Our mail system has many addresses which we accept and forward to our
users' addresses on other systems (ie, domain.com is a virtual alias
domain, and we forward [EMAIL PROTECTED] to [EMAIL PROTECTED], or whatever).

We use both reject_unverified_sender and reject_unverified_recipient to
check that the envelope from/to addressed are valid, so we never
generate the "obvious" backscatter where you're an naive MX that accepts
mail for a bogus addresses at the final destination, and then generate
bounces when the recipient turns out to be invalid.

However, we're now finding that more and more of our users' final mail
destinations are applying SMTP-time spam filtering, so actually reject
the message we're trying to forward due to it's contents.  Because we've
already queued the mail, we then end up generating bogus bounce messages
to the original (obviously forged) senders.

We're now starting to get people reporting us to the likes of SpamCop
for originating backscatter, but in this case I really don't know how we
can avoid it. I've considered things like (in order of preference):
* forwarding the mail whilst the original sender is connected, and
  returning any errors to them, so never spooling the mail
* never actually generating and sending bounce /mails/ except to local
  users/destinations (who are using us as their smarthost), because we
  never accept mail for invalid destinations in any other circumstances
* sending the bounces for the user's forwarders to the user's local
  mailbox, rather than the original sender

Unfortunately I don't really know how to do any of them with postfix. I
did have a vague idea of doing some address re-writing, and then
pretending that the remote MX is a content_filter, but this absolutely
terrifies me. Does anyone have any better ideas?

Obviously we can look at deploying our own SMTP-time spam rejection with
content_filter, but that seems to a) pass the problem on to anyone who's
trying to forward mail to us, b) not solve the real issue which is that
we can never have /exactly/ the same definition spam as the systems
we're forwarding to.

Thanks,
Rob

-- 
Robert McQueen
Bluelinux Internet Services Ltd.

Reply via email to