Kurt Seifried, [email protected], Copyright Kurt Seifried 2001


Most, if not all the readers of this column run a mail server, and more then likely it is running Sendmail. In all fairness Sendmail is a damn good MTA (Mail Transfer Agent), Eric Allman originally wrote it with one main goal in mind: the mail must get through. Unfortunately, when Sendmail was originally written security wasn't a major concern on the Internet and it shows. Sendmail runs almost exclusively as the root user on most systems, meaning any flaws are potentially very serious. In addition to this Sendmail isn't very good at handling high loads. New mailers, such as Postfix, Zmailer, and Qmail are several times faster then Sendmail on the same hardware. Until recently most of the alternative mailers to Sendmail were not drop-in replacements, to replace Sendmail was a painful task, and the new software typically behaved differently then Sendmail. Postfix was designed from the start to address all these problems.


Postfix does not run exclusively as root, instead a master program (called "master") runs as root and spawns off processes to handle incoming, outgoing and local mail delivery as needed. Using a series of modular components, each task is handled by a separate program (which makes auditing it easier), for example outgoing email is dumped into the queue directory by your software, where "pickup" gets it and hands it to "cleanup", which hands it to "trivial-rewrite" which handles the headers, and finally is given to "smtp" if bound for a foreign system. Postfix is also much easier to setup for a chroot'ed environment than Sendmail is, simply make a few edits to the file (typically in /etc/postfix) and Postfix will run chroot'ed in its defined queue directory (usually /var/spool/postfix). It is also possible to set process limits for individual portions of postfix, again in the file. You can also easily set which user the Postfix programs run as, typically it defaults to "postfix" (a user similar in concept to the "nobody" user for apache), which has access to certain queue directories. Another major benefit of Postfix is the clarity of the configuration files, if you haven't yet looked at the guts of a file I would recommend that you don't (O'Reilly has an 800 page book on Sendmail).

The file:

# ==========================================================================
# service type	private	unpriv	chroot	wakeup	maxproc	command + args
# (yes) (yes) (yes) (never) (50)
# ==========================================================================
smtp	inet	n	-	n	-	-	smtpd
pickup	fifo	n	n	n	60	1	pickup
cleanup	unix	-	-	n	-	0	cleanup
qmgr	fifo	n	-	n	300	1	qmgr
rewrite	unix	-	-	n	-	-	trivial-rewrite
bounce	unix	-	-	n	-	0	bounce
defer	unix	-	-	n	-	0	bounce
smtp	unix	-	-	n	-	-	smtp
showq	unix	n	-	n	-	-	showq
error	unix	-	-	n	-	-	error
local	unix	-	n	n	-	-	local
cyrus	unix	-	n	n	-	-	pipe
	flag=R user=cyrus argv=/usr/cyrus/bin/deliver -e -q -m ${extension} ${user}
uucp	unix	-	n	n	-	-	pipe
	flags=F user=uucp argv=uux -r -n -z -a$sender - $nexthop!rmail ($recipient)
ifmail	unix	-	n	n	-	-	pipe
	flags=F user=ftn argv=/usr/lib/ifmail/ifmail -r $nexthop ($recipient)

Replacing Sendmail

As stated before, replacing Sendmail with another MTA used to be a very painful task, many mailers approached the "problems" in Sendmail by doing things quite differently and require a pretty extensive overhaul of the mail system, and a very different set of configuration files. With Postfix you can use most of your existing configuration files (such as access, aliases, virtusertable, etc.) simply by defining them appropriately in the file. In addition to this Postfix behaves like Sendmail, you can invoke it using "sendmail", get a listing of the mailq with "mailq", feed it pretty much the same options and so forth. I find a typical Postfix conversion takes around 10 minutes for most sites (assuming you use an RPM and don't have to compile it), and have yet to encounter any major disasters (although I have found several small to medium sized glitches). 

New problems

Of course replacing one software package with another will solve certain problems, and create new ones, to which Postfix is no exception. Partly due to it's design as a secure MTA you may experience some minor problems with Postfix. The best example is email to root, Postfix, by default, does not like to deliver email with elevated privileges (necessary to send email to root typically). You will need to define an alias for root in the aliases file, such as: "root: someuser". This also leads to problems with several mailing list packages, especially SmartList, which by default does all sorts of funky things that Postfix will not stand for. In any case I would advise switching to Majordomo, it is easier to configure and maintain via email for owners of mailing lists.


