Postfix Revision as of Sunday, 20 December 2015 at 19:56 UTC

Written for CentOS 6.4, Postfix 2.6.6.

• Server is example.com.
• Mail users map to local accounts (i.e., in /etc/passwd).
• DNS looks like this:

 [me@example.com ~]$ dig example.com -t MX +short
 10 mail.example.com.

• Made sure that reverse DNS is set up properly.

Overview

Things were easier to set up after understanding these things, even
cursorily¹.

Mail Transfer - Basic Idea

Lots of formal abbreviations! They are, luckily enough, quite
sensible. Here’s the basic
flow:

   Sender > Server > Server > ... > Server > Receiver
   (MUA)    (MTA)    (MTA)          (MTA)     (MUA)

You can be a bit more granular:

      Sender   >  Server > Server > ... >   Server     | Delivery |  >   Receiver
   (MUA > MSA)    (MTA)    (MTA)          (MTA > MDA)  | complete |     (MRA > MUA)

(MDAs, when local, are also called LDAs.)

This separation of purpose is good since you can use a variety of
applications and
topologies
at each stage. Lot of possibilities. E.g.

   Sender > Postfix > Procmail > Clam Anti-Virus > SpamAssassin > Procmail > Fetchmail > Receiver
   (MUA)    (MTA)     (                       MDA                        )     (MRA)       (MUA)

Open Relays

Where it is not required to (1) authenticate to your server, and/or (2)
be in the same network as the server to send email. This is very bad for
public-facing mail servers. From a simpler time when there were very few
email servers and everybody was nice to each other.

Mailbox Formats

There are quite a few, each
with its own pros and cons. I personally like
Maildir.

Installation

 yum install postfix cyrus-sasl
 ln -s /usr/sbin/sendmail.postfix /etc/alternatives/mta --force

Configuration

The Basics

Postfix ships with sane and secure defaults. Here’s stuff I changed in
/etc/postfix/main.cf

First set the hostname and domain

 myhostname = example.com
 mydomain = $myhostname

Mail from this server will come from this domain.

 myorigin = $mydomain

Accept mail on specified interface² and all protocols (IPv4 and
IPv6³)

 inet_interfaces = all
 inet_protocols = all

This server will think itself the final MTA (in the chain above) for
these domains:

 mydestination = $mydomain, localhost

The server will only trust itself⁴

 mynetworks_style = host

Use the Maildir format for message delivery

 home_mailbox = Maildir/

Change the banner for fun (and no version information)

 smtpd_banner = $myhostname ESMTP Why, hello there!

Now edit [http://www.postfix.org/master.5.html /etc/postfix/master.cf]
to enable the submission port⁵.

   submission inet n       -       n       -       -       smtpd

Uncomment any options ("-o"); we’ll take care of these in later.

Testing

Restart the postfix service. Then from another computer,

   orangebox:~$ telnet example.com 25
   Trying 96.126.123.32...
   Connected to example.com.
   Escape character is '^]'.
   EHLO example.com
   250-example.com
   250-PIPELINING
   250-SIZE 10240000
   250-VRFY
   250-ETRN
   250-ENHANCEDSTATUSCODES
   250-8BITMIME
   250 DSN
   MAIL FROM: testuser@internet.com
   250 2.1.0 Ok
   RCPT TO: me@example.com
   250 2.1.5 Ok
   DATA
   354 End data with .
   Subject: Hello!
   How have you been?
   .
   250 2.0.0 Ok: queued as 7602C72C2
   QUIT

This should:

• Work for a valid user

 MAIL FROM: postmaster@example.com
 250 2.1.0 Ok

• Not work an invalid user

 RCPT TO: nonexistent@example.com
 550 5.1.1 nonexistent@example.com: Recipient address rejected: User unknown in local recipient table

• Deliver a mail to your home folder! The “Maildir” folder will be
  created automagically.

 ~/Maildir
         ├── cur
         ├── new
         │   └── 1377029606.Vca00I4025fM640219.example.com
         └── tmp

But this is done insecurely. Let’s fix that.

