Re: [Qemu-devel] [RFC] net: Peer with existing NIC in netdev_add

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

On Thu, Nov 01, 2012 at 10:53:52AM +0100, Stefan Hajnoczi wrote:
> On Wed, Oct 31, 2012 at 06:34:07PM +0200, Michael S. Tsirkin wrote:
> > On Wed, Oct 31, 2012 at 03:51:08PM +0100, Stefan Hajnoczi wrote:
> > > On Wed, Oct 31, 2012 at 10:57:24AM +0200, Michael S. Tsirkin wrote:
> > > > On Wed, Oct 31, 2012 at 09:07:27AM +0100, Stefan Hajnoczi wrote:
> > > > > On Tue, Oct 30, 2012 at 05:24:06PM +0200, Michael S. Tsirkin wrote:
> > > > > > On Wed, Oct 24, 2012 at 02:49:21PM +0200, Stefan Hajnoczi wrote:
> > > > > > > Allow netdev_del followed by netdev_add to re-peer a NIC and its netdev:
> > > > > > > 
> > > > > > >   (qemu) info network
> > > > > > >   virtio-net-pci.0: type=nic,model=virtio-net-pci,macaddr=52:54:00:12:34:56
> > > > > > >    \ netdev0: type=user,net=10.0.2.0,restrict=off
> > > > > > > 
> > > > > > >   (qemu) netdev_del netdev0
> > > > > > > 
> > > > > > >   (qemu) netdev_add socket,id=netdev0,listen=:1234
> > > > > > > 
> > > > > > >   (qemu) info network
> > > > > > >   virtio-net-pci.0: type=nic,model=virtio-net-pci,macaddr=52:54:00:12:34:56
> > > > > > >    \ netdev0: type=socket,
> > > > > > > 
> > > > > > > This makes it possible to switch netdev while the guest is running.  It
> > > > > > > is not necessary to reset the NIC.
> > > > > > > 
> > > > > > > Note that the NIC's link goes down in netdev_del and back up again in
> > > > > > > netdev_add.  Therefore the guest becomes aware that the network has
> > > > > > > changed, although this depends on the emulated NIC model providing link
> > > > > > > status change interrupts.
> > > > > > > 
> > > > > > > Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
> > > > > > 
> > > > > > I'd be surprised if this patch worked when one or both backends are tap.
> > > > > > tap supports offloads but slirp doesn't, since guest
> > > > > > probes offloads at startup, it assumes it can use offloads.
> > > > > > We also program tap during device operation e.g. on set features.
> > > > > > vhost operation could also be interesting, have not looked into it.
> > > > > 
> > > > > Yes, I left a TODO in the RFC patch and described the issue below.
> > > > > We'll have to reject incompatible netdevs.
> > > > 
> > > > Ideally, we'd probe all backend capabilities at init time.
> > > > However, looks like we allowed netdev and device creation in any order.
> > > > Can we change this and require netdev always be there before device?
> > > 
> > > I don't think the order is a problem.  The relaxed order is only
> > > relevant during startup from main() - but in that case we have no
> > > constraints yet anyway.
> > > The problem only occurs when netdev_add is used to create an
> > > incompatible netdev after devices have initialized.  We should be able
> > > to check and error out in the code that my RFC patch modifies.  If
> > > constraints are violated then netdev_add can fail with an error (the new
> > > netdev is not created and the QMP client needs to try again with a
> > > compatible netdev configuration).
> > > 
> > > Maybe I'm misunderstanding your point?
> > > 
> > > Stefan
> > 
> > OK so if we basically require same type backend then I think it's mostly
> > fine.  I was trying to think of a way to allow changing backend type,
> > this becomes messy very quickly.  In partuclar macvtap probably
> > shouldn't be swapped with tap even though they are the same type
> > formally.
> 
> As long as they are offload-compatible, I think they can be swapped.
> It's up to the user or the management stack to make sure switching
> netdevs makes "sense".  So the network may be different and the guest
> needs to DHCP again, but that's the user's problem.

I think a simple rule like "use same backend type" is better than
an opaque one "are offload-compatible" - user has no idea
which offloads do each of the frontends and backends support.
Also if in future we add offloads to backend X suddenly we
break ability to swap with backend Y.
Let's keep it simple.

> Is there a reason why macvtap <-> tap won't work given compatible
> vnet_hdr offload?
> 
> Stefan

There's no guarantee they will support same offload options in all
kernels, in fact thats' not the case in some kernel.org kernels.


-- 
MST