Date: 2006-08-28 17:56:55 -0400
Subject: milter-abook: skipping Cw, adding MAIL FROM to cache tuple
More information..: http://www.milter.info/#Support
I'm intending to use the milter-abook database to bias spamassassin scores
downward. If sender exists in milter-abook.db *and*:
1) sender SPF is good, give -0.5 point;
2) sender has no SPF record but is in AWL, give -0.5 point;
3) milter-abook to: counter > 10, give another -1 point
4) milter-abook to: counter > 10 and SPF_PASS, give another -2 points
I want to give a small immediate boost as soon as my users communicate with
someone, but I want to to avoid giving a lot of credit to an address that has
only receieved a handful of messages (perhaps accidental, perhaps "REMOVE ME
FROM YOUR LIST!!!"). I want to void giving points when frequent
correspondents are spoofed, hence the SPF.
1) Any initial comments about this scheme? I keep playing with the exact
scores to use, but it seems to be working in practice.
2) I've noticed that when my smtp-authed clients send mail to other local
addresses, those RCPT TO's end up in the abook cache. I can't immediately
find a way to avoid that; milter-abook-To: does something else. I guess it's
not really a problem because I can skip Cw in my SA plugin, but having
milter-abook skip Cw would save a mutex, etc.
3) I would like to give additional points iff the traffic is perfectly
symmetric, i.e, from/to match the previous to/from. If email@example.com
communicates with firstname.lastname@example.org regularly, then it is likely that
email@example.com will welcome some.dude's mail, but it is near certain
that firstname.lastname@example.org will. The obvious way to do this would be to have
milter-abook store both what it currently does, keyed on <email>, and tuples
with key <authed mail from:>-<email>.
Is there a theoretical or performance reason you didn't do this originally? Do
you have any interest in adding it as a feature? Something like
Copyright 2009, 2012 by SnertSoft. All rights reserved.