SUMMARY: Weird mail problem

Date: Thu Jun 13 1991 - 21:12:04 CDT

>On one 3/60 (a diskless client) mail to another machine on
>campus is sent to our mailhost ok and on to destination
>On the problem 3/60 (a disk server) mail to another machine
>is attempted to send direct.
>The mail files just mount up in the /var/spool/mqueue directory and
>sendmail keeps on trying and getting refused.

> My comments on lines after each ">" . This fix is the last one sent. I
> have included all replies as they all were different and looked like they would
> fix some particular problem.
> Thanks All
> Peter Glassenbury

From: (Jim Hendrickson, FMC-Minneapolis)
 Sounds like the problem is in the gateway link. Two things are possibilities:
 - The networks are different (or netmasks are different), resulting in a need
   for routing, and there is no specific or default route.
>All networking is ok - this would probably show up somewhere else as well
 - If the message is "connection refused" it means that the socket isn't
   configured for the service (/etc/services should show port 25 for mail),
   or if it is something like "Host name configuration error", it usually
   means that the name you use for the host isn't the same as that the remote
   machine knows *that interface* by. This often occurs when there are two
   ethernet cards in one machine.
>Were getting a message like this but were supposed to as the machine
>shouldn't have tried to send the mail direct to that machine.

From: pxl@odin.LANL.GOV (Peter Lomdahl, Los Alamos National Laboratory )
Sounds like you have different entries in /usr/lib/mailhosts. Any host
mentioned there is tried for direct delivery. Hope this helps.
> We didn't have a /usr/lib/mailhosts at all - adding one didn't make
> any difference.

From: (Ian MacPhedran, Engineering Computer Centre,
                                University of Saskatchewan.)
.. you have defined the "T" macro to point to a file of
acceptable destinations, and these files are different between the two
machines. Look for a line starting with "FT=" in your
If you don't define this macro, the machine picks the names it's allowed
to send to from YP/NIS and will throw mail at any machine it knows about.
> This seems similar to above but adding the macro didn't seem to help

The server thinks it's mailhost...
.... try
 A) make sure "mailhost" is defined somewhere, & that it's defined properly.
    (ie, either via yp or in /etc/hosts).
 B) Change the definition of the "R" macro in to the real name
    of the mailhost machine, or change all instances in where the
    appropriate mailer is invoked (presumably "ether" or "ddn" or "tcp")
    so that mail is always pointed at your mailhost machine.
> A - it was. B - I didn't want to get into this just yet
    Run sendmail in test mode
    /usr/etc/sendmail -bt -C/etc/
> 0
> I had tried sendmail with full debugging and it gave exactly the same for
> both machines - it just did different things after that. Thats what made
> me think it was not a sendmail problem but some sort of networking one.
> Here is the clue that seemed to get the fix
> - The directory /var/spool/mail is a link to a globally mounted partition

From: ( Dan K.)
You neglected to mention whether the two Sun3s were mounting the same
/var/spool/mail. sendmail senses whether this is an NFS mount and
forwards to the NFS server if so...
> I think this is also part of the answer
> Below is the fulldescription and fix. I suppose this could be called a bug
> in SunOS4.0.3 because the problem didn't happen on another server with
> similar setup but running SunOS 4.1.1

From: (Michael Shon, GCA/Tropel Division)
> This almost didn't make it but a postmaster caught it and sent it to me
> which was just as well as this was the fix for our problem

If your uses option 'R' ,as in
        # Remote mode - send through server if mailbox directory is mounted
then it could be that on the diskless machine, since the directory is
never a local filesystem, it always formards to its server. On the diskfull
machine, sendmail behaves differently when it sees that /var/spool/mail is
a link, rather than a mount. It does not matter that the link points to a
mount, it is very particular.
This bit me when I tried to make our /var/spool/mail a pointer to a mount
(to protect us from server failures). All the machines that used to work
suddenly tried to make direct connections.

Option R is supposed to permit a server name
so I tried to force it to forward to the server by making /var/spool/mail
NFS mounted on the local system, but it ignored the servername on my 4.0.3
systems. Damn!

I hope this is not your problem, because I couldn't think of a fix, except
to mount on /var/spool/mail. If I thought that was a good idea, I wouldn't
have tried to find a way around it.

If I guessed right, please let me know.
I like to think I know what I'm doing.
Good luck.

> That was it - Thanks Michael - I have mounted the mail partition directly
> on /var/spool/mail - It works- Why did you think it was such a bad idea
> Is it just because it will hang the server if the mailhost hangs? This
> wouldn't matter too much in our situation but may be nasty at other
> sites

This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:06:15 CDT