[Dovecot] Dovecot LDA and address extensions

[Tom Metro]

I'm converting over a virtual user installation that used Postfix and 
maildrop with a MySQL back-end for mailbox lookups to use Dovecot LDA. 
The maildrop setup used a database table like:

+-------------------------+--------+-------------+--------------------+
| address                 | owner  | domain      | mailbox            |
+-------------------------+--------+-------------+--------------------+
| postmaster at example.com  | tmetro | example.com | .admin.postmaster/ |
| root at example.com        | tmetro | example.com | .admin.root.lex/   |
| tmetro at example.com      | tmetro | example.com | .INBOX/            |
| tmetro+spam at example.com | tmetro | example.com | .spam.tmetro/      |
[...]


where maildrop was handed ${recipient} from postfix, and it queries the 
above table matching ${recipient} to the address column, getting back 
the expression CONCAT_WS('/','/var/mail',domain,owner) as the home 
directory, and CONCAT_WS('/','/var/mail',domain,owner,mailbox) as the 
mailbox. The UID/GID are fixed, as this is a virtual user setup.

So looking up the mailbox for the user postmaster at example.com returns:
/var/mail/example.com/tmetro/.admin.postmaster/

Currently I'm using a static user database with Dovecot. This SQL 
appears to replicate the maildrop setup:
user_query = SELECT CONCAT_WS('/','/var/mail',domain,owner) as home,
     CONCAT_WS('/','maildir:/var/mail',domain,owner,mailbox) as mail,
     5000 as uid, 5000 as gid FROM vmbx2 WHERE userid = '%u'

And this works for mail deliveries.

The problem with this is that not only is a database lookup overkill for 
the IMAP server, as the root mail location for IMAP users follows a 
simple pattern adequately represented by the static userdb, it also 
breaks IMAP, because the returned 'mail' overrides mail_location, 
(currently set to maildir:/var/mail/%d/%n/), so for user 
tmetro at example.com it returns:
maildir:/var/mail/example.com/tmetro/.INBOX/

instead of:
maildir:/var/mail/example.com/tmetro/

(which is the same as the home directory).

Can the "userdb sql" directive be added to the "protocol lda" section? 
Or perhaps instead create a new "auth lda" section, and move the "userdb 
sql" and "socket listen" directives to that section?


The second issue is how to best deal with address extensions, something 
that wasn't fully implemented in the maildrop setup. Ideally I'd like a 
setup where a lookup is first performed for user+ext at domain, and if it 
fails to return a mailbox, then lookup user at domain. Assuming Dovecot can 
handle a multi-row query result (grabbing the first record and ignoring 
the rest), this can probably be hacked using an appropriate WHERE 
expression.

But the difficulty is getting the address components. Postfix will 
provide them, but Dovecot's deliver doesn't seem to have a way of 
receiving them and passing them on to the user query. The example LDA 
setup shown here:
http://wiki.dovecot.org/LDA/Postfix

suggests passing the extension as an argument to the mailbox (-m) 
switch. But this assumes that one wants a one-to-one direct mapping 
between extensions and mailbox names.

I suppose I could switch from:
-d ${user}@${nexthop}
  to:
-d ${recipient}
and add even more complexity to the SQL to parse out the extension. But 
I'd rather avoid embedding the extension separator character in the SQL, 
and would rather eliminate it from the database as well.

Maybe Sieve is the answer. I presume the argument to -m could be 
inspected in a Sieve script, which could then map to the desired target 
mailbox. The non-IMAP users (like postmaster at example.com) could be 
reimplemented as aliases that redirect to a user+ext at example.com address 
for the user responsible for that domain. But using Sieve scripts for 
extension to mailbox mappings seems far less elegant than using a database.

Maybe I need to wedge a wrapper program between Postfix and deliver that 
will accept the address components from postfix, perform the lookup, and
convert the results into a -d argument pointing to the "owner" (see 
table above) and a -m argument pointing to the mailbox.


On a side note...

When I used this command in postfix's master.cf:
/usr/lib/dovecot/deliver -d ${user}@${nexthop} -n -m ${extension}

as recommended in the wiki, the following error gets logged:

   ... status=bounced (command line usage error. Command output: Fatal:
   Unknown argument: -d Usage: deliver [-c <config file>]
   [-d <destination user>] [-m <mailbox>]                [-f <envelope
   sender>] )

which doesn't make a lot of sense. I suspect what it is trying to say is 
that it didn't like something about the argument to the -d switch. Maybe 
that should be ${domain} instead of ${nexthop}.

(What's up with all the whitespace between "[-m <mailbox>]" and "[-f 
<envelope sender>]"? Also, might be a good idea to have deliver always 
write errors - even startup errors - to the specified logging facility, 
given that not all programs that launch it will report what it writes to 
STDERR. I notice most postfix commands - even the ones people interact 
with - log errors.)

I ended up switching to -d ${recipient} to emulate the old maildrop 
setup for now, and that worked fine.


The sample dovecot.conf has:

   ## dovecot-lda specific settings
   ##
   # socket listen {
   #   master {
   #     path = /var/run/dovecot/auth-master

I uncommented the first two lines, but left the 'path' directive 
commented, under the assumption that the comment was showing the 
default. But without it uncommented, the socket doesn't get created. 
This is an inconsistency compared to how defaults are shown elsewhere in
dovecot.conf.

  -Tom