Mick examines sendmail's security controversies and builds an SMTP gateway for handling internet mail.
Ah, sendmail. You either love it for being so versatile and ubiquitous, or you hate it for being bloated, complicated and insecure. Or perhaps you're a complete newcomer to the e-mail server game and would like to give sendmail a try (after all, sendmail is arguably the most popular open-source software package on the Internet).
Well, contrary to popular belief, sendmail isn't a total loss where security is concerned, nor does it require learning the arcane syntax of sendmail.cf (although hardcore sendmail gurus do indeed master it). This month we examine these and other sendmail security controversies, using sendmail's handy m4 macros to rapidly build a secure but functional Simple Mail Transport Protocol (SMTP) gateway to handle internet mail.
Sendmail is one of the most venerable internet software packages still in widespread use. It first appeared in 4.1c BSD UNIX (April 1983), and to this day it has remained the most relied-upon application of its kind. Among message transfer agents (MTAs), sendmail is the great workhorse of the Internet, transferring e-mail between networks dependably and (to end users) transparently. But sendmail has both advantages and disadvantages.
On the good side, sendmail has a huge user community, with the result that it's easy to find both free and commercial support for it, not to mention a wealth of electronic and print publications. It's also stable and predictable, due to its maturity.
On the negative side, sendmail has acquired a certain amount of cruft (old code) over its long history, with the result that it has a reputation of being insecure and bloated. Both charges are open to debate, however. It's true that it has had a number of significant vulnerabilities over the years. However, these have been brought to light and fixed rapidly.
As for the “bloatware” charge, it's true that sendmail has a much larger code base than other MTAs such as qmail and Postfix, and a larger RAM footprint too. This probably has at least as much to do with the fact that sendmail is monolithic (one executable provides most of sendmail's functionality) as it does with cruft. Indeed, sendmail's code has been scrutinized so closely by so many programmers over the years that it's a little hard to believe that too much cruft (i.e., code that is strictly historical and obsolete) has survived intact over the past 20 years.
It's much more useful to observe that being monolithic, sendmail must run as root if any portion of its required functionality does, i.e., writing mail to multiple users' home directories. For this reason, sendmail can run only as an unprivileged user on systems on which it's to act solely as an e-mail relay or gateway.
Sendmail also is criticized for its complexity. The syntax of its configuration file, sendmail.cf, is non-instinctive to say the least—in my opinion, its difficulty ranks somewhere between C programming and regular expressions. As with these, this is because of how powerful sendmail is (though many of us do wish sendmail used C, regular expressions or some other standard configuration language in sendmail.cf rather than its own equally complex but much more proprietary syntax). Nowadays, though, this point is largely moot. Recent versions of sendmail can be configured via m4 macros, which provide a much less user-hostile experience than editing sendmail.cf directly.
Regardless of one's opinions on sendmail, it's unquestionably a powerful and well-supported piece of software. If sendmail's benefits are more compelling to you than the negatives, you're in good company. But you'll be in even better company if you learn to run sendmail securely.
As mentioned earlier, sendmail is monolithic in that it does all its real work with one executable, sendmail. Sendmail has two modes of operation: it can be invoked as needed, in which case it will process any queued mail and then quit; or it can be run as a persistent background dæmon.
Dæmon mode is required only when sendmail's role is to receive mail from external hosts; if all you use sendmail for is sending mail, you shouldn't run sendmail as a dæmon, and in fact you can probably stop reading now, because sendmail really doesn't need any customization to do this unless you wish to run it chroot-ed.
The way sendmail works, then, depends on how it's being run. If it's running as a dæmon (i.e., with the -bd flag), it listens for incoming SMTP connections on TCP port 25 and periodically tries to send any outbound messages in its queue directory /var/spool/mqueue. If it's being invoked on the fly, it attempts to deliver the outbound message it's been invoked to send and/or checks /var/spool/mqueue for other pending outbound messages.
