Linux Virtualization - Aug 2023 - [PATCH net] virtio-net: set queues after driver_ok

On Wed, Aug 9, 2023 at 2:23?PM Dragos Tatulea <dtatulea at nvidia.com>
wrote:
>
> On Tue, 2023-08-08 at 23:13 -0400, Jason Wang wrote:
> > Commit 25266128fe16 ("virtio-net: fix race between set queues and
> > probe") tries to fix the race between set queues and probe by
calling
> > _virtnet_set_queues() before DRIVER_OK is set. This violates virtio
> > spec. Fixing this by setting queues after virtio_device_ready().
> >
> > Fixes: 25266128fe16 ("virtio-net: fix race between set queues and
probe")
> > Reported-by: Dragos Tatulea <dtatulea at nvidia.com>
> > Signed-off-by: Jason Wang <jasowang at redhat.com>
> > ---
> > The patch is needed for -stable.
> > ---
> >  drivers/net/virtio_net.c | 4 ++--
> >  1 file changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> > index 1270c8d23463..ff03921e46df 100644
> > --- a/drivers/net/virtio_net.c
> > +++ b/drivers/net/virtio_net.c
> > @@ -4219,8 +4219,6 @@ static int virtnet_probe(struct virtio_device
*vdev)
> >         if (vi->has_rss || vi->has_rss_hash_report)
> >                 virtnet_init_default_rss(vi);
> >
> > -       _virtnet_set_queues(vi, vi->curr_queue_pairs);
> > -
> >         /* serialize netdev register + virtio_device_ready() with
ndo_open()
> > */
> >         rtnl_lock();
> >
> > @@ -4233,6 +4231,8 @@ static int virtnet_probe(struct virtio_device
*vdev)
> >
> >         virtio_device_ready(vdev);
> >
> > +       _virtnet_set_queues(vi, vi->curr_queue_pairs);
> > +
> >         /* a random MAC address has been assigned, notify the device.
> >          * We don't fail probe if VIRTIO_NET_F_CTRL_MAC_ADDR is
not there
> >          * because many devices work fine without getting MAC
explicitly
>
> Thanks for the quick fix. Doesn't this fix though bring back the
original race
> that was fixed in commit 25266128fe16 ("virtio-net: fix race between
set queues
> and probe")? Or is being under the same rntl_lock session as
register_netdev
> enough to avoid the race?
Yes, rtnl needs to be held for userspace requests to change the number
of queues. So we are serialized in this way.

Thanks
>
> Thanks,
> Dragos

On Wed, Aug 09, 2023 at 02:45:17PM +0800, Jason Wang
wrote:
> On Wed, Aug 9, 2023 at 2:23?PM Dragos Tatulea <dtatulea at
nvidia.com> wrote:
> >
> > On Tue, 2023-08-08 at 23:13 -0400, Jason Wang wrote:
> > > Commit 25266128fe16 ("virtio-net: fix race between set
queues and
> > > probe") tries to fix the race between set queues and probe
by calling
> > > _virtnet_set_queues() before DRIVER_OK is set. This violates
virtio
> > > spec. Fixing this by setting queues after virtio_device_ready().
> > >
> > > Fixes: 25266128fe16 ("virtio-net: fix race between set
queues and probe")
> > > Reported-by: Dragos Tatulea <dtatulea at nvidia.com>
> > > Signed-off-by: Jason Wang <jasowang at redhat.com>
> > > ---
> > > The patch is needed for -stable.
> > > ---
> > >  drivers/net/virtio_net.c | 4 ++--
> > >  1 file changed, 2 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> > > index 1270c8d23463..ff03921e46df 100644
> > > --- a/drivers/net/virtio_net.c
> > > +++ b/drivers/net/virtio_net.c
> > > @@ -4219,8 +4219,6 @@ static int virtnet_probe(struct
virtio_device *vdev)
> > >         if (vi->has_rss || vi->has_rss_hash_report)
> > >                 virtnet_init_default_rss(vi);
> > >
> > > -       _virtnet_set_queues(vi, vi->curr_queue_pairs);
> > > -
> > >         /* serialize netdev register + virtio_device_ready() with
ndo_open()
> > > */
> > >         rtnl_lock();
> > >
> > > @@ -4233,6 +4231,8 @@ static int virtnet_probe(struct
virtio_device *vdev)
> > >
> > >         virtio_device_ready(vdev);
> > >
> > > +       _virtnet_set_queues(vi, vi->curr_queue_pairs);
> > > +
> > >         /* a random MAC address has been assigned, notify the
device.
> > >          * We don't fail probe if VIRTIO_NET_F_CTRL_MAC_ADDR
is not there
> > >          * because many devices work fine without getting MAC
explicitly
> >
> > Thanks for the quick fix. Doesn't this fix though bring back the
original race
> > that was fixed in commit 25266128fe16 ("virtio-net: fix race
between set queues
> > and probe")? Or is being under the same rntl_lock session as
register_netdev
> > enough to avoid the race?
> 
> Yes, rtnl needs to be held for userspace requests to change the number
> of queues. So we are serialized in this way.
> 
> Thanks
maybe post v2 adding this in the commit log.

> >
> > Thanks,
> > Dragos