On Sat, 2006-02-18 at 02:15, Bob Dodds wrote:

> qpsmtpd catches a lot of spam and phishing at the
> protocol stages. It's not just a content filter.

Agreed, but so does mimedefang. It gives you hooks
to every part of the smtp operation. 

> "More efficiently" is troll bait, considering how much
> inefficiency is due to spam, phish, and worm load and
> how much "access" is determined by lookup calls to
> the same other tools by either qpsmtpd or sendmail.

The 'access' in question is a dbm style database and
sendmail does matches against it in c code.  I know there
are valid arguments about perl code being as fast as
c sometimes, but sendmail does this pretty quickly.
My point was simply that if you can reject based on
this before the message even gets to the slower network
tests or content scans it will be more efficient.
For example, if you normally relay to an internal server
and a lot of mail still comes addressed to an ex-user,
rather than doing the remote smtp check for every message
you can put:
To:[EMAIL PROTECTED] ERROR:550 user unknown
in the access database on the outside machine to bounce
it without going any farther.

> qpsmtpd can use milters and mimedefang, so neither
> of those is worth points in the sendmail column.

This would be more convincing if you verified that
someone has done it.  I apologize for bringing it up
and not actually making it work.  Mimedefang appears
to exercise most of the possibilities of the milter
interface so I'll be surprised if it works on the
first attempt.  Some things may turn out to be fairly
intimate with sendmail like the way you would pass,
whether the connection was using a TLS certification or
was authenticated through the milter interface. 

> And
> I just implied in the previous column, ldap and sql
> are not worth points in the sendmail column, since
> qpsmtpd can use any lookup and auth method that
> sendmail can.

Of course - it is possible to write just about anything
in perl.  But most things people need have already been
done when you combine the features of sendmail and mimedefang.
I notice there is still work in progress getting remote
smtp recipient verification to work before accepting a message
to rely.  Mimedefang has had that for some time although
it doesn't have to perform the relay operation because
that's built into sendmail.

> Just what does sendmail exclusively have, and show
> me its realtime smtp protocol filtering.

It already has the ability to listen for smtp/smtps/submission
at the same time without losing the relay IP address as you
do with stunnel and can process starttls. It can use PAM based
authentication (maybe it needs sasl for that..). It understands
relatively complex schemes of virtual users, aliases, .forward
files and the necessary recursive expansions to process them.

> I know, you
> are touting the "efficiency" of a vacuum, of a lack of
> protocol filtering?! Must be.

Mimedefang does the realtime protocol filtering part.
You can filter based on the relay address, the envelope
sender - actually any of the ESMTP arguments can be
accessed during filter_sender(), the recipient, or
you can add custom code.  For example you can call
md_check_against_smtp_server() inside filter_recipient
to reject unknown users early if you are going to
relay to another machine for final delivery.  If you
continue on to content filtering, mimedefang breaks
out the attachments in temporary files so you can
efficiently run your choice of scanners before the
SMTP accept happens.  If, in fact mimedefang will work
with the qpsmtpd milter interface, none of this part
would be exclusive to sendmail.  

-- 
  Les Mikesell
   [EMAIL PROTECTED]


Reply via email to