Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!uunet!seismo!rutgers!topaz.rutgers.edu!hedrick
From: hedrick@topaz.rutgers.edu (Charles Hedrick)
Newsgroups: comp.bugs.4bsd
Subject: Re: various bugs in 4.x (4.3?) TCP code
Message-ID: <14008@topaz.rutgers.edu>
Date: Sun, 16-Aug-87 22:50:25 EDT
Article-I.D.: topaz.14008
Posted: Sun Aug 16 22:50:25 1987
Date-Received: Mon, 17-Aug-87 04:45:35 EDT
References: <13996@topaz.rutgers.edu>
Organization: Rutgers Univ., New Brunswick, N.J.
Lines: 20

After some additional testing and purusing packet logs, I have
convinced myself that the meltdown described in the my previous
message will be prevented if either of the hosts involved in the
conversation has the fixes to tcp_input.c and tcp_fsm.h posted in my
previous message.  In that message I suggested that it might be
dangerous to put up Van Jacobson's Sun TCP modifications unless you
could make sure that every host in the world got the fix at the same
time.  Fortunately, this turns out not to be the case.  Since it is
only necessary for the fix to be present on one end, this means that
it is safe to put up the new TCP as long as all the hosts you put it
up on get the fix.  I have so far been unable to explain what it is
about the new TCP that causes the problem to happen.  As far as I can
see, the bug was present in SunOS 3.2, which means it was almost
certainly present in 4.2BSD.  As far as I can tell, the only reason it
hasn't turned up before is that timings happened to work out.

I should note in passing that this entire business may be unimportant
to other sites.  Problems that happen every few hours at Rutgers
happen every few months elsewhere, and every few months, things happen
here that are simply impossible.  It must be the air.