Re: [PATCH] drm: do not use device name as a format string

[Nicolas Iooss <nicolas.iooss_linux-oWGTIYur0i8@public.gmane.org>]

On 12/07/2015 01:31 PM, Thierry Reding wrote:
> On Mon, Dec 07, 2015 at 12:46:52PM +0100, Daniel Vetter wrote:
>> On Mon, Dec 07, 2015 at 11:53:01AM +0200, Jani Nikula wrote:
>>> On Mon, 07 Dec 2015, Daniel Vetter <daniel-/w4YWyX8dFk@public.gmane.org> wrote:
>>>> On Sun, Dec 06, 2015 at 11:16:32AM +0100, Nicolas Iooss wrote:
>>>>> On 12/06/2015 10:35 AM, Daniel Vetter wrote:
>>>>>>> On 11/18/2015 06:58 PM, Nicolas Iooss wrote:
>>>>>>>> drm_dev_set_unique() formats its parameter using kvasprintf() but many
>>>>>>>> of its callers directly pass dev_name(dev) as printf format string,
>>>>>>>> without any format parameter.  This can cause some issues when the
>>>>>>>> device name contains '%' characters.
>>>>>>>>
>>>>>>>> To avoid any potential issue, always use "%s" when using
>>>>>>>> drm_dev_set_unique() with dev_name().
>>>>>>
>>>>>> Not sure this is worth it really, normally people don't place % characters
>>>>>> into their device names, ever. And if they do it'll blow up. There's also
>>>>>> no security issue here since userspace can't set this name.
>>>>>>
>>>>>> If the maintainers of the affected drivers don't want this I won't merge
>>>>>> this patch.
>>>>>
>>>>> Actually I had the same opinion before I began to add __printf
>>>>> attributes and "%s" in several places in the kernel to make
>>>>> -Wformat-security useful.  This led me to discover some funny issues
>>>>> like the one fixed by commit 3958b79266b1 ("configfs: fix kernel
>>>>> infoleak through user-controlled format string",
>>>>> https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=3958b79266b14729edd61daf9dfb84de45f4ec6d
>>>>> ).  The patch I sent is in fact a very small step towards making
>>>>> -Wformat-security useful again to detect "real" issues.
>>>>>
>>>>> Of course, if you do not feel it is worth it and believe that dev_name
>>>>> is fully controlled by trusted sources which will never introduce any %
>>>>> character, I understand your will of not merging my patch.
>>>>
>>>> Ah, that's the context I was missing, that really should be in the commit
>>>> message. If this is part of an overall effort to enable something useful
>>>> it makes more sense to get it in.
>>>>
>>>> On the patch itself it feels rather funny to do a "%s", str); combo, maybe
>>>> we should have a drm_dev_set_unique_static instead? Including kerneldoc
>>>> explaining why there's too.
>>>
>>> No caller of drm_dev_set_unique() actually uses the formatting for
>>> anything... so you'd end up with drm_dev_set_unique_static() and an
>>> orphaned drm_dev_set_unique()...
>>
>> Ok, then I guess we can just ditch the printf stuff from set_unique.
>> Nicolas, you're up for that?
> 
> Looking at all the callsites of drm_dev_set_unique() it seems like all
> of the drivers (with the exception of vgem) use dev_name() on the same
> device that's already passed into drm_dev_alloc(), so perhaps another
> alternative would be to have drm_dev_alloc() set the unique name by
> default and keep the function for cases where it needs to be set
> explicitly (like for vgem). vgem passes drm_dev_alloc() a NULL device,
> so that could serve as condition.

I have written a patch which removes the printf format processing from
drm_dev_set_unique().  I will test it as soon as possible and depending
on the results, send it or explain what went wrong.  If no one does the
work before me, I'll also take a look at calling drm_dev_set_unique() in
drm_dev_alloc(), and this would be an other patch.

Thanks,
Nicolas