[MlMt] SMTP Timeout

[Bill Cole]

On 4 Sep 2014, at 17:03, Benny Kjær Nielsen wrote:

> On 4 Sep 2014, at 22:18, John Grasty wrote:
>
>> Every once in a while a message will timeout and not get sent. The 
>> activity log shows everything down to the closing . , but the mail 
>> server never seems to receive the .
>
> It's not a known issue. Could you send me an example of such a log? 
> Off list if you like.

This has been seen historically in situations where a stupid 
"transparent" proxy (notably some versions of Cisco's PIX/ASA "smtp 
fixup" malware) is mangling SMTP interactions at the application layer. 
The common error is failure to anticipate the possibility of the 
terminating <cr><lf>.<cr><lf> sequence being split between 2 TCP 
segments and/or read() calls (or their logical equivalents), resulting 
in the proxy not seeing the whole termination sequence at once. Some 
SMTP client implementations have adapted to this careless coding by 
making sure that the message itself is pushed in one segment and the 
terminating sequence in a segment of its own. This assures that the full 
sequence isn't split between TCP segments and makes it very unlikely 
(but not quite impossible) that it will not be returned intact in a 
single read().

That tactic also slightly mitigates a similar problem when a sending 
system uses "Path MTU Discovery" and sets the "DONT FRAGMENT" flag on 
packets, while some intermediary is dropping ICMP "MUST FRAGMENT" 
replies of some distant device that can't deal with large packets. That 
problem usually hits every message over 1500 bytes to a particular SMTP 
server, instead of the "every once in a while" occurrence of the split 
termination sequence problem.