On Tue, 26 Mar 2002, Matthew Byng-Maddick wrote:
> On Tue, Mar 26, 2002 at 11:05:04AM +0100, Patrick von der Hagen wrote:
> > But second, I'd expect exim to detect a timeout and react accordingly.
> > Since TCP is no stateless connection, the sending of a 250-message to
> > accept a mail should return an error-code. I'd be very surprised if
> > Phil did not check this error code.
>
> This would potentially break reliable mail delivery. Think about the
> situation where it's actually the link that's gone down in between the
> two machines, and you don't know at what time the other end has stopped
> being able to see your machines. By the time you send 250, the message
> should be queued for delivery.
Indeed. Exim cannot know if its message has got back to the sending
client. In fact, it never checks any writing to the TCP/IP socket - it
detects errors when it fails to *read* from the client. In this endgame
situation, if local_scan() says "ok", it takes responsibility for the
message whatever happens about sending a response.
--
Philip Hazel University of Cambridge Computing Service,
ph10@??? Cambridge, England. Phone: +44 1223 334714.
