Re: [Intel-gfx] [PATCH] drm/i915: Sanity check incoming ioctl data for a NULL pointer

[Chris Wilson <chris@chris-wilson.co.uk>]

On Thu, Mar 14, 2013 at 09:50:04PM -0700, Ben Widawsky wrote:
> On Thu, Mar 14, 2013 at 12:59:57PM +0000, Chris Wilson wrote:
> > In order to prevent a potential NULL deference with hostile userspace,
> > we need to check whether the ioctl was passed an invalid args pointer.
> > 
> > Reported-by: Tommi Rantala <tt.rantala@gmail.com>
> > Link: http://lkml.kernel.org/r/CA+ydwtpuBvbwxbt-tdgPUvj1EU7itmCHo_2B3w13HkD5+jWKow@mail.gmail.com
> > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> > ---
> >  drivers/gpu/drm/i915/i915_gem_execbuffer.c |   11 +++++++++--
> >  1 file changed, 9 insertions(+), 2 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
> > index 365e41a..9f5602e 100644
> > --- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c
> > +++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
> > @@ -1103,7 +1103,11 @@ i915_gem_execbuffer(struct drm_device *dev, void *data,
> >  	struct drm_i915_gem_exec_object2 *exec2_list = NULL;
> >  	int ret, i;
> >  
> > -	if (args->buffer_count < 1) {
> > +	if (args == NULL)
> > +		return -EINVAL;
> > +
> > +	if (args->buffer_count < 1 ||
> > +	    args->buffer_count > INT_MAX / sizeof(*exec2_list)) {
> >  		DRM_DEBUG("execbuf with %d buffers\n", args->buffer_count);
> >  		return -EINVAL;
> >  	}
> > @@ -1182,8 +1186,11 @@ i915_gem_execbuffer2(struct drm_device *dev, void *data,
> >  	struct drm_i915_gem_exec_object2 *exec2_list = NULL;
> >  	int ret;
> >  
> > +	if (args == NULL)
> > +		return -EINVAL;
> > +
> >  	if (args->buffer_count < 1 ||
> > -	    args->buffer_count > UINT_MAX / sizeof(*exec2_list)) {
> > +	    args->buffer_count > INT_MAX / sizeof(*exec2_list)) {
> >  		DRM_DEBUG("execbuf2 with %d buffers\n", args->buffer_count);
> >  		return -EINVAL;
> >  	}
> 
> Why did you change UINT_MAX to INT_MAX?

Because we check later against INT_MAX, and I didn't like the confusion.
If we are going to pick an arbitrary limit, lets at least be consistent.

> TBH, I'm confused what we're
> trying to achieve, and why we need anything other than:
> if (!args->buffer_count)

Because we then promptly do a u32 multiply and we need to be sure that
userspace can't trigger an overflow there and cause us to read
unallocated memory later.
> 
> I'm also not seeing how the NULL checks are needed since at least it
> seems to be for execbuffer (IOW) we could never have NULL args.

That's what I thought too. Looking at the stack trace, the empirical
evidence is that we need the check.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre