Re: rfc 1122 and ICMP SOURCE HOST ISOLATED

barns@cove.mitre.org Tue, 15 November 1994 17:49 UTC

Received: from ietf.nri.reston.va.us by IETF.CNRI.Reston.VA.US id aa06182; 15 Nov 94 12:49 EST
Received: from CNRI.Reston.VA.US by IETF.CNRI.Reston.VA.US id aa06178; 15 Nov 94 12:49 EST
Received: from venera.isi.edu by CNRI.Reston.VA.US id aa10565; 15 Nov 94 12:48 EST
Received: from mwunix.mitre.org by venera.isi.edu (5.65c/5.61+local-19) id <AA15695>; Tue, 15 Nov 1994 09:29:15 -0800
Return-Path: barns@cove.mitre.org
Received: from gateway.mitre.org (gateway.mitre.org [128.29.31.10]) by mwunix.mitre.org (8.6.4/8.6.4) with SMTP id MAA20934; Tue, 15 Nov 1994 12:28:37 -0500
Sender: ietf-archive-request@IETF.CNRI.Reston.VA.US
From: barns@cove.mitre.org
Received: from cove.mitre.org.mitre.org by gateway.mitre.org (4.1/SMI-4.1) id AA22429; Tue, 15 Nov 94 12:28:34 EST
Received: by cove.mitre.org.mitre.org (4.1/SMI-4.1) id AA02999; Tue, 15 Nov 94 12:28:21 EST
Message-Id: <9411151728.AA02999@cove.mitre.org.mitre.org>
To: ietf-hosts@isi.edu
Cc: Gabriel.Montenegro@eng.sun.com, braden@isi.edu, Steve.Drach@eng.sun.com, postel@isi.edu, Kent@bbn.com, barns@cove.mitre.org
Subject: Re: rfc 1122 and ICMP SOURCE HOST ISOLATED
In-Reply-To: Your message of "Fri, 11 Nov 94 16:56:16 PST." <199411120056.AA21711@zephyr.isi.edu>
Date: Tue, 15 Nov 1994 12:28:20 -0500

Sigh, I propose 6-10 are "hard", 11-12 are "soft", and 13 (if it was
ever really allocated???  "Communication administratively prohibited")
is "hard".

It seems practical to treat 6 as "hard", but we're somewhat
inconsistent with our religion here.  Isolated is not administrative,
it's an error condition in the network; it means "almost all
destinations are unreachable due to network or host glitch".  In
retrospect, I wonder if the code is even a good idea; not only is
the means of detection net-specific (which is OK with me) but the
meaning of the message is somewhat net-specific.

If there is no carrier on your ethernet interface should your kernel
synthesize Dest Unreachable code 8 when you try to send out that ethernet
interface?  Does anyone actually do this?

ICMP sure is a mess.  But this seems to be a morass for everyone (look
at X.25 and ISO 8208 - 10 times as many error codes, same inconsistency
among implementations, too many codes, people don't use specific codes
when they should, there are not enough codes to describe all diagnostic
information without loss of information, ...)  There's probably a paper
in this somewhere - "Self-Similar Disparity in Error Code Allocations"...