fk@kos.rci.dk (Fleming Kraglund) (03/02/90)
Well, I don't know if anybody has heard about this before but this is something i wouldn't wish on my worst enemy. So i post this here so that everybody can read it and enjoy. First of all, the setup: I run a SysV3.2 (386/ix) with sendmail 5.51 (yes i know it's old) I deliver in the background meaning that after address validation sendmail forks a child to do the actual delivery, thus returns quick to the user. Sendmail supports the Errors-To header, and my email address is there, this means that i get a copy of every letter that fails delivery (if the final mailer supports it), this helps to find first time users and other things. Allso some wise guy at isc thougth that definening LOG is a waste of space, [not to mention DEBUG in the 2.x version grrrrr] Ok, here goes: Last friday i made some changes in the aliases-file. Later i've found out that i made a empty line that started with a space, sounds harmless but see what the result was. Monday morning there was a lot of error mail in my mailbox. I even had error mail on error mail on error mail..., until maxhop stopped that. Well I've seen that before, that was when i made my smtp interface program for our LAN, what i did then was to exit with EX_TEMPFAIL when something went wrong, such as the connection broke or there was a failure when closeing the connection. But oh no, when using a smtp mailer everything about how it went is known to sendmail. But if the mailer exits with non zero, sendmail gets confused and thinks that it went wrong, queues the letter for another try [nice], and sends a copy to the Errors-To [nice?], all this in spite the fact that it has just talked to the recieving sendmail which had just said "recipent@host.dom... Sent." which is smtp for mail delivered. Worst of all my address is on another machine so gues whats happening to the error copy's..... After carefully checking that my mailer really did exit with 0, and from the many number of persons comming to inform me that when they try to send mail their letters was returned, it didn't seem to matter which mailer was used. (and actualy their letter was sent.) Trying sendmail with the -odq to just queue the letter stil faulted, and even trying to verify failed?. Well the problem was as i said in the beginning that there was a line with a space in the aliases-file, which did set the exit value when parsed, but sendmail continued to do it's job, and delivered the mail. After delivery sendmail checks to se if *any* error has occured and if this is the case assumes that delivery failed and did a 'return to sender'. So i hope this is fixed in the new version [which i'm going to get now], b.r Flemming ps: I have a compleate debug flag list for version 5.51 if anybody can use it mail me. -- This is my OWN opinion fk@rci.dk (I hate .sig files) -------------------------------------------------------------------------------- 1/2 * PS = PS/2 (half a computer) | 1/2 * OS = OS/2 (half an operating system)