Path: utzoo!attcan!uunet!wuarchive!zaphod.mps.ohio-state.edu!uakari.primate.wisc.edu!uflorida!kluge!scs!ibarrac
From: ibarrac@scs.fiu.edu (Carlos Ibarra)
Newsgroups: comp.mail.sendmail
Subject: Problem with MXs in Sun sendmail
Message-ID: <1267@kluge.fiu.edu>
Date: 6 Sep 90 14:49:15 GMT
Sender: news@kluge.fiu.edu
Reply-To: ibarrac@scs.fiu.edu (Carlos Ibarra)
Distribution: na
Organization: Florida International University, Miami
Lines: 34

At Florida International University, we have a network of Sun workstations
and a mailhost machine. I have set up the nameserver so that each workstation
has two MX RRs, one with preference 10 for itself, and one of preference 20,
for the mailhost machine. The idea is that the workstations are down more
often than the mailhost, so if a workstations is down and some outside machine
is trying to send mail to it, the outside machine should try the workstation
first, then the mailhost, which should accept the message.

The problem is, this does not work. If a workstation is down, the mailhost
machine's sendmail is contacted by the outside machine and receives the
message. Then, the mailhost sees that the destination machine is not itself,
and tries to contact the destination machine. Since that is still down, it then
tries the next MX which is itself. At that point, it fails with this error:
   ----- Transcript of session follows -----
Connected to serss0.fiu.edu:
>>> HELO fiu.edu
<<< 553 fiu.edu host name configuration error
554 cova@polar ... Service unavailable

Here, polar.fiu.edu is the destination machine, serss0.fiu.edu is the mailhost,
which also responds to the name fiu.edu.

Sendmail tries to contact itself through a TCP connection, therefore when it
recognizes that it is talking to itself, it cuts the connection to avoid a mail
loop.

According to RFC974 (Mail Routing and the Domain System), if the sender's own
name appears among the MX RR's for a destination machine, all MX RRs with
preference >= the sender's MX should be discarded. Does Sun's sendmail not
do this? If it did, it would not work anyway, right?

How can I make the mailhost machine accept the message which was originally
intended for the other machine (which is down) and not attempt to forward it?