Re: [exim] SMTP connection lost

Todd Lyons wrote:
> On Mon, Aug 9, 2010 at 11:21 PM, r00f <r00f@???> wrote:
>> Well, thank you for tip but looks like ECN wasn't enabled on that host
>>
>> #sysctl -A | grep ecn
>> net.ipv4.tcp_ecn = 0
>
> I'm having an issue with a particular site being unable to send emails
> that are larger than about 40K. And same configuration here, ECN not
> enabled on either the load balancers nor the mail servers.
>
>> Servers negotiate TCP session with MSS 1360. Then they start SMTP session,
>> which goes normal during HELO. But when the sending server sends me
>> message
>> headers, I receive:
>
> I'm having trouble with one site sending to my exim servers. They use
> a Barracuda firewall. We have lots of people who send to us, and they
> send out to lots of people. We just experience problems when he's
> sending email to me. On my side, I see large gaps inbetween packets
> that are sent. On their side, they see lots of TCP retransmissions.
> It appears that, for those retransmitted packets, my side never gets
> that first packet.
>
> The MTU is 1390 bytes for this particular host (seems to be 1420 for
> most everybody else). I can find nothing that is causing this.

Grep the list archives. ISTR an issue two or more years ago wherein some (many?)
broadband providers were - in essence - 'stealing' bytes for one overhead reason
or another, AND ICMP was now-and-then firewalled (or ignored?), so the hosts
could not negotiate a mutually useful MTU size - at least for anything that
needed a full MTU (HELO and other basic 'handshakes' would not, but headers
probably would, 'DATA' almost always would do...).

Not (exclusively) an Exim issue - it's 'sub-optimal' configuration of the
underlying TCP/IP...

... but it sounds familiar.

HTH

Bill Hacker


> If
> you find anything in your case, please do detail it as it could be
> related to my case.
>