Before we go any further, I should make clear what we're about to build. I've chosen the SMTP gateway scenario because it's such a common role for sendmail and because it's so dependent on good security (in most organizations, publicly accessible mail servers face scarier if not more numerous threats than internal mail servers do).
An SMTP gateway typically needs meticulous attention to privilege levels, file permissions and in general, only as much enabled functionality as is needed to route mail. On such a host, sendmail should run as an unprivileged user where possible; it should run chroot-ed (in a subset of /) at least when writing files, and it should be configured to relay mail only for your organization, not for spammers.
It takes very little tweaking to harden sendmail on Red Hat 7 for SMTP gateway use and only a little more on SuSE and other distributions.
I can state with absolute certainty that your Linux distribution of choice includes one or more packages for sendmail. Whether it's presently installed on your system and whether it's an appropriate version for you to use, however, is another matter.
If you use an RPM-based distribution (Red Hat, Mandrake, SuSE, etc.), you can see whether you've got sendmail installed and which version by issuing the command rpm -qv sendmail. Note that Red Hat and its derivatives split sendmail into three packages: sendmail, sendmail-cf and sendmail-doc. SuSE uses a single package, sendmail.
So, what version should you run? As of this writing, the latest version of sendmail is 8.12.2. Red Hat 7 and SuSE 7, however, still support variants of sendmail version 8.11. As far as I can tell, there's nothing wrong with sticking with your distribution's supported sendmail package if it's version 8.11.0 or higher. There were no major security problems in the 8.10 or 8.11 releases; 8.11, in fact, was a “features” release: rather than being released to patch security holes, it was released because the sendmail team had added support for TLS encryption and for the SMTP AUTH extension to SMTP.
If you've got the time and/or inclination, though, it's never a bad idea to compile and install the latest stable version. For sanity's sake, I'll assume for the remainder of this article that you're using sendmail 8.10.0 or higher (unless otherwise noted).
Debian GNU/Linux v2.2 (Potato) still supports sendmail v.8.9.3. Although this is a stable and apparently secure release, it's now two major versions old (if one considers the second numeral to represent a major version, which I do because the first numeral has been eight for half a decade). In addition, 8.9.3 doesn't support TLS or SMTP AUTH.
If you want TLS or SMTP AUTH, or are uncomfortable running an older version, you always can uninstall the package, download the latest source code tarball from www.sendmail.org and compile and install sendmail from source. The source code tarball is well documented and compiles easily under Linux, assuming you've got a working gcc installation.
Once you've installed sendmail, either in the form of a binary package from your distribution or from a source code tarball you've compiled yourself, you've still got a couple of tasks left before you can configure and run the sendmail executable as a dæmon.
If you're a SuSE user, become root if you aren't already. Next, open /etc/rc.config with your text editor of choice and set the variable SMTP to “yes”. This is necessary for sendmail's startup script in /etc/init.d to run at boot time.
In addition, you need to edit the file /etc/rc.config.d/sendmail.rc.config so that the variable SENDMAIL_TYPE is set to “no”. Doing so essentially will disable SuSEconfig's use of /etc/rc.config.d/sendmail.rc.config, which in other circumstances can be set up to generate a simple sendmail configuration automatically. We're going to set up an SMTP gateway/relay, which is a bit beyond the scope of sendmail.rc.config. But if your host is to act only as a simple SMTP server for its own local users, it will probably suffice to edit this file (having first set its SENDMAIL_TYPE variable to “yes”); if so, you'll find sendmail.rc.config's full documentation in /etc/mail/README.
After editing rc.config and sendmail.rc.config, run SuSEconfig. This will propagate the changes you just made to rc.config and sendmail.rc.config. To start the dæmon you can enter the command /etc/init.d/sendmail start, but I recommend you wait until sendmail is fully configured before you do so.
If you're a Red Hat user, you have only one step prior to configuring sendmail: edit /etc/sysconfig/sendmail so that the variable DAEMON is set to “yes”. This will tell the startup script /etc/init.d/sendmail to start sendmail as a dæmon at boot time.
