persoanlizing SPAM filters with dovecot's ldat delivery for virtual users

[Chris Kottaridis]

On Thu, 2011-07-28 at 07:32 -0430, Patrick O'Callaghan wrote:
> On Thu, 2011-07-28 at 00:43 -0600, Chris Kottaridis wrote:
> > My question is whether there is anyway for the virtual mail users I
> > now have to do a similar kind of learning ?
> 
> What is a "virtual user"?

They don't have accounts on the machine so they can't login. But, Imap
recognizes them as users. They do have a home directory from imap's
persective.

I got the term from the dovecot documentation here:

http://wiki2.dovecot.org/VirtualUsers

I also did find out something last night on further search that I can
potentially connect spam to an SQL database. With that tool I may be
able to run sa-learn on virtual users mail to define virtual user
specific filtering. Have the users store mail they consider to be spam
in a known mailbox name. Then run a cron job to run sa-learn on each
user's mailbox telling it's --spam and run sa-learn on the other
mailboxes telling sa-learn it is --ham and run sa-learn with the -u
option.

From what I can tell sa-learn will then store user specific spam info in
the database. I still need a way to run spamassassin on incoming mail to
let it know what the username is that the email is destined for. So,
maybe I need to create a wrapper to make a local mailer that runs spamc
on mail that hasn't already been tagged as spam using the username and
then fed that into dovecot-lda for the actual delivery.

This is one of the pages that talks about spamassassin and databases:

http://vps.donwyatt.com/My_SpamAssassin_MySQL_How-To.html
http://www.wistful.net/wiki/My_SpamAssassin_MySQL_How-To
http://wiki.apache.org/spamassassin/UsingSQL

The last one has a section entitled:
"Global, Per-Domain, and Per-User Preferences via SQL"

I don't know for sure yet if this will give me what I want, because I
haven't really had time to play with anything. But, I do have hope.

Thanks
Chris Kottaridis

> 
> poc
>