There is a some devices that have conflicting
> numbers, having a way to know exactly which device it is, like gather
> major/minor number, would allow to have the right set of ioctl. As it
> is costly to do that, we can, for example enable this only when verbose
> mode is activated, and fallback on just displaying the ioctl number on
> the other cases.

Interesting, but major/minor will not help in the DRM case. I had to
look at sysfs to figure out what driver is backing the device. Each
driver has its own set of private ioctls that overlap
(DRM_COMMAND_BASE to DRM_COMMAND_END). See 2/5 and 3/5 in my i915/drm
patch series.

-Patrik

>
> I am preparing an rfc for all that.
>
> --
> Gabriel Laskar