Il 19/07/2012 20:20, Ed Maste ha scritto:
It would be a good idea to update netdev_bsd_send() to fix the issue
>mentioned by the comment:
>
> /* XXX should support sending even if 'ethertype' was
NETDEV_ETH_TYPE_NONE.
> */
>
>because according to commit 76c308b50d (netdev-linux: Support 'send'
>for netdevs opened with NETDEV_ETH_TYPE_NONE.), which fixed the same
>issue in netdev-linux, bonding will not work without this issue being
>fixed.
Hi,
we are looking at this issue right now, but we are not sure we
understand it correctly. NETHDEV_ETH_TYPE_NONE is no longer used in the
current code, and I have been unable to setup a scenario that triggers
the code introduced by commit 76c308b50d. I have done something like this:
(tap0 and tap1 already exist and have some processes listening on the
corresponding file descriptors)
ovs-vsctl add-br br0 -- set bridge br0 datapath_type=netdev
ovs-vsctl -- --fake-iface add-bond br0 bond0 tap0 tap1 # also w/o
--fake-iface
ifconfig br0 10.0.0.1/24
ping 10.0.0.2
I can see packets coming out of tap0, say. If I kill the process
listening on tap0, the packets come out of tap1. In all cases,
ovs-vswitchd had called netdev_linux_listen() on both tap0 and tap1, and
therefore the branch in netdev_linux_send that uses the global AF_PACKET
socket was not taken.
Can you please point out a scenario that actually makes use of the
global socket? That would be very helpful.
Giuseppe
--
Dr. Ing. Giuseppe Lettieri
Dipartimento di Ingegneria della Informazione
Universita' di Pisa
Largo Lucio Lazzarino 2, 56122 Pisa - Italy
Ph. : (+39) 050-2217.649 (direct) .599 (switch)
Fax : (+39) 050-2217.600
e-mail: g.letti...@iet.unipi.it
_______________________________________________
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
