Running pixiecore on a specific NIC

[danderson]

Running pixiecore on a specific NIC

[hangxie]

Here are logs:

2017/01/04 23:17:25 [DHCP] Offering to boot 80:e0:1d:d7:2a:5c
2017/01/04 23:17:25 [DHCP] Ignore request from unexpected NIC enp21s0/10.10.10.86
2017/01/04 23:17:25 [DHCP] Offering to boot 80:e0:1d:d7:2a:5c
2017/01/04 23:17:29 [TFTP] Sent "80:e0:1d:d7:2a:5c/0" to 172.16.10.15:2070
2017/01/04 23:17:29 [TFTP] clamping blocksize to "172.16.10.15:2071": 1456 -> 1450
2017/01/04 23:17:29 [TFTP] Sent "80:e0:1d:d7:2a:5c/0" to 172.16.10.15:2071
2017/01/04 23:17:29 [DHCP] Offering to boot 80:e0:1d:d7:2a:5c
2017/01/04 23:17:29 [DHCP] Ignore request from unexpected NIC enp21s0/10.10.10.86
2017/01/04 23:17:29 [DHCP] Offering to boot 80:e0:1d:d7:2a:5c
2017/01/04 23:17:29 [HTTP] Sending ipxe boot script to 172.16.10.15:15471

And netstat:

tcp        0      0 172.16.10.30:18081      0.0.0.0:*               LISTEN      28791/pixiecore
udp        0      0 172.16.10.30:69         0.0.0.0:*                           28791/pixiecore
udp        0      0 172.16.10.30:4011       0.0.0.0:*                           28791/pixiecore
raw        0      0 0.0.0.0:17              0.0.0.0:*               7           28791/pixiecore

[panchoh]

I think this bit me a while a go, but I failed to report it here. Thanks for reporting and fixing it!

i suggest that you create a pull request with your changes, so that danderson can process it more easily. Also, note that you will have to sign a "Contributor License Agreement' (CLA). Don't worry, the googlebot will guide you through it.

Cheers!

[danderson]

Thanks for the report! I need to debug some more and figure out why the raw socket behavior is weird. Your patch seems fine, although I will probably write my own by adding more logic to the BPF program, if I can't make the socket work properly.

[danderson]

Ah, I suspect I know. My guess is that binding the raw socket to a specific IP filters out broadcast packets, so the dhcp4 package never receives the DHCPDISCOVER. Fortunately, that's easy to fix with a small extension of the BPF program. I'll test that now.

[danderson]

@hangxie I think the commit I just pushed fixes -l to work correctly. I've tested both with and without --dhcp-no-bind, and now I can boot machines with a pixiecore listening on a specific address. Can you confirm?

[hangxie]

It works, thanks for the quick fix.