Fixed: http://hg.dovecot.org/dovecot-2.2/rev/9091d0f2d971

And for LDAP: http://hg.dovecot.org/dovecot-2.2/rev/939aa051e3f1

On 6.7.2013, at 1.18, Axel Luttgens <AxelLuttgens@swing.be> wrote:

Hello,

Let's say dovecot.conf contains:

mail_uid = dovemailer mail_gid = dovemailer mail_home = /some/path/%n mail_location = mbox:~/mboxes:INBOX=~/mboxes/inbox

and that the password database query is of the form:

password_query =
SELECT
passwd AS password,
nickname AS user,
mail_home AS userdb_home,
mail_location AS userdb_mail,
WHERE
...

The database initially comes with NULL for both mail_home and mail_location, the goal being to be able to progressively replace legacy settings.

With the above, one gets such entries in the logs upon a pop or imap connection:

auth-worker(11262): Debug: auth(u12345678,127.0.0.1): username changed u12345678 -> john.doe auth: Debug: auth(u12345678,127.0.0.1,<dsNsasbgRQB/AAAB>): username changed u12345678 -> john.doe [...] auth: Debug: prefetch(john.doe,127.0.0.1,<dsNsasbgRQB/AAAB>): passdb didn't return userdb entries, trying the next userdb

and, of course, the userdb_query fails since it isn't supposed to be invoked under such circumstances.

Of course, the userdb_query could be adapted so as to handle pop/imap connections in addition to say, lmtp or doveadm connections, but this would anyway raise the question: why bother with a prefetch database setup?

In fact, it seems that the problem comes from the fact that the password_query returns NULL values (i.e. "do not override dovecot.conf settings") for all userdb_xxx settings even if, technically speaking, it returns such columns.

A slight yet somewhat silly modification of the password_query, such as this one:

password_query =
SELECT
passwd AS password,
nickname AS user,
'dovemailer' AS userdb_uid,
mail_home AS userdb_home,
mail_location AS userdb_mail,
WHERE
...

indeed seems to bring back all the expected behavior: now, the "passdb returns userdb entries" and, for example, the config's mail_home expands to the expected value /some/path/john.doe.

Could it be that the case "userdb_xxx columns returned, even if all with NULL values" has been somehow overlooked in the code? Or am I erring with my interpetation of all those matters?

TIA, Axel