Doing things securely

Generate a self-signed certificate⁶

 openssl req -new -x509 \
             -newkey rsa:4096 \
             -days 3650 \
             -nodes \
             -out /etc/pki/tls/certs/dovecot.crt \
             -keyout /etc/pki/tls/private/dovecot.key

 chmod o= /etc/pki/tls/private/dovecot.pem

Now configure Postfix to use these certificates for TLS

 smtp_tls_security_level = may
 smtpd_tls_security_level = may
 smtpd_tls_cert_file = /etc/pki/tls/certs/postfix.crt
 smtpd_tls_key_file = /etc/pki/tls/private/postfix.key
 smtpd_tls_auth_only = yes
 smtpd_tls_loglevel = 3
 smtpd_tls_received_header = yes
 smtpd_tls_session_cache_timeout = 3600s
 tls_random_source = dev:/dev/urandom

Restart Postfix. As always, see /var/log/maillog for any errors.

Now test. Important You have to use the OpenSSL client instead of
telnet from this point on! Watch out for non-zero “Verify return codes”.
Avoid these by downloading the generated “postfix.crt” using it with
your OpenSSL command.

 openssl s_client -starttls smtp \
                  -CAfile /path/to/postfix.crt \
                  -connect example.com:25

Another important warning is to keep the former telnet commands
lowercase. Else, the client will renegotiate every time you type
RCPT TO. OpenSSL can waste your time like
that!

Some restrictions

Stepping throught the telnet output in the previous section, start
adding some restrictions to the client connection⁷:

 smtpd_client_restrictions = reject_unknown_client_hostname, permit

Then the
[http://www.postfix.org/postconf.5.html#smtpd_helo_restrictions HELO]
command

 smtpd_helo_required = yes
 smtpd_helo_restrictions = reject_unknown_helo_hostname, reject_non_fqdn_helo_hostname, reject_invalid_helo_hostname, permit

[http://www.postfix.org/postconf.5.html#smtpd_sender_restrictions MAIL FROM]

 smtpd_sender_login_maps = pcre:/etc/postfix/login_maps.pcre
 smtpd_sender_restrictions = reject_non_fqdn_sender, reject_sender_login_mismatch, reject_unknown_sender_domain, permit

And finally,
[http://www.postfix.org/postconf.5.html#smtpd_recipient_restrictions RCPT TO].
This will allow you to relay messages (i.e. send email to other
domains) if you’re SASL-authenticated.

 smtpd_recipient_restrictions = permit_sasl_authenticated, reject_unauth_destination, permit

Add this to /etc/postfix/login_maps.pcre⁸.

 /^(.*)@example.com$/   ${1}

Test away! You should see good errors like:

 450 4.1.8 askljdas@lksjdklfsdjf.com: Sender address rejected: Domain not found
 450 4.7.1 : Helo command rejected: Host not found
 553 5.7.1 me@example.com: Sender address rejected: not logged in

Now add a way in which you an log in to the server remotely to send
messages through it.

SASL Authentication

Will use Cyrus.
Postfix uses it by
default.
You can see what other libraries Postfix was compiled with support for
as well:

 [root@example ~]# postconf -a
 cyrus
 dovecot

Install Cyrus

 yum install cyrus-sasl

You can then see what authentication methods Cyrus supports:

 [root@example !]# saslauthd -v
 saslauthd 2.1.23
 authentication mechanisms: getpwent kerberos5 pam rimap shadow ldap

Install the appropriate package. Since I’m using plain auth,

 yum install cyrus-sasl-plain

Since we’re dealing with local accounts, let’s tell Cyrus to use
/etc/shadow. Open /etc/sysconfig/saslauthd:

 SOCKETDIR=/var/run/saslauthd
 MECH=shadow
 FLAGS=

Start the service

 service saslauthd start

Make sure it starts when you reboot your server

 chkconfig saslauthd on

Test!

 [root@example !]# testsaslauthd -u testuser -p
secretpassword
 0: OK "Success."

