[1418] in linux-net channel archive
new ARP policy
daemon@ATHENA.MIT.EDU (Werner Almesberger)
Thu Nov 23 14:53:42 1995
From: Werner Almesberger <almesber@lrc.epfl.ch>
To: linux-net@vger.rutgers.edu
Date: Thu, 23 Nov 1995 15:27:50 +0100 (MET)
I've had a look at the recent ARP changes and there are a few things that
look problematic to me. Unfortunately, they'd only show up on multihomed
hosts and I don't have any multihomed hosts under 1.3.4x at the moment,
so I can't test my assumptions. (Well, I could test case 2 ...)
Suspected problem 1: assume two Ether devices eth0 and eth1. If I try to
create ARP entries for both without explicitly specifying arp_dev,
wouldn't one of them fail because dev_getbytype will always return the
same device ? Shouldn't ip_rt_local be used instead ?
Suspected problem 2: can I still proxy ARP on network X.0 for a machine
with address X.N which I reach over some other network ? The tests at
the beginning of arp_req_set seem to disallow that. Such proxies are
useful if I need to have a layer 3 separation (e.g. firewalling or use
of a different network type) without being able to use the appropriate
layer 3 mechanism (e.g. subnetting of X or allocation of a different
network).
By the way, shouldn't ARP entries also be flushed when the IP address
or the netmask of an interface change ?
- Werner
--
_________________________________________________________________________
/ Werner Almesberger, DI-LRC,EPFL,CH werner.almesberger@lrc.di.epfl.ch /
/_IN_R_133__Tel_+41_21_693_6621__Fax_+41_21_693_6610_____________________/
