Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!linus!philabs!cmcl2!seismo!hao!hplabs!sri-unix!mogul@SU-SHASTA.ARPA
From: mogul@SU-SHASTA.ARPA
Newsgroups: net.unix-wizards
Subject: Re: internet broadcast addresses
Message-ID: <1277@sri-arpa.UUCP>
Date: Mon, 25-Jun-84 19:38:00 EDT
Article-I.D.: sri-arpa.1277
Posted: Mon Jun 25 19:38:00 1984
Date-Received: Sat, 30-Jun-84 01:30:44 EDT
Lines: 30

From:  Jeff Mogul <mogul@SU-SHASTA.ARPA>

    I understand that NIC is recommending the use of all 1 bits in the host
    portion of an internet address to denote a broadcast datagram.  (That is
    the low order 24, 16 or 8 bits depending on whether the network is class
    A, B, or C.)  Unix 4.2bsd uses all zero instead of all one bits for that
    purpose, and to convert would take more than a simple redefinition of the
    constant INADDR_ANY because of the variable length of the hosts field.

There was an IEN (out of BBN) issued a few years ago, on an unrelated
topic, that specified this form for a broadcast address.  I discussed
this with Jon Postel when I decided to write an RFC on broadcasting, and
he more or less declared "all ones" to be the standard.  My RFC has not
yet been issued, but I am fairly sure that any "standard" for IP
broadcasting willl be all 1's rather than all 0's.

This is yet another case where Berkeley disregarded informal standards,
to the detriment of all 4.2BSD users (Another case is the infamous "trailer
protocols", but at least you can turn those off.)

    Has anybody had a clash of software incompatibilty over this point?  I am
    expecting to run into the problem soon at this university.

I'm about to test some code that should solve the problem.  When I get
it working, I'll announce it over various lists.  The fixes involve
minor changes to all device drivers, larger changes to a few other kernel
modules, and probably large changes to programs such as rwhod that
use broadcasts.

-Jeff