Re: [apps-discuss] I-D Action: draft-ietf-appsawg-greylisting-01.txt

["John Levine" <[e-mail]>]

Based on my experience, here's some advice for how to make greylisting
work.  My goal is to use it to recognize broken mail clients that
don't retry.  Once a client has shown that it knows how to retry,
there is no further benefit from greylisting it.  (It may be sending
spam, but you'll have to recognize it other ways.)

* Greylist on the triple [IP, MAIL FROM, RCPT TO].  After a successful
retry, whitelist all further mail from the IP.  I don't feel strongly
whether you should put just one RCPT TO in the triple or wait for DATA
and put all of them.

* The greylister should have a time window within which retries will
be accepted.  On mine, the minimum is one minute, the maximum is 24
hours.  I think those are reasonable.  You want a minimum because some
spamware retries immediately, you want a maximum both to keep the
retry table from filling up with junk and so two spams sent a week
apart aren't interpreted as a retry.  Some people set their minimum
much longer, several hours, but I don't see any benefit to that unless
you believe in the blacklist theory.

* Whitelist entries should time out eventually.  I delete them when
there's been no mail from the IP for a week.

* If you have multiple incoming servers at the same MX distance, they
should share a single greylist database.  This handles the case where
the retry goes to a different server, and lets all the servers share
the IP whitelist.  I do this by making the greylister a standalone
daemon, and use a very simple UDP protocol to query it.  If you're
Google or Yahoo and have multiple server pools all over the world,
I suppose a database per pool would do.

* Since some senders have multiple outgoing servers, allow approximate
matches on the retry IP.  I take anything in the same /24 which seems
to work OK.  I only whitelist the IP that retries, but I should
probably whitelist both.

* The greylister MUST have a manual whitelist.  Some legit senders
just don't work with greylisting for a variety of reasons, most of
which are not in conflict with RFC 5321.

R's,
John