Path: utzoo!utgpu!attcan!uunet!lll-winken!ames!mailrus!cornell!uw-beaver!rice!sun-spots-request
From: leres@helios.ee.lbl.gov (Craig Leres)
Newsgroups: comp.sys.sun
Subject: Re: Setting netmask
Message-ID: <8901010244.AA00363@helios.ee.lbl.gov>
Date: 7 Jan 89 03:19:21 GMT
Sender: usenet@rice.edu
Organization: U of Maryland, Dept. of Computer Science,gs
Lines: 27
Approved: Sun-Spots@rice.edu
Original-Date: Sat, 31 Dec 88 18:44:36 PST
X-Sun-Spots-Digest: Volume 7, Issue 91, message 8 of 13

Gunnar Lindberg writes:

> I guess this depends on what SunOS you run, but for 3.x (x >= 4) the
> discless clients simply broadcast an "icmp(ICMP_MASKREQ)" on the net
> and use the first reply. File servers don't - instead they depend on

This isn't quite correct. I don't remember if 3.3 or 3.4 was the first
version to support subnets (whichever it was, it the support was in the
kernel, not the documentation) but all pre-3.5 systems that support
subnets will gleefully accept and use any icmp netmask reply they receive.
This generally means that they will use the LAST reply they receive.
(Unless, of course, they get a bogus reply that renders their ethernet
interface useless.) In 3.5, code was added to ignore obviously bogus
netmask replies. Also, this version won't accept a netmask reply if the
netmask has already been set. So it will use the FIRST reply and if you
explicitly ifconfig the netmask, you don't have to worry about it changing
on you.

>     2)	If the network is large, many hosts may try to answer, which
> 	gives quite a peak in network traffic (first one ARP for
> 	each reply, than the reply itself). This was discussed in
> 	"comp.sys.sun" a while ago.

In pre-4.0 systems, the symptom of too many replies was "nd boot error N"
where N is an errno value (typically 12 or "not enough core").

	Craig