On Fri, Mar 30, 2012 at 03:13:23PM -0700, Ansis Atteka wrote: > On Fri, Mar 30, 2012 at 2:06 PM, Ben Pfaff <b...@nicira.com> wrote: > > > On Fri, Mar 23, 2012 at 06:04:50PM -0700, Ansis Atteka wrote: > > > This patch removes port from datapath before vswitchd sets ofport column > > > to -1 (e.g. if user-space attempted to create a duplicate tunnel then > > > kernel would reject config changes and leave the old tunnel with its old > > > config. Hence the user-space have to remove it) > > > > > > Signed-off-by: Ansis Atteka <aatt...@nicira.com> > > > > I don't yet understand the problem. Can you explain, step-by-step, > > how this comes into play and what difference it makes? > > > By the way, this patch must be applied together with the other patch I sent > out previously: http://openvswitch.org/pipermail/dev/2012-March/016053.html. > Otherwise user-space would not set ofport column to -1 at all after > executing > following commands: > > > ovs-vsctl add-br ovsbr > > ovs-vsctl add-port ovsbr port1 > > ovs-vsctl add-port ovsbr port2 > > ovs-vsctl set Interface port1 options:remote_ip=2.1.1.1 options:key=123 > > type=gre > > ovs-vsctl set Interface port2 options:remote_ip=1.1.1.1 options:key=123 > > type=gre > > ovs-vsctl set Interface port2 options:remote_ip=2.1.1.1 options:key=123 > > type=gre > > > After you would apply that datapath patch, ofport column would > get set to -1 correctly for port2. But the datapath would still have > the port2 hanging there with the old config: > > > root@aatteka-MacBookPro:~# ovs-dpctl show > > system@ovsbr: > > lookups: hit:0 missed:0 lost:0 > > flows: 0 > > port 0: ovsbr (internal) > > port 5: port1 (gre: key=123, remote_ip=2.1.1.1) > > * port 6: port2 (gre: key=123, remote_ip=1.1.1.1)* > > root@aatteka-MacBookPro:~# ovs-vsctl show > > b3e2d947-529c-4bf3-80e0-d3e360a8ab55 > > Bridge ovsbr > > Port "port2" > > Interface "port2" > > type: gre > > * options: {key="123", remote_ip="2.1.1.1"}* > > Port "port1" > > Interface "port1" > > type: gre > > options: {key="123", remote_ip="2.1.1.1"} > > Port ovsbr > > Interface ovsbr > > type: internal > > ovs_version: "1.6.90" > > > Assuming that interface is considered non-operational, when ofport > column gets set to -1. I thought that more correct behavior would be to > make user-space remove the old port2 from datapath (because it is > out of sync from OVS DB anyway).
Ah. I understand now. Thank you. Will you add the above, or a summary of it, to the commit log? The only issue I noticed in the patch itself was that one of the lines appeared to be incorrectly indented (perhaps you used tabs instead of spaces). Thanks, Ben. _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
