Re: [exim] binning mails on delivery based on headers

On Fri, 30 Sep 2005, John Oxley wrote:

> Okay I have mails coming through a spam/virus filter to a separate
> box running exim 4.52 and courier-IMAP/POP on FreeBSD 5.4.

Then, to be candid, I'd want to redesign the system. exim is an MTA:
it's supposed to provide a reliable mail service, which IMHO means
either positively accepting, or positively rejecting, mail items.
Silently dropping items into a black hole (automatically, without
postmaster/admin intervention) isn't an option that I'd care to
consider for the MTA, due to considerations of false positives: (if
individual users want to construct their own mail filters to do that,
on their own authority, then that's different).

By all means have a separate back-end engine that does the scanning,
if it leads to a more balanced server load, but I'd want the actual
MTA to consult the scanner at SMTP time (i.e during the DATA phase)
and give the offering MTA a definite answer one way or the other.

If that leads to too much system load, maybe there are some
more-effective tests that one can carry out at an earlier stage to
discern and reject abusive offers, thus limiting the number of items
which have to get fed to the scanner. In our case, only a minority of
abusive offers ever get to the DATA stage, and a proportion of those
get rejected relatively cheaply too, before the overhead of
spamassassin is contemplated.

Of course the (unstated) assumption of your posting is correct, that
once an item has been accepted it's dangerous to automatically bounce
it, due to faked envelope sender addressses; so, once it's accepted,
if it's considered abusive then the only options would seem to be
silently discarding it (as you've proposed) or passing it for
postmaster/admin inspection. Since neither of those seem IMHO
particularly attractive, my "take" is that one should avoid designing
a system such that it provokes that situation. And that, to me, means
designing the system such that a positive acceptance or a positive
rejection is feasible at SMTP time.

Sorry if this seems so dogmatic, but that's my best counsel.

regards