I've been doing some tests with normal (foo@domain), catch-all
(@domain), and plussed (foo+foo@domain) addresses, with the virtual(8)
delivery agent, and virtual_alias_maps and virtual_mailbox_maps.
The idea is to check the setup of users who want both redirection and
delivery to a local mailbox (with Dovecot, and Maildir format).
This worked as I expected, except in one case: when mail is sent to an
unknown recipient, and the catch-all setup is used, it's not possible to
both redirect the incoming mail, *and* have it delivered to a local
mailbox. Is this expected? It looks like a bug, because the destination
MTA (the one that receives the redirect) gets a badly-formed RCPT.
Tests as follows, with the virtual_alias_maps file being 'valias', and
the virtual_mailbox_maps file being 'vmailbox', and the local domain
being example.com:
(1) Mail to known user '[email protected]':
valias: "[email protected] [email protected], [email protected]"
vmailbox: "[email protected] example.com/foo/"
This works: external.org gets the mail, and Dovecot also gets the mail
from mailbox 'foo/'.
(2) Mail to unknown user '[email protected]':
valias: "@example.com @example.com, [email protected]"
vmailbox: "@example.com example.com/foo/"
This fails: external.org doesn't get the mail, but Dovecot does get the
mail from mailbox 'foo/'.
The mail log at external.org shows that Postfix did try to redirect the
mail, but sent it to an invalid address:
Recipient address rejected: User unknown in virtual mailbox table;
from=<[email protected]> to=<"[email protected], foo"@external.org>
Second question: with my current setup, an entry is created in
virtual_alias maps even if the user doesn't want redirection, but only
wants local delivery. In other words, if user 'bar' wants local
delivery, then the file entries are:
valias: "[email protected] [email protected]"
vmailbox: "[email protected] example.com/bar/"
This works, and doesn't seem to cause a problem. I don't really want to
change the software to remove this (unnecessary) entry in valias. Are
there likely to be any problems with this?
