Re: [PATCH 5.10] gpu/i915: fix error return in mmap_offset_attach()

["Ville Syrjälä" <ville.syrjala@linux.intel.com>]

On Thu, Oct 02, 2025 at 06:20:54PM +0300, Ville Syrjälä wrote:
> On Thu, Oct 02, 2025 at 09:55:58AM -0400, Rodrigo Vivi wrote:
> > On Thu, Oct 02, 2025 at 11:48:26AM +0300, Alexandr Sapozhnkiov wrote:
> > > From: Alexandr Sapozhnikov <alsp705@gmail.com>
> > 
> > About the subject, this is not just a 5.10 kernel issue.
> > This code is the current code in our tip.
> > So this needs to target drm-tip branch, and then Cc stable
> > and perhaps a Fixes: tag.
> > 
> > > 
> > > In the drm_vma_node_allow function, kmalloc may 
> > > return NULL, in which case the file element will not be 
> > > added to the mmo->vma_node list. It would be good to 
> > > not ignore this event, but at least log an error message.
> > > 
> > > Found by Linux Verification Center (linuxtesting.org) with SVACE.
> > > 
> > > Signed-off-by: Alexandr Sapozhnikov <alsp705@gmail.com>
> > > ---
> > >  drivers/gpu/drm/i915/gem/i915_gem_mman.c | 7 +++++--
> > >  1 file changed, 5 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/drivers/gpu/drm/i915/gem/i915_gem_mman.c b/drivers/gpu/drm/i915/gem/i915_gem_mman.c
> > > index a2195e28b625..adaef8f09d59 100644
> > > --- a/drivers/gpu/drm/i915/gem/i915_gem_mman.c
> > > +++ b/drivers/gpu/drm/i915/gem/i915_gem_mman.c
> > > @@ -706,8 +706,11 @@ mmap_offset_attach(struct drm_i915_gem_object *obj,
> > >  	mmo = insert_mmo(obj, mmo);
> > >  	GEM_BUG_ON(lookup_mmo(obj, mmap_type) != mmo);
> > >  out:
> > > -	if (file)
> > > -		drm_vma_node_allow_once(&mmo->vma_node, file);
> > > +	if (file) {
> > > +		err = drm_vma_node_allow_once(&mmo->vma_node, file);
> > > +		if (err)
> > 
> > perhaps we also need to drm_vma_offset_remove here?
> > I mean... honest question, doubt here. Is there any further clean-up needed?
> 
> Yeah, mmo->vma_node has already been linked to varius places here.
> So this will lead to use-after-free in short order.
> 
> With the current code if this fails then I think all that ends up
> happening is that subsequent mmap() will fail. Maybe that's just
> fine?

Hmm, or maybe it would be better to just directly return an error
here without freeing/cleaning up anything. Looks to me like that
should work fine as well, and userspace would get the error
immediately instead of later.

> 
> > 
> > > +			goto err;
> > > +	}
> > >  	return mmo;
> > >  
> > >  err:
> > > -- 
> > > 2.43.0
> > > 
> 
> -- 
> Ville Syrjälä
> Intel

-- 
Ville Syrjälä
Intel