Re: Patch for ARP code to listen more...

Vipul Gupta (vgupta@cs.binghamton.edu)
Thu, 11 Apr 1996 21:29:59 -0400 (EDT)

It is good to see this finally being addressed. The Linux ARP
code currently does not follow RFC826. In particular, the RFC states
that if a host sees an ARP messages (could be either a reply or
a request), it should look at the source IP and source Hwaddr
(lets say a.b.c.d and p:q:r:s:t:u) and update its cache if an
entry for a.b.c.d exists. Note that the RFC implies that a host
A should update its ARP cache (assuming it has an entry for
a.b.c.d) even when the incoming ARP message is a request to
resolve an IP address other than A's own.

Would be nice if you could extend your patch to include this
behaviour as well. It helps a lot with the 'gratuitous ARP' feature
crucial for the correct implementation of Mobile-IP.

Vipul Gupta
========================================================================
Vipul Gupta
Dept. of Computer Science Ph: (607) 777-2943, Fax: (607) 777-4822
State Univ. of New York E-mail: vgupta@cs.binghamton.edu
Binghamton, NY 13902-6000 URL: http://cs.binghamton.edu/~vgupta

On Thu, 11 Apr 1996, Tim Morley wrote:

>
> A situation i have come across uses arp transmittions to enable to
> swapping between duplicate routers, to allow dynamic changeover in
> event of a failure. (E.G.routing through address a, with duplicate
> routers with addresses b and c (a!=b!=c), but b and c broadcast arp
> packets for a at regular intervals, so routing transparently hops from
> one to the other).
>
> I've modified things as follows, when an arp reply is recieved (for
> any IP address, not just my own), it checks if its in the arp table
> already, and if so updates it. If it isn't it checks to see if the
> reply is to my address and if so adds it to the table.
>
> Its only a little patch, but I would appreciate it if it could go into
> the next kernel.
>
> Tim Morley