Re: [PATCH 3/5] vfio: Don't leak a group reference if the group already exists

[Jason Gunthorpe <jgg@nvidia.com>]

On Mon, Oct 04, 2021 at 04:25:43PM -0600, Alex Williamson wrote:
> On Fri,  1 Oct 2021 20:22:22 -0300
> Jason Gunthorpe <jgg@nvidia.com> wrote:
> 
> > If vfio_create_group() searches the group list and returns an already
> > existing group it does not put back the iommu_group reference that the
> > caller passed in.
> > 
> > Change the semantic of vfio_create_group() to not move the reference in
> > from the caller, but instead obtain a new reference inside and leave the
> > caller's reference alone. The two callers must now call iommu_group_put().
> > 
> > This is an unlikely race as the only caller that could hit it has already
> > searched the group list before attempting to create the group.
> > 
> > Fixes: cba3345cc494 ("vfio: VFIO core")
> > Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
> >  drivers/vfio/vfio.c | 18 +++++++++---------
> >  1 file changed, 9 insertions(+), 9 deletions(-)
> > 
> > diff --git a/drivers/vfio/vfio.c b/drivers/vfio/vfio.c
> > index 1cb12033b02240..bf233943dc992f 100644
> > +++ b/drivers/vfio/vfio.c
> > @@ -338,6 +338,7 @@ static void vfio_group_unlock_and_free(struct vfio_group *group)
> >  		list_del(&unbound->unbound_next);
> >  		kfree(unbound);
> >  	}
> > +	iommu_group_put(group->iommu_group);
> >  	kfree(group);
> >  }
> >  
> > @@ -389,6 +390,8 @@ static struct vfio_group *vfio_create_group(struct iommu_group *iommu_group,
> >  	atomic_set(&group->opened, 0);
> >  	init_waitqueue_head(&group->container_q);
> >  	group->iommu_group = iommu_group;
> > +	/* put in vfio_group_unlock_and_free() */
> > +	iommu_group_ref_get(iommu_group);

      ^^^^^^^^^^^^^^^^^

> >  	group->type = type;
> >  	BLOCKING_INIT_NOTIFIER_HEAD(&group->notifier);
> >  
> > @@ -396,8 +399,8 @@ static struct vfio_group *vfio_create_group(struct iommu_group *iommu_group,
> >  
> >  	ret = iommu_group_register_notifier(iommu_group, &group->nb);
> >  	if (ret) {
> > -		kfree(group);
> > -		return ERR_PTR(ret);
> > +		group = ERR_PTR(ret);
> > +		goto err_put_group;
> >  	}
> >  
> >  	mutex_lock(&vfio.group_lock);
> > @@ -432,6 +435,9 @@ static struct vfio_group *vfio_create_group(struct iommu_group *iommu_group,
> >  
> >  	mutex_unlock(&vfio.group_lock);
> >  
> > +err_put_group:
> > +	iommu_group_put(iommu_group);
> > +	kfree(group);
> 
> ????
> 
> In the non-error path we're releasing the caller's reference which is
> now their responsibility to release,

This release is paried with the get in the same function added one
hunk above

> but in any case we're freeing the object that we return?  That
> can't be right.

Yes, that is a rebasing mistake pulling this back from the last patch
that had a "return ret" here, thanks

> > @@ -776,10 +780,6 @@ static struct vfio_group *vfio_group_find_or_alloc(struct device *dev)
> >  
> >  	/* a newly created vfio_group keeps the reference. */
> 
> This comment is now incorrect.  Thanks,

Indeed

Jason