[705] in linux-net channel archive
Re: Linking 2 Ethernets using SLIP
daemon@ATHENA.MIT.EDU (Mark W. Eichin)
Sun Jul 16 23:41:34 1995
Date: Sun, 16 Jul 95 19:24:08 -0400
From: "Mark W. Eichin" <eichin@cygnus.com>
To: fpriano@ull.es
Cc: linux-net@vger.rutgers.edu
In-Reply-To: <480*/S=fpriano/O=ull/PRMD=iris/ADMD=mensatex/C=es/@MHS> (message
from Felix Herrera Priano on Sun, 16 Jul 1995 9:09:49 UTC)
>the figure of an old message to this list from eichin@cygnus.com...thank you
[aside: I'd buy a tool that made it easy to do charts like that :-)
Emacs picture mode doesn't really cut it. Email me if you know of any.]
Since I do have a working setup like this (that's why I drew the
picture, after all, to help debug it :-) I'll braindump the details I
remember; I haven't touched the setup in a while (other than to
replace the dialup SLIP line with a Metricom Ricochet radio modem.)
I chose to 'pseudo-bridge' the two networks, namely, use the same
subnet for A+m+n+o as I did for B+x. In order to make this work, I
have a machine on the Bx net (it can be *any* machine on that net)
performing "proxy arp" [accomplished on a sun by filling in an
/etc/hosts.arp file.] The idea is that this machine has "published"
entries in its arp cache for all of the Amno IP addresses [and note
that A may not be necessary since B will probably take care of that
one automatically] that point to B's ethernet address. Thus, normal
routing will get a packet for m to the Bx gateway, which will then
'arp' for m's address, get B's address, send the packet to B, which
will forward it to A [minor correction, I've got Amno set up as a fake
subnet of Bx-net, in that only B knows about it, and thus forwards
correctly.] A also knows about the fake subnet, and has a default
route pointing to B; mno likewise know about the fake subnet, and have
a default route pointing to A.
This is somewhat of a kludge, and is probably a "bad" design. One
notable side effect is that in order to talk to solaris boxes (and
possibly other boxes doing MTU discovery?) I had to set the SLIP MTU
to be the same as the ethernet MTU on both sides (1500, rather than
something better for interactive performance.) I haven't determined if
this is a plausible though unfortunate side effect, bugs in the
network code on A (a linux 1.2 box) or bugs in the network code on B
(a Portmaster.)
I hope this is enough detail; in my case, Bx is my office, Amno is
"home" with a sparc-sunos, 3 linux boxes, and a sun3-sunos/xkernel
(the sun3 doesn't actually succeed in talking to the outside world,
but one day I'll upgrade it to fix that. In the meantime, it is safer
this way :-)
_Mark_ <eichin@cygnus.com>
