[PATCH] drm/vgem: Close use-after-free race in vgem_gem_create

[Daniel Vetter]

On Sun, Feb 02, 2020 at 05:37:31PM +0000, Chris Wilson wrote:
> Quoting Daniel Vetter (2020-02-02 13:21:33)
> > There's two references floating around here (for the object reference,
> > not the handle_count reference, that's a different thing):
> > 
> > - The temporary reference held by vgem_gem_create, acquired by
> >   creating the object and released by calling
> >   drm_gem_object_put_unlocked.
> > 
> > - The reference held by the object handle, created by
> >   drm_gem_handle_create. This one generally outlives the function,
> >   except if a 2nd thread races with a GEM_CLOSE ioctl call.
> > 
> > So usually everything is correct, except in that race case, where the
> > access to gem_object->size could be looking at freed data already.
> > Which again isn't a real problem (userspace shot its feet off already
> > with the race, we could return garbage), but maybe someone can exploit
> > this as an information leak.
> > 
> > Cc: Dan Carpenter <dan.carpenter at oracle.com>
> > Cc: Hillf Danton <hdanton at sina.com>
> > Cc: Reported-by: syzbot+0dc4444774d419e916c8 at syzkaller.appspotmail.com
> > Cc: stable at vger.kernel.org
> > Cc: Emil Velikov <emil.velikov at collabora.com>
> > Cc: Daniel Vetter <daniel.vetter at ffwll.ch>
> > Cc: Sean Paul <seanpaul at chromium.org>
> > Cc: Chris Wilson <chris at chris-wilson.co.uk>
> > Cc: Eric Anholt <eric at anholt.net>
> > Cc: Sam Ravnborg <sam at ravnborg.org>
> > Cc: Rob Clark <robdclark at chromium.org>
> > Signed-off-by: Daniel Vetter <daniel.vetter at intel.com>
> > ---
> >  drivers/gpu/drm/vgem/vgem_drv.c | 9 ++++++---
> >  1 file changed, 6 insertions(+), 3 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/vgem/vgem_drv.c b/drivers/gpu/drm/vgem/vgem_drv.c
> > index 5bd60ded3d81..909eba43664a 100644
> > --- a/drivers/gpu/drm/vgem/vgem_drv.c
> > +++ b/drivers/gpu/drm/vgem/vgem_drv.c
> > @@ -196,9 +196,10 @@ static struct drm_gem_object *vgem_gem_create(struct drm_device *dev,
> >                 return ERR_CAST(obj);
> >  
> >         ret = drm_gem_handle_create(file, &obj->base, handle);
> > -       drm_gem_object_put_unlocked(&obj->base);
> > -       if (ret)
> > +       if (ret) {
> > +               drm_gem_object_put_unlocked(&obj->base);
> >                 return ERR_PTR(ret);
> > +       }
> >  
> >         return &obj->base;
> >  }
> > @@ -221,7 +222,9 @@ static int vgem_gem_dumb_create(struct drm_file *file, struct drm_device *dev,
> >         args->size = gem_object->size;
> >         args->pitch = pitch;
> >  
> > -       DRM_DEBUG("Created object of size %lld\n", size);
> > +       drm_gem_object_put_unlocked(gem_object);
> > +
> > +       DRM_DEBUG("Created object of size %llu\n", args->size);
> 
> I was thinking we either should return size from vgem_gem_create (the
> strategy we took in i915) or simply remove the vgem_gem_create() as that
> doesn't improve readability.
> 
> -static struct drm_gem_object *vgem_gem_create(struct drm_device *dev,
> -                                             struct drm_file *file,
> -                                             unsigned int *handle,
> -                                             unsigned long size)
> +static int vgem_gem_dumb_create(struct drm_file *file, struct drm_device *dev,
> +                               struct drm_mode_create_dumb *args)
>  {
>         struct drm_vgem_gem_object *obj;
> -       int ret;
> +       u64 pitch, size;
> +       u32 handle;
> +
> +       pitch = args->width * DIV_ROUND_UP(args->bpp, 8);
> +       size = mul_u32_u32(args->height, pitch);
> +       if (size == 0 || pitch < args->width)
> +               return -EINVAL;
> 
>         obj = __vgem_gem_create(dev, size);
>         if (IS_ERR(obj))
> -               return ERR_CAST(obj);
> +               return PTR_ERR(obj);
> +
> +       size = obj->base.size;
> 
> -       ret = drm_gem_handle_create(file, &obj->base, handle);
> +       ret = drm_gem_handle_create(file, &obj->base, &handle);
>         drm_gem_object_put_unlocked(&obj->base);
>         if (ret)
>                 return ERR_PTR(ret);
> 
> -       return &obj->base;
> -}
> -
> -static int vgem_gem_dumb_create(struct drm_file *file, struct drm_device *dev,
> -                               struct drm_mode_create_dumb *args)
> -{
> -       struct drm_gem_object *gem_object;
> -       u64 pitch, size;
> -
> -       pitch = args->width * DIV_ROUND_UP(args->bpp, 8);
> -       size = args->height * pitch;
> -       if (size == 0)
> -               return -EINVAL;
> -
> -       gem_object = vgem_gem_create(dev, file, &args->handle, size);
> -       if (IS_ERR(gem_object))
> -               return PTR_ERR(gem_object);
> -
> -       args->size = gem_object->size;
> +       args->size = size;
>         args->pitch = pitch;
> +       args->handle = handle;
> 
> 
> At the end of the day, it makes no difference,

Yeah there's room for more polish, but didn't want to do that in the cc:
stable patch.

> Reviewed-by: Chris Wilson <chris at chris-wilson.co.uk>

Thanks for your review, finally applied to drm-misc-next-fixes now that CI
has blessed me with its attention for a bit!
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch