Re: xen vs bridges

[Manuel Bouyer <bouyer%antioche.eu.org@localhost>]

On Sat, Jun 21, 2008 at 10:49:00AM -0400, der Mouse wrote:
> >>> Nothing in the xennet protocol prevents us from using MTU up to 2k.
> >>> The linux frontends and backends limit it to 1500 though, so we'd
> >>> need a way to know what MTU backends support.
> 
> Would we?  What happens if we send more than they're expecting?  Does

I guess it'd just drop the packet. 


> it work the way it does if you connect a vlan-capable interface to a
> non-capable one with physical interfaces?  That's what I'd expect it to
> do, on the "xennet and xvif are just a crossed cable between two
> interfaces" principle.

Yes, maybe I was too conservative here.

> 
> >> For my purposes, I just want it to work with NetBSD and I don't care
> >> about compatability with other domU OSes.  Will it work to just
> >> patch the xennet and xvif drivers to give them ETHERCAP_VLAN_MTU,
> > Yes, it should work
> 
> And it does!
> 
> --- /sys/arch/xen/xen/if_xennet_xenbus.c=     2006-11-02 22:03:32.000000000 
> -0500
> +++ /sys/arch/xen/xen/if_xennet_xenbus.c      2008-06-21 10:32:21.000000000 
> -0400
> @@ -322,6 +322,7 @@
>           ether_sprintf(sc->sc_enaddr));
>       /* Initialize ifnet structure and attach interface */
>       memcpy(ifp->if_xname, sc->sc_dev.dv_xname, IFNAMSIZ);
> +     ((struct ethercom *)ifp)->ec_capabilities = ETHERCAP_VLAN_MTU;
>       ifp->if_softc = sc;
>       ifp->if_start = xennet_start;
>       ifp->if_ioctl = xennet_ioctl;
> 
> --- /sys/arch/xen/xen/xennetback_xenbus.c=    2006-10-15 09:34:17.000000000 
> -0400
> +++ /sys/arch/xen/xen/xennetback_xenbus.c     2008-06-21 10:27:21.000000000 
> -0400
> @@ -310,6 +310,7 @@
>           (int)domid, (int)handle);
>       printf("%s: Ethernet address %s\n", ifp->if_xname,
>           ether_sprintf(xneti->xni_enaddr));
> +     ((struct ethercom *)ifp)->ec_capabilities = ETHERCAP_VLAN_MTU;
>       ifp->if_flags =
>           IFF_BROADCAST|IFF_SIMPLEX|IFF_NOTRAILERS|IFF_MULTICAST;
>       ifp->if_snd.ifq_maxlen =
> 
> gives the behaviour I wanted.
> 
> I actually don't know whether it passes oversized packets correctly,
> because the vlan interface I'm using in the bridge is an untagged
> interface (see the thread on tech-net about untagged vlans).  Hmm, I
> should improve that so that an untagged vlan interface does not
> subtract off encapsulation overhead from its parent's MTU; that would
> have done as well in this case (but I think xennet and xvif should
> support VLAN MTUs in any case).

You can tcpdump on the xvif interface, and see if oversized packets reaches
it.

-- 
Manuel Bouyer <bouyer%antioche.eu.org@localhost>
     NetBSD: 26 ans d'experience feront toujours la difference
--