[wip] Implement L2fwd

[dpino]

I made this temporary PR to discuss implementation of L2Fwd.

I want to make the L2Fwd a two-way forwarder as suggested by Nikolay (and actually that's how DPDK's l2fwd seems to work) . My setting environment is the following (arrow means NICs are wired with one another).

0000:04:00.0 <-> 0000:05:00.0
0000:04:00.1 <-> 0000:05:00.1

I blast packets on 0000:04:00.0. I create a l2fwd between 0000:05:00.0 and 0000:04:00.1, which forwards packets received in the former NIC to the latter. Packets are received in 0000:04:00.1 but when I try to send back the packets from 0000:04:00.1 to 0000:05:00.0 the link reports 0 packets transmitted:

link report:
           5,493,356 sent on nic1.tx -> l2fwd1.input (loss rate: 0%)
           5,493,356 sent on l2fwd1.output -> nic2.rx (loss rate: 0%)
                   0 sent on nic2.tx -> l2fwd2.input (loss rate: 0%)
                   0 sent on l2fwd2.output -> nic1.rx (loss rate: 0%)

What I noticed is that packets received in 0000:04:00.1 are actually transmitted to 0000:05:00.1 (because cards are wired together). I made a custom app that listens on 0000:05:00.1 and it reports received packets:

link report:
           2,841,169 sent on nic.tx -> sink.rx (loss rate: 0%)
load: time: 3.00s  fps: 1,507,050 fpGbps: 5.747 fpb: 10  bpp: 467  sleep: 0   us

After this observation, I think a deployment of a l2fwd involves 4 NICs:

• A traffic generator that sends packets to a NIC. For instance, packeblaster blasts packets into 04:00.0.
• A l2fwd app forwards packets received in 05:00.0 to 04:00.1.
• A second l2fwd sends packets received in 05:00.1 back to 05:00.0, so packets arrive to their point of origin, 04:00.0.

I think that's what the diagram at http://dpdk.org/doc/guides/sample_app_ug/l2_forward_real_virtual.html explains. Comments? Thoughts?

On the other hand, I'm not sure what's the the best command interface for this app. Should we request the user to introduce 3 PCI addresses? Spare the use of packetblaster, generate packets on our own and request 4 PCIs?

[dpino]

More or less I'm convinced this is the way to go. Use of packetblaster + l2fwd involves 4 ports (NICs). I will update the upstream PR with more changes. I think I will send a patch to the packetblaster too so it can report on received packets (as it's a two-way forwarder packets will eventually get back to the port of the packetblaster, completing the whole loop).