First, postfix has a special transport, called ‘relay’, which is the same as SMTP, except that it doesn’t send to the backup MX. In this case, that means it doesn’t send to itself, avoiding loops. The relay transport is the default for all the domains you specify in the relay_domains parameter.

Should the machine have a relayhost defined, you need to disable that for each of the domains you’re a backup MX for. You can do that with transport_maps. In my case, the transport_maps is hash:/etc/postfix/transport_maps. That is a key-value file on which you run ‘postmap’ after you’ve edited it. In it, specify this (the comments explain it):

# When you specify a transport without nexthop, it resets the relay to the
# recipient domain (see man 5 transport). And, when the transport is relay,
# postfix will not relay to the backup MX, to prevent loops back to itself. So,
# because this host has a default relayhost, use the folowwing when you want to
# specify a domain for which we are backup MX: 
# example.com            relay 

The next step you want to do is take measures to prevent mails to non-existent users piling up in the queue. Because spam is sent to the backup MX all the time, it will relay it to the primary, which rejects it. Your backup host will then bounce all kinds of spam mails…

To fix that, we instruct postfix to use recipient address verification. This causes it to probe the primary host to check the address exists (and caches that info) before relaying.

To enable it, do this:

smtpd_recipient_restrictions = permit_mynetworks, reject_unauth_destination, reject_unknown_recipient_domain, reject_unverified_recipient
unverified_recipient_reject_reason = Address lookup failed
# Prevent caching non-existent users, to prevent delivery failures when new users are unknown still.
address_verify_negative_cache = no
# Setting to 550 to make a reject a fatal error, not a defer.
unverified_recipient_reject_code = 550
# The point of a backup MX is to accept mail when the primary is down, so setting this prevents incoming mail being deferred when the address probe cannot be done.
# TODO: find out how to actually implement it, because it doesn't work; permit it not a valid option here.
#unverified_recipient_tempfail_action = permit 

The reject_unknown_recipient_domain prevents probes to domains that don’t exist.

Some extra options, not strictly related to being a backup MX:

smtpd_sender_restrictions = reject_unknown_sender_domain, reject_unknown_helo_hostname
unknown_address_reject_code = 550
# Override the default of 5 days, because the point of a backup MX is to keep it around for a while.
maximal_queue_lifetime = 21d

when you type mailq on a postfix server, the first address listed is the sender, the second the recipient.

Source.

aptitude -P install postfix-policyd-spf-perl

Add this to master.cf (but perhaps change the path to the script):

policy  unix  -       n       n       -       -       spawn
        user=nobody argv=/usr/bin/perl /usr/lib/postfix/policyd-spf-perl

Add this to main.cf, directly below/after reject_unauth_destination (if you do it before, you are an open relay):

check_policy_service unix:private/policy

So:

smtpd_recipient_restrictions = permit_mynetworks,
  permit_sasl_authenticated,
  reject_unauth_destination,
  check_policy_service unix:private/policy
  reject_unauth_pipelining,
  reject_non_fqdn_recipient

The source article has stuff about testing.

• “postsuper -r ALL”. Requeues all messages.
• “postcat [file]“. Views queue files in /var/spool/bla.
• “postqueue”. Deal with queue.

I believe there were more, but I can’t remember them. Wish I had blogged earlier…

First install libsasl2 and configure it. Enable it in /etc/default/sasl.

First make the sasl config file in /etc/postfix/sasl which says:

pwcheck_method: saslauthd
mech_list: PLAIN LOGIN

Then configure the ssl paramters in postfix (the following is deprecated. See aforementioned official postfix docs for good way):

# According to official docs, this should be in one pem file.
smtpd_tls_cert_file=/etc/ssl/certs/ssl-cert-snakeoil.pem
smtpd_tls_key_file=/etc/ssl/private/ssl-cert-snakeoil.key
# This should be smtp_tls_security_level = may, because use_tls is deprecated.
smtpd_use_tls=yes
smtpd_tls_session_cache_database = btree:${queue_directory}/smtpd_scache
# I don't know if this one is also needed.
smtp_tls_session_cache_database = btree:${queue_directory}/smtp_scache
# When TLS encryption is optional in the Postfix SMTP server, do not announce or accept SASL authentication over unencrypted connections. 
smtpd_tls_auth_only=yes

Then enable the three smtps lines in master.cf:

smtps     inet  n       -       -       -       -       smtpd
  -o smtpd_tls_wrappermode=yes
  -o smtpd_sasl_auth_enable=yes

Then you need to do some hacking to get the jailed postfix to access /var/run/saslauthd:

rm -r /var/run/saslauthd/
mkdir -p /var/spool/postfix/var/run/saslauthd
ln -s /var/spool/postfix/var/run/saslauthd /var/run
chgrp sasl /var/spool/postfix/var/run/saslauthd
# Add user postfix to group sasl
adduser postfix sasl

Then you should be good to go. Start all daemons.

poll server user "user" pass "secret" mda "formail -A 'X-Zimbra-To: user@domain.org'| /opt/zimbra/postfix/sbin/sendmail -i -t service@sicirec.org"

The -i tells sendmail to ignore a single dot on a line, because that would normally mean end of mail. The -t is “to” (not the header “To:“).

It is a bit unclear why postfix delivers locally to Zimbra, since doing mail user@ourdomain.org routes through an external SMTP server, which is configured in Zimbra to be used as MTA for outgoing mail. It is configured as ‘webmail MTA’.

I have a zimbra server fooled into thinking it hosts a particular domain. Part of the fooling involves setting a different SMTP server than localhost for all outgoing mail. Luckily, Zimbra can do that.

The downside of that is that when you send mail to “root”, the other SMTP server qualifies it with its domain and the mail appears to be coming from the wrong server.

To fix it, specify this in the /opt/zimbra/postfix/conf/main.cf:

myorigin = example.com

This seems to work without caveats. However, I don’t know if zimbra overwrites this config file at some point.

As always, pick a domain that exists, otherwise a lot of mailservers won’t accept it. You don’t even need an MX record, A or CNAME if enough.