[Qemu-devel] Re: [PATCH] net: delay freeing peer host device

["Michael S. Tsirkin" <mst@redhat.com>]

On Mon, Sep 27, 2010 at 02:18:37PM -0600, Alex Williamson wrote:
> On Mon, 2010-09-27 at 15:00 +0200, Michael S. Tsirkin wrote:
> > With -netdev, virtio devices present offload
> > features to guest, depending on the backend used.
> > Thus, removing host netdev peer while guest is
> > active leads to guest-visible inconsistency and/or crashes.
> > 
> > As a solution, while guest (NIC) peer device exists,
> > we prevent the host peer from being deleted.
> > This patch does this by adding peer_deleted flag in nic state:
> > if host device is going away while guest device
> > is around, set this flag and keep a shell of
> > the host device around for as long as guest device exists.
> > 
> > The link is put down so all packets will get discarded.
> > 
> > At the moment, management can detect that device deletion
> > is delayed by doing info net. As a next step, we shall add
> > commands that control hotplug/unplug without
> > removing the device, and an event to report that
> > guest has responded to the hotplug event.
> > 
> > Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
> > ---
> >  net.c |   40 +++++++++++++++++++++++++++++++++-------
> >  net.h |    1 +
> >  2 files changed, 34 insertions(+), 7 deletions(-)
> > 
> > diff --git a/net.c b/net.c
> > index 3d0fde7..edb87af 100644
> > --- a/net.c
> > +++ b/net.c
> > @@ -281,29 +281,55 @@ NICState *qemu_new_nic(NetClientInfo *info,
> >      return nic;
> >  }
> >  
> > -void qemu_del_vlan_client(VLANClientState *vc)
> > +static void qemu_free_vlan_client(VLANClientState *vc)
> >  {
> >      if (vc->vlan) {
> >          QTAILQ_REMOVE(&vc->vlan->clients, vc, next);
> >      } else {
> > +        QTAILQ_REMOVE(&non_vlan_clients, vc, next);
> >          if (vc->send_queue) {
> >              qemu_del_net_queue(vc->send_queue);
> >          }
> > -        QTAILQ_REMOVE(&non_vlan_clients, vc, next);
> >          if (vc->peer) {
> >              vc->peer->peer = NULL;
> >          }
> >      }
> > -
> > -    if (vc->info->cleanup) {
> > -        vc->info->cleanup(vc);
> > -    }
> > -
> >      qemu_free(vc->name);
> >      qemu_free(vc->model);
> >      qemu_free(vc);
> >  }
> >  
> > +void qemu_del_vlan_client(VLANClientState *vc)
> > +{
> > +    /* If there is a peer NIC, delete and cleanup client, but do not free. */
> > +    if (!vc->vlan && vc->peer && vc->peer->info->type == NET_CLIENT_TYPE_NIC) {
> > +        NICState *nic = DO_UPCAST(NICState, nc, vc->peer);
> > +        if (nic->peer_deleted) {
> > +            return;
> > +        }
> > +        nic->peer_deleted = true;
> > +        /* Let NIC know peer is gone. */
> > +        vc->peer->link_down = true;
> > +        if (vc->peer->info->link_status_changed) {
> > +            vc->peer->info->link_status_changed(vc->peer);
> 
> Are we potentially introducing more races here?  Now we have the tap
> trying to call link change on a peer device that's potentially in the
> middle of going away.  Thanks,
> 
> Alex

You mean in guest?
This could always happen anyway: user can call link state change
command. So drivers must be ready for this.
There's no race in qemu because everything is singlethreaded.

> > +        }
> > +        if (vc->info->cleanup) {
> > +            vc->info->cleanup(vc);
> > +        }
> > +        return;
> > +    }
> > +
> > +    /* If this is a peer NIC and peer has already been deleted, free it now. */
> > +    if (!vc->vlan && vc->peer && vc->info->type == NET_CLIENT_TYPE_NIC) {
> > +        NICState *nic = DO_UPCAST(NICState, nc, vc);
> > +        if (nic->peer_deleted) {
> > +            qemu_free_vlan_client(vc->peer);
> > +        }
> > +    }
> > +
> > +    qemu_free_vlan_client(vc);
> > +}
> > +
> >  VLANClientState *
> >  qemu_find_vlan_client_by_name(Monitor *mon, int vlan_id,
> >                                const char *client_str)
> > diff --git a/net.h b/net.h
> > index 518cf9c..44c31a9 100644
> > --- a/net.h
> > +++ b/net.h
> > @@ -72,6 +72,7 @@ typedef struct NICState {
> >      VLANClientState nc;
> >      NICConf *conf;
> >      void *opaque;
> > +    bool peer_deleted;
> >  } NICState;
> >  
> >  struct VLANState {
> 
>