[comp.mail.sendmail] Caught in a virtual pinball machine

kls@ditka.UUCP (Karl Swartz) (09/03/89)

In article <3897@ditka.UUCP> I write:
>Over the past month or so a handful of messages have ended up
>in the uucp mailbox here on ditka ...

>All seem to have bounced several times before finally finding
>a home here, such that it is.  All seem to have bounced off of
>rt1 early along the way, for reasons which also elude me.  All
>have involved sundc.East.Sun.COM early in the process ...

I spent several hours yesterday and today going over all the
logs I could find.  The first such message came thru ditka at
10:22 pm on August 8th:

    From: pacbell!csed-1!sundc.East.Sun.COM!sundc!
	Sun.COM!claris!ames!postmaster    (wrapped for clarity)
    To: rt1!hc!cs!rock

The rt1!hc link disappeared about that time; rt1 *should* have
stripped the hc, found the cs (which it doesn't know), and sent
it to ditka for routing.  (I know, re-routing is bad, but quite
a lot of traffic went via rt1!hc and the link was rather abruptly
dropped, certainly before the maps were updated.)

Instead, a few minutes later I see the following going thru ditka:

    From: rt1!ditka!pacbell!csed-1!sundc.East.Sun.COM!
        sundc!Sun.COM!claris!ames!postditka    (wrapped for clarity)
    To: rt1!hc!cs!rock

Note that the "From:" username has been changed.  Apparently one
of the many bugs that IBM introduced in sendmail for AIX 2.2.1 is
a buffer size of only 70 characters for the "From:" address at
some point ("ditka!pacbell ... ames!post" is exactly 70 characters
long; other trashed messages confirm this).

I've been sorely tempted to trash the IBM sendmail in favor of an
unmolested 5.61 version; maybe this is the excuse I've been waiting
for.  Of course there are probably many other hacks that will keep
5.61 from working on AIX without a great deal of painful work.

>I see no reason why either of the messages
>should have passed thru ditka or rt1 at all

This part part still remains a mystery.  I can't find any other
hc's in the maps, and none of the sites involved would have been
routed via the real hc.  If it helps, here is a 'uniq -c' on all
the rmails directed thru rt1 and hc that seemed to lead to trouble:

   1     rmail rt1!hc!4!30
   1     rmail rt1!hc!ajpo!greeneh
   6     rmail rt1!hc!ajpo.sei.cmu!ploedereder
   4     rmail rt1!hc!cs!rock
  17     rmail rt1!hc!csed-1stoltz!mohar
   1     rmail rt1!hc!csed-37!mohar
   1     rmail rt1!hc!fatvax!ebailey
   4     rmail rt1!hc!ida!rwex
   5     rmail rt1!hc!ncs.dnd.ca!smichell
   1     rmail rt1!hc!oids-8!wheeler
   1     rmail rt1!hc!philabs.philips.labs!caj
   1     rmail rt1!hc!philabs.philips.labs!dpb
   1     rmail rt1!hc!philabs.philips.labs!ewk
   1     rmail rt1!hc!philabs.philips.labs!mat
   1     rmail rt1!hc!philabs.philips.labs!mmr
   2     rmail rt1!hc!umass!gsg
   1     rmail rt1!hc!ut.edu.ray!issi

All of these passed thru sundc early in the process, for whatever
that might be worth.

-- 
Karl Swartz		|UUCP		ames!lll-winken!pacbell!ditka!kls
1-505/667-7777 (work)	|Internet	kls@rt1.lanl.gov
1-505/672-3113 (home)	|BIX		kswartz
"I never let my schooling get in the way of my education."  (Twain)

clarke@acheron.uucp (Ed Clarke/10240000) (09/04/89)

From article <3940@ditka.UUCP>, by kls@ditka.UUCP (Karl Swartz):
> 
> I've been sorely tempted to trash the IBM sendmail in favor of an
> unmolested 5.61 version; maybe this is the excuse I've been waiting
> for.  Of course there are probably many other hacks that will keep
> 5.61 from working on AIX without a great deal of painful work.
> 

I picked up the sources from Berkeley.  Once you figure out the correct
configuration options it works without any changes.  I had to pick up
bind 4.8 at the same time to get some missing resolver subroutines ( I
think that they're normally in libc.a on a bsd system ).  I have not
tried the IDA modifications.
-- 
Ed Clarke
acheron!clarke