And now, at last, we configure sendmail to act as our domain's SMTP gateway. What follows applies to any installation of sendmail 8.9 or higher (you shouldn't in any circumstances run 8.8).
The simplified method of generating sendmail configurations (sendmail.cf and accompanying files) consists of these steps:
Enable needed features in sendmail.mc.
Set up domain-name masquerading, if needed, in sendmail.mc.
Run m4 to generate sendmail.cf from sendmail.mc.
Configure delivery rules by editing mailertable.
Configure relaying rules by editing access.
Configure local user-aliases in aliases.
Convert mailertable, access and aliases to databases.
Define all local hostnames in local-host-names.
The first and probably longest task in setting up an SMTP gateway is generating /etc/sendmail.cf. This is done most easily by editing /etc/mail/sendmail.mc (or on SuSE systems, /etc/mail/linux.mc—it may be different on other distributions).
Depending on which Linux distribution you use, configuration information for sendmail.mc can be found in /usr/share/doc/sendmail/README.cf (Red Hat and its derivatives), /usr/share/sendmail/README (SuSE) or some variant thereof. I don't have enough space to describe the syntax of the many settings in this file in detail. We will, however, look at some that are useful for security and for modularizing our configuration a bit.
In addition to sendmail.cf itself, we can configure sendmail to read several other files for configuration information. This is useful for two reasons. First, editing sendmail.cf directly is unpleasant and even regenerating it from sendmail.mc isn't always desirable. Second, if your SMTP gateway has multiple administrators with varying privileges, you may wish to keep sendmail.mc and sendmail.cf locked down but allow other administrators to edit user aliases or mail delivery rules (i.e., /etc/mail/access and /etc/mail/mailertable, respectively).
The most useful external configuration files to enable are mailertable, which defines local mail-delivery rules; virtusertable, which defines virtual domain mappings on a per-user and per-domain level; and access, which defines which hosts may use the server as an SMTP relay.
The sendmail.mc directives for enabling these files are shown below:
FEATURE(`mailertable',`hash -o /etc/mail/mailertable.db')dnl FEATURE(`virtusertable',`hash -o /etc/mail/virtusertable.db')dnl FEATURE(`access_db',`hash -o /etc/mail/access.db')dnl
(Note that the mailertable and access_db features are enabled by default under Red Hat, but that virtusertable must be added manually.)
Each of these lines tells sendmail to reference the specified file (although the access database is called access, not access_db), to use a hash database and the path of the respective file. We'll see how to use these files shortly, but first we've got a few more things to attend to in sendmail.mc.
If your users' e-mail addresses are generic to your domain rather than specific to the hosts they log on to, for example, firstname.lastname@example.org rather than email@example.com, you probably want the From: fields of their outbound e-mail to reflect this. (Receiving e-mail addressed to those generic names requires user aliases—see below.)
Following are some sendmail.mc lines that tell our example SMTP gateway to rewrite the From: fields of polkatistas.org's users in this manner. All the lines in the example below must be added (or uncommented):
MASQUERADE_AS(`polkatistas.org')dnl MASQUERADE_DOMAIN(`.polkatistas.org')dnl EXPOSED_USER(`root')dnl FEATURE(`masquerade_entire_domain')dnl FEATURE(`masquerade_envelope')dnl
The MASQUERADE_AS directive specifies the fully qualified domain name you wish to appear in applicable From: addresses. The MASQUERADE_DOMAIN directive specifies the hosts to which MASQUERADE_AS is applicable. Note that the “.” preceding polkatistas.org indicates that all hostnames with this domain name are to be masqueraded.
EXPOSED_USER specifies a user name for whom the From: address should not be masqueraded. root is a popular candidate for this because e-mail from root often contains alerts and warnings; if you receive one, you generally want to know which host sent it.
The feature masquerade_entire_domain causes MASQUERADE_DOMAIN to be interpreted as an entire domain rather than a hostname; masquerade_envelope applies the masquerading not only to the SMTP header but to the envelope as well.
