[ovs-discuss] Question about OFPP_LOCAL/OFPP_CONTROLLER

[Ben Pfaff]

On Thu, Nov 03, 2011 at 04:35:09AM +0100, Pino de Candia wrote:
> On Wednesday, November 2, 2011 at 4:36 PM, Ben Pfaff wrote: 
> > On Wed, Nov 02, 2011 at 10:34:56AM +0100, Pino de Candia wrote:
> > > I was wondering if you could give me some guidance on how to
> > > investigate this issue further: some packets forwarded to
> > > OFPP_CONTROLLER are arriving with their in_port set to OFPP_LOCAL (not
> > > the actual in_port).
> > 
> > The details you provide are good, except that you didn't say what
> > version of Open vSwitch you are using.
> 
> Oh, sorry about that. It's definitely happening with v1.2.2, but I think we started seeing it in v1.2.1 (I'll double-check with my colleagues). 
> > 
> > Does this seem to happen consistently, that is, for every packet in a
> > flow processed with the actions that you mentioned, or does it only
> > happen for sporadically, etc.?
> 
> It happens for every packet. 

OK, I tried to reproduce this.  I built 1.2.2, created a bridge with
an in-band controller, and created a flow:

    # ovs-ofctl add-flow br0 actions=123,controller

I started a "ping" across that bridge and used "ovs-ofctl snoop br0"
to watch the traffic to the controller.  But the in_ports in the
"packet_in"s look OK to me:

    OFPT_PACKET_IN (xid=0x0): total_len=60 in_port=3 (via action) data_len=60 (unbuffered)

The flow I created was really in use:

    # ovs-ofctl dump-flows br0
    NXST_FLOW reply (xid=0x4):
     cookie=0x0, duration=96.82s, table=0, n_packets=83, n_bytes=4980,
     actions=output:123,CONTROLLER:65535

I also tried with the tip of master but I got the same results.  What
else is necessary to reproduce the problem?

Thanks,

Ben.