Re: [Qemu-devel] [PATCH] virtio-balloon: stats vq migration spec compliance

[Michael S. Tsirkin]

On Tue, Aug 16, 2016 at 11:30:16AM +0100, Stefan Hajnoczi wrote:
> On Tue, Aug 16, 2016 at 12:32:55AM +0300, Michael S. Tsirkin wrote:
> > virtio spec says that devices must not touch VQs
> > before DRIVER_OK is set.
> > 
> > Additionally, balloon must not touch stats VQ
> > unless the stats feature bit has been negotiated.
> > 
> > Cc: Ladi Prosek <address@hidden>
> > Signed-off-by: Michael S. Tsirkin <address@hidden>
> > ---
> >  hw/virtio/virtio-balloon.c | 4 +++-
> >  1 file changed, 3 insertions(+), 1 deletion(-)
> > 
> > diff --git a/hw/virtio/virtio-balloon.c b/hw/virtio/virtio-balloon.c
> > index 65457e9..7189260 100644
> > --- a/hw/virtio/virtio-balloon.c
> > +++ b/hw/virtio/virtio-balloon.c
> > @@ -427,6 +427,7 @@ static void virtio_balloon_vmstate_cb(void *opaque, int 
> > running,
> >                                        RunState state)
> >  {
> >      VirtIOBalloon *s = opaque;
> > +    VirtIODevice *vdev = VIRTIO_DEVICE(s);
> >  
> >      if (!running) {
> >          /* put the stats element back if the VM is not running */
> > @@ -436,7 +437,8 @@ static void virtio_balloon_vmstate_cb(void *opaque, int 
> > running,
> >              s->stats_vq_elem = NULL;
> >          }
> >  
> > -    } else {
> > +    } else if (balloon_stats_supported(s) &&
> > +               (vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
> >          /* poll stats queue for the element we may have discarded
> >           * when the VM was stopped */
> >          virtio_balloon_receive_stats(VIRTIO_DEVICE(s), s->svq);
> 
> Another suspect case is when the feature bit was negotiated,
> VIRTIO_CONFIG_S_DRIVER_OK is set, but the virtqueue PFN wasn't set up.
> 
> In that case virtqueue_pop() will access junk values.
> 
> I'm not sure if anything terrible can happen but a check that
> vq->vring.desc != 0 would be nice - just like we do in
> virtio_queue_notify_vq().
> 
> Can you add a virtio_queue_get_addr() call to see if the PFN has been
> set?
> 
> Stefan

DRIVER_OK is set after VQs are initialized so I think that's enough.