Re: [PATCH v3 2/5] iommufd: Destroy vdevice on idevice destroy

[Xu Yilun <yilun.xu@linux.intel.com>]

On Mon, Jun 30, 2025 at 06:27:51AM +0000, Tian, Kevin wrote:
> > From: Xu Yilun <yilun.xu@linux.intel.com>
> > Sent: Friday, June 27, 2025 11:38 AM
> > 
> > +static void iommufd_device_remove_vdev(struct iommufd_device *idev)
> > +{
> > +	struct iommufd_vdevice *vdev;
> > +
> > +	mutex_lock(&idev->igroup->lock);
> > +	/* vdev has been completely destroyed by userspace */
> > +	if (!idev->vdev)
> > +		goto out_unlock;
> > +
> > +	vdev = iommufd_get_vdevice(idev->ictx, idev->vdev->obj.id);
> > +	if (IS_ERR(vdev)) {
> > +		/*
> > +		 * vdev is removed from xarray by userspace, but is not
> > +		 * destroyed/freed. Since iommufd_vdevice_abort() is
> > reentrant,
> > +		 * safe to destroy vdev here.
> > +		 */
> > +		iommufd_vdevice_abort(&idev->vdev->obj);
> > +		goto out_unlock;
> > +	}
> 
> let's add a comment that vdev is still freed in iommufd_destroy()
> in this situation.

Yes.

> 
> > -void iommufd_vdevice_destroy(struct iommufd_object *obj)
> > +void iommufd_vdevice_abort(struct iommufd_object *obj)
> >  {
> >  	struct iommufd_vdevice *vdev =
> >  		container_of(obj, struct iommufd_vdevice, obj);
> >  	struct iommufd_viommu *viommu = vdev->viommu;
> > +	struct iommufd_device *idev = vdev->idev;
> > +
> > +	lockdep_assert_held(&idev->igroup->lock);
> > +
> > +	/*
> > +	 * iommufd_vdevice_abort() could be reentrant, by
> > +	 * iommufd_device_unbind() or by iommufd_destroy(). Cleanup only
> > once.
> > +	 */
> > +	if (!viommu)
> > +		return;
> 
> Just check idev->vdev, to be consistent with the other path.

I think there is problem here. From your comments above, vdev could be
aborted/destroyed by iommufd_destroy() again *after* idev is freed.
That means in iommufd_vdevice_abort/destroy(), usage of vdev->idev or
idev->vdev or vdev->idev->igroup->lock may be invalid.

I need to reconsider this, seems we need a dedicated vdev lock to
synchronize concurrent vdev abort/destroy.

Thanks,
Yilun