A major issue with Sendmail used to be scalability, and even now the default configuration tends to require significant tuning to get fast performance. Another issue is sites are starting to handle email for 10's of thousands of users on a single server, and using flat text files for configuration details such as mapping outgoing user email addresses (to make bob appear as from [email protected] for example). This file could become several hundred thousand lines long in the future and slow the system down. Postfix can however be integrated with a database backend (currently MySQL is supported) to host it's configuration files, which not only scales better but allows you to do things such as dynamically blocking email to and from certain sites, and allow users to administrate their own accounts more easily (a database beats a flat file any day). 


The IBM Open Source License is surprisingly liberal. Previous versions had a rather ugly termination clause, which prevented wide spread acceptance of Postfix, however this has been removed and Postfix is now "safe" to use. You can distribute the software, develop it, make changes and so forth, the only catch being that you must contribute any changes back to IBM (rather reasonable since they paid Wietse to develop it). 


Some of you are probably wondering why I haven't mentioned Qmail yet, or written an article "Qmail - the Sendmail replacement". I have tested Qmail, and used it for a while, in general I found (and several sites I communicated with, one being a large Linux vendor) Qmail to be very painful to configure and maintain. In addition to this the Qmail license is very unclear, and doesn't even ship with the software. To quote the author: "If you want to distribute modified versions of qmail (including ports, no matter how minor the changes are) you'll have to get my approval." This and other issues have hindered Qmail's acceptance.




Technical side of Postfix

This article was written using Postfix snapshot 20001030. Since then several things (like virtual) have changed, making some points in this article incorrect.


This is probably the best way of restricting incoming email to valid email accounts only. Let's assume you have a decent-sized corporate LAN based on Windows and are using Exchange server for email. Exchange can only validate incoming email based on the domain, not the user, and since it will attempt to deliver the email for 48 hours, your system can get quickly clogged up - with no easy way to clean it out. Place your Exchange server behind a firewall so no one on the Internet can connect to it directly, and then place a Postfix server on the public side. Add this to your

smtpd_sender_restrictions = check_recipient_access

Then in your /etc/postfix/access-inbound file, simply put,

[email protected]	OK		REJECT

You will also need to create the hash file. The following command will do so:

postmap /etc/postfix/access

You could also use the generic access file, but splitting it up allows for extremely selective access controls. You can then have the mail delivered locally or forwarded to another (internal) system. If an email is sent to an email address not listed specifically, and the domain is covered by a reject rule, the sender will receive an email with an error like,

The Postfix program
<[email protected]>: 
host[] said: 554
<[email protected]>: 
Recipient address rejected: Access denied

You can also specify a custom error such as "useraccount does not exist." However, a spammer could theoretically use this to build an address list by simply testing all the email addresses - those that generate an error message do not exist, and those that do not generate one do exist.


This feature can be used as intended, for having virtual users (i.e., if you handle multiple domains and more than one wants a [email protected] email address) as well as for protecting internal servers.

virtual_maps = hash:/etc/postfix/virtual

Then in your /etc/postfix/virtual simply remap email:

[email protected]	[email protected]
[email protected]	[email protected]
[email protected]	[email protected]		bounce-local

You will also need to create the hash file:

postmap /etc/postfix/virtual

If incoming email does not match a virtual user, mapping it is sent to another address, in this example a local user called "bounce-local." It may then be blackholed to /dev/null, simply bounced (no such user), or sent to an admin email account - where it is deleted or forwarded to the correct person, if an obvious typo was made in the address, or whatever. You can also use a database backend instead of a hash or dbm file. This is definitely the way to go for large installations. See man virtual(5) for more information. Simply use the same machine as the outbound mail server (i.e., smart host) and it will rewrite the email addresses outgoing from to (or whatever you want).


This keyword lets you allow only local delivery of email to valid users and/or definitions in the aliases map. Thus any email bound for a nonexistent user gets bounced immediately with an error message saying the user does not exist. This is useful if you the postmaster do not want to receive as many error messages. Potentially, however, an attacker could use this to find valid names (anything valid won't generate an error). This is probably most useful for large installations such as ISPs and large corporations. An additional consideration would be to use relocated_maps. Simply put this in your

local_recipient_maps = $alias_maps unix:passwd.byname

This will accept any email defined in aliases or for user accounts in the password database.


