dovecot - Jun 2008 - spamassassin, postfix with dovecot lda?

Hi all,

I am trying to setup a mail server using postfix (virtual_mailboxes), 
spamassassin and dovecot, along with SQL where appropriate. From my 
research so far I realized that, for several reasons, it is required for 
me to use dovecot LDA instead of maildrop, local, procmail or other 
alternatives.

My configuration was working flawlessly, until spamassassin per-user 
configuration came to play. If I just wanted an MTA anti-spam gateway, I 
could directly call spamassassin via master.cf, or through some helper 
"content-filter-application" (like amavis or amavisd-new), and 
everything would work just fine (I tried and tested many such 
configurations with success). But when antispam per user preferences 
became my concern (Bayesian filters and classifier), I realized that 
spamd should be called by the LDA. With procmail this was a trivial 
issue, with dovecot-lda I was unable to find any solution on the web or 
other documentation.

So my question is as follows:
    Is there a way to call spamc from dovecot-LDA (and/or 
dovecot-sieve), or is there an alternate way to do this for a site with 
virtual mailboxes configured with dovecot-lda?

Thank you all in advance,

regards,

-- 
George Mamalakis
IT Officer, 
Department of Electrical and Computer Engineering,
Aristotle Univ. of Thessaloniki, Greece

George Mamalakis wrote:
> Hi all,
>
> I am trying to setup a mail server using postfix (virtual_mailboxes), 
> spamassassin and dovecot, along with SQL where appropriate. From my 
> research so far I realized that, for several reasons, it is required 
> for me to use dovecot LDA instead of maildrop, local, procmail or 
> other alternatives.
>
> My configuration was working flawlessly, until spamassassin per-user 
> configuration came to play. If I just wanted an MTA anti-spam gateway, 
> I could directly call spamassassin via master.cf, or through some 
> helper "content-filter-application" (like amavis or amavisd-new),
and
> everything would work just fine (I tried and tested many such 
> configurations with success). But when antispam per user preferences 
> became my concern (Bayesian filters and classifier), I realized that 
> spamd should be called by the LDA. With procmail this was a trivial 
> issue, with dovecot-lda I was unable to find any solution on the web 
> or other documentation.
>
> So my question is as follows:
>    Is there a way to call spamc from dovecot-LDA (and/or 
> dovecot-sieve), or is there an alternate way to do this for a site 
> with virtual mailboxes configured with dovecot-lda?
>
> Thank you all in advance,
>
> regards,
>
Hi George. You seem to have a similar setup as I have here. I found 
www.postfixvirtual.net most useful! Do have a look there.

I just had a 12 hour fight to get all running again after a upgrade from 
Mandriva 2008.0 to 2008.1. That should have been trivial but I had all 
sort of issues with amavisd missing perl modules. Well, finally solved 
that one again.

Egbert Jan (NL)

George Mamalakis <mamalos at eng.auth.gr> writes:
> So my question is as follows:
>    Is there a way to call spamc from dovecot-LDA (and/or
> dovecot-sieve), or is there an alternate way to do this for a site
> with virtual mailboxes configured with dovecot-lda?
Hello,

I am currently running a similar setup, with spamc called per user at
delivery time, from postfix. I use the following settings (on a Debian
Etch system, so installation paths must be adapated).

- dovecot's deliver binary must be set setuid root,

- /etc/postfix/main.cf:

mailbox_transport = dovecot
dovecot_destination_recipient_limit = 1


- /etc/postfix/master.cf:

#
# dovecot
#
dovecot         unix    -       n       n       -       1       pipe
  flags=Rh user=dovecot argv=/usr/bin/spamc -u ${user} -e
/usr/lib/dovecot/deliver -d ${user} -f ${sender}


-- 
Nicolas