Now tell Postfix to use Cyrus in /etc/postfix/main.cf

 smtpd_sasl_auth_enable = yes
 smtpd_sasl_type = cyrus

Set some security
options

 smtpd_sasl_security_options = noanonymous

Restart the Postfix service. Test:

 [root@toolkit ~]# openssl s_client -starttls smtp -CAfile /path/to/postfix.crt -connect example.com:25
 (Certificate, connection info)
 ---
 250 DSN
 helo example.com
 250 example.com
 auth plain An8o0tjsHojfDausWtzblk4bnZA
 235 2.7.0 Authentication successful

Generate the funky MD5 output with your username and password:

 echo -ne '\000user\000password' | base64

Preventing Spam, Bad Email, and DOS Attacks

Using Blocklists

Change smtpd_recipient_restrictions to add some blocklists⁹ and
other stringent policies:

 smtpd_recipient_restrictions = 
   permit_sasl_authenticated, 
   reject_invalid_hostname, 
   reject_non_fqdn_sender,
   reject_non_fqdn_recipient,
   reject_unauth_destination,
   reject_rbl_client zen.spamhaus.org,
   reject_rbl_client psbl.surriel.com,
   reject_rbl_client bl.spamcop.net,
   permit

Most residential IPs are banned by blocklists, so keep that in mind when
testing your setup:

 554 5.7.1 Service unavailable; Client host [173.29.77.33] blocked using zen.spamhaus.org; 
 http://www.spamhaus.org/query/bl?ip=173.29.77.33

Using SPF

Sender Policy Framework
prevents fake sender addresses from your domain. It’s a great idea and
is something everyone should do¹⁰.

To empower Postfix with SPF, first install some required packages from
EPEL:

 yum install perl-core perl-Mail-SPF --enablerepo=epel

I’m going to try the Perl implementation of
SPF. ¹¹ Download,
extract, move to a good place:

 tar -xvzf postfix-policyd-spf-perl-2.010.tar.gz
 cd postfix-policyd-spf-perl-2.010
 mv postfix-policyd-spf-perl /usr/local/bin/

Now set up /etc/postfix/main.cf. Add to smtpd_recipient_restrictions

 # Other options not shown for brevity
 smtpd_recipient_restrictions = 
   check_policy_service unix:private/policy-spf,
   policy_time_limit = 3600 # Default is 1000; too short¹²

Then add to /etc/postfix/master.cf

 policy-spf unix  -       n       n       -       0       spawn
     user=nobody  argv=/usr/bin/perl /usr/local/bin/postfix-policyd-spf-perl

Restart the service. Send yourself an email from another service Gmail,
and look for SPF output in /var/log/maillog

Some Limits on Interaction

 # Set the time unit
 anvil_rate_time_unit = 120s
 
 # This allows 20 connections per second
 smtpd_client_connection_rate_limit = 2400
 
 # Don't allow more than 100 messages per second
 smtpd_client_message_rate_limit = 12000
 
 # Messages can't be more than 20M in size
 message_size_limit = 2048000
 
 # Can tweak these depending on use
 smtpd_soft_error_limit = 10
 smtpd_hard_error_limit = 20
 smtpd_error_sleep_time = 60

Prevent Abuse

Greylisting is a great
approach to fighting spam. The idea is that spammy mail servers do not
respect the RFC spec that, if an email couldn’t be delivered initially,
they are to re-attempt delivery later.
Postgrey works well for this. By
default, it asks MTAs to attempt redelivery in 5 minutes.

   yum install postgrey
   service postgrey start
   chkconfig postgrey on

This will run on a Unix socket. The next step is to get Postfix to use
it. Edit /etc/postfix/main.cf.

   # Other options not shown for brevity
   smtpd_recipient_restrictions =
       check_policy_service unix:postgrey/socket

I lowered the default wait time to a minute by creating
/etc/sysconfig/postgrey and adding this:

   OPTIONS=$OPTIONS" --delay=60"

Restart Postfix and Postgrey. You’ll see something like this in
maillog to make sure it’s working:

   postgrey[12582]: action=pass, reason=client whitelist, client_name=mail-qc0-f169.google.com, 
       client_address=209.85.216.169, sender=anand.nikhil@gmail.com, recipient=test@example.com

Miscellaneous

“warning: dict_nis_init:”

Disable NIS lookups

 alias_maps = hash:/etc/aliases

“Relay Access Denied”

Usually
something
quite
simple.

Footnotes