Chaosnet over Ethernet

On 2005-11-13, Pat Barron

Read Stevens-Wright, TCP-IP Illustrated, Volume 2. At least everything between ipoutput() and etheroutput() and etherinput() and the ARP stuff. There are nice call diagrams, too.

Then read my paper "The New Link-Level Independent ARP Subsystem of NetBSD" 1

I've changed NetBSD ARP to be able to use non-Ethernet as the layer below. Back then, the Free- and OpenBSD network gurus I concated seemed not to be interested, although I think that some FreeBSD activity happened later; the guy in question pointed out a potential problem to me.

I didn't think about a different upper layer. But the method would be the same:

Basically, it's easy - remove the fixed-size field definitions for the ARP payload, create access macros, use them everywhere.

ARP packets have the length of the addresses embedded, so you can use them everywhere once you have set them.

So you'll need changes to the PFINET-AFINET part of etheroutput(), which are minor, and add the PFCHAOS part.

Of course, you'd have to check for consistency, especially on received packets; and add the PFCHAOS case to the switch in the ARP-input function in the ethernet input path.

No real performance problem should happen, as ARP results are cached for long times (several seconds at least).

Regards, -is