On Wed May 13, 2009 at 15:53:34 +0200, Ernesto wrote:

> breaking the RFC by guessing how a spam issueing address might look is
> not good.

  Indeed, which is why I wondered how other people solve this
 specific problem.

> Better is to reject a spam mail directly during connection phase,
> because once your mail server has accepted the mail, you _must_ deliver
> it to the user anyway (with special tags i.e.), because of data
> protection laws!

  (This particular case is for my own domains - the only user is myself.)

  I appreciate that I must make the decision to reject during
 the SMTP phase.  My specific problem is that unsolicited/faked
 bounce messages are essentially a different category of mail.

  It is not sufficient to pipe the mails through spambayes,
 spamassassin, crm114, etc.  They do a reasonable job of recognising
 these messages but none are perfect.  Given that in qpsmtpd we
 have easy access to the envelope sender we can reliable identify
 bounces - it mostly becomes a policy decision on what to do with
 them.

  So far I've been accepting them but storing them in a throwaway
 account via tagging + procmail.  (/home/automated/bounces/{new cur
 tmp}).  But when there are 2000 plus messages arrived overnight
 from a spam run you didn't sent it is galling to have accepted them.

  Still it seems that the standard wisdom is that I either accept
 all bounces, or have a list of addresses which send mail,
 or I implement some kind of outgoing signatures which I can
 validate upon the receipt of a bounce.

  None of those solutions are terribly great, but if that
 is the best I can do then so be it.

> Most of the spam is issued by robots working on hijacked Windoze-PCs and
> is sent without using standard mail protocolls - re-trying delivery
> i.e.. So greylisting is doing a good job, but binds many resources
> (database processing, disk space etc.)

  Greylisting seems less effective these days, but given that most of
 the bounces come from spam victims running real mailservers, rather
 than zombie machines, I suspect it would just delay the inevitable.

> Please have a look at my spam statistics **):

  I think I win on sheer numbers, rejecting just over 8.5 million
 spam messages in the past 30 days..

Steve
-- 
Debian GNU/Linux System Administration
http://www.debian-administration.org/

Reply via email to