This is an ideal feature for large companies that want to remap users without too much trouble. It allows you to specify the original email address and the new email address. For example, user joebob goes to another ISP instead of forwarding all their mail, which results in senders not realizing it has changed. They get an error message specifying the new email address. This can be used for wholesale domain moves as well. Simply add the following to your

relocated_maps = hash:/etc/postfix/relocated

Then put the email address, username or domain name, followed by the email address or domain name, and anyone sending email in will get a nice error telling them the new address. See man relocated(5) for more information.

Blocking Spam

This is probably one of the favorite features for administrators; under Postfix it is trivial to implement. Simple add the following to your

maps_rbl_domains =, 

The rbl is the real-time blackhole list, basically a list of known spammers and open relays that spammers use. The dul is the dialup list. Generally speaking, you shouldn't be receiving to much email directly from dialup users, i.e., people using their own servers. However, quite a few legitimate users do set up their own email servers for use on dialup links, and blocking them inadvertently may be a problem.

Database Backends

One thing I really love about Postfix is the ability to use databases instead of flat text files or hash and dbm files. Currently only MySQL is supported, but that is more than sufficient for most users. You must first compile in support for MySQL. See the "MYSQL_README" file for more information on this. Then you simply create a table in the MySQL database of usernames, virtual mappings or whatever. The cool thing is, this allows for very efficient sharing of configuration files between servers; and since you can specify multiple MySQL servers, you can replicate the database and avoid a single point of failure as well as being able to vary the order the databases are listed in on various servers, as a simplistic form of load balancing.

MySQL has plans to add database replication to its available features. For now, you will have to create your own solution, such as a simple tool that connects to and updates all the databases at once. The configuration is reasonably simple. In your, put something like,

alias_maps = mysql:/etc/postfix/

And then in the, put

user = someone
password = some_password
dbname = customer_database
table = mxaliases
select_field = forw_addr
where_field = alias
additional_conditions = and status = 'paid'
hosts =

This would allow you to have a table called "mxaliases" in the database called "customer_database," where the field "forw_addr" matches and where the "status" field is set to "paid" (so you can easily block email to customers who don't fork over). The really useful thing is that you can easily update configuration files on the fly. As soon as it is updated in the database, it's ready to go. Additionally, this makes giving users control over their own accounts much easier. A Web hosting provider could easily let people handle virtual user mappings on their own domain, for example, through a cgi interface.


Transport Layer Security is to email as SSL is to Web browsing. TLS allows you to encrypt email transfers from server to server, but more importantly, it allows you to add authentication to the mail server. Instead of having to allow access based on IP and hostname, you can use usernames and passwords. That way people can connect securely from off-site - while using dialup on the road - and spammers are not able to use you as a relay.

There is an add-on TLS package for Postfix (see URL at bottom) available from Germany. (Germany is very pro-encryption; the federal government has even gone so far as to sponsor GnuPG development.) TLS is becoming more common now that the RSA patent has expired. Red Hat 7.1 ships Postfix configured for TLS out of the box.

Regular Expressions

Postfix supports the use of regular expressions for header rewriting and other neat tricks. There's support for basic regex (regexp), and support for perl-compatible regular expressions (PCRE). The first are simpler. To use them, just put this in your

header_checks = regexp:/etc/postfix/header-checks

And then add rules to your header-checks file; the target can be REJECT, OK or a custom error.

/^Subject: Make money fast/		REJECT
/^X-Mailer: Microsoft Outlook Express/	REJECT

See man regexp_table(5) for more information. You can also use PCRE by simply specifying pcre: instead of regexp: in your The rules are basically the same, except that the syntax used for pattern matching is a bit more advanced.

/^[email protected](?!my.domain).*$/ 550 Stick this in your pipe $0

See man pcre_table(5) for more information.


Postfix is ideal for large installations, with its database backends and extremely tight control of mail delivery. Additionally, it supports numerous security features, such as TLS and even the ability to specify which users are allowed to send mail off-site and which aren't - again, in a very selective manner. Qmail has quite a few of these features, but has one significant problem: The license makes it extremely difficult to distribute in a binary format, which is what most people want.

Postfix, on the other hand, comes under the IBM Public License, which is surprisingly considerate to end users. My real favorite is the simplicity of configuration. An average is under 30 lines of configuration directives. Hopefully more vendors will start shipping Postfix with their distributions.


Reference links:

Postfix: -

Other Sendmail alternatives: (GPL) (license is unclear)