Four other directives, one logistical and the other three security-related, are shown in Listing 1. The always_add_domain feature is enabled by default under Red Hat and SuSE; use_cw_file and smrsh are enabled in Red Hat but not SuSE; confSAFE_FILE_ENV always must be defined manually.
The always_add_domain feature simply forces the local host's domain name to be appended to any e-mail originating from a host that identifies itself without a domain. For example, if the SMTP gateway receives mail from the user “bobo” on a host identified only as “whoopeejohn”, sendmail will rewrite the From: field to read firstname.lastname@example.org rather than bobo@whoopeejohn (but of course masquerading directives still apply).
The use_cw_file feature tells sendmail to refer to the file /etc/mail/local-host-names for a list of hostnames sendmail should consider local. The file /etc/mail/local-host-names is a text file containing hostnames listed one per line. Suppose our example SMTP gateway receives e-mail not only for the domain polkatistas.org, but also tubascoundrels.net. If our gateway's hostname is “mail”, then its local-host-names file will look like this:
localhost.localdomain mail.polkatistas.org mail.tubascoundrels.net
The third feature enabled in Listing 1 is smrsh, the sendmail restricted shell. This is an important security control that restricts the commands that may be executed from a user's .forward file.
The fourth line in Listing 1 tells sendmail to set sendmail.cf's SafeFileEnvironment variable to, you guessed it, some subdirectory of / that sendmail will chroot to (sort of). Actually, this only will happen when sendmail writes files. If you think about it, though, this 50% chroot makes sense: file-writes are what we're probably most worried about, and creating this sort of chroot environment is a lot simpler than your typical chroot jail (which must contain copies of every file hierarchy, file, executable and device your chroot-ed program needs).
Listing 2 shows a recursive listing (ls -lR) of my example SafeFileEnvironment /var/mailjail.
Sendmail created the files /var/mailjail/var/spool/mqueue/bobo and .../root. Beforehand, I had created the entire chroot jail with only four commands:
mkdir -p /var/mailjail/var/spool/mail /var/mailjail/var/spool/mqueue cd /var/mailjail chown -R mail:mail * chmod -R 700 *
If you're concerned about unsolicited commercial e-mail, there's some good news. By default, sendmail doesn't allow SMTP relaying, a common technique of spammers. This can be disabled in sendmail.mc, but you won't find out how from me. Leave this alone. In addition, you can direct sendmail to reject mail from known sources of spam, per the Realtime Blackhole List (RBL), by adding or uncommenting this line:
FEATURE(`dnsbl')For this to work, however, you need to subscribe to the RBL. See Resources for a link to its home page, where you'll find subscription and use instructions and some important disclaimers. (Note that while RBL subscriptions are free for “Individual/Hobby Sites”, there is a fee-schedule associated with this service.) Using the RBL can block e-mail from some legitimate users as well as from spammers, so proceed with caution.
If you run Red Hat 7.1 or 7.2, there's one more sendmail.mc parameter to check, this time one that needs to be commented out. If your /etc/mail/sendmail.mc contains a line like this:
Then you need to comment it out by appending the string dnl to the beginning of the line. If active, this line will cause sendmail to accept only connections on the loopback interface and not from external hosts. Needless to say, for an SMTP gateway this is undesirable (though it unquestionably enhances security).
Those are the most important sendmail.mc settings for our purposes. There are others relevant to security, especially for nongateway roles (local delivery, etc.). For more information see the README.cf or README file I alluded to at the beginning of this section.
To compile our macro-configuration file into sendmail.cf, we use this command:
m4 /etc/mail/sendmail.mc > /etc/sendmail.cf
If your macro-configuration file's name isn't sendmail.mc, substitute it with linux.mc or whatever your macro-configuration file is called. Sendmail expects its configuration file to be named sendmail.cf; however, it looks for it in /etc, so that part of the command is the same regardless of your distribution or even your version of sendmail.
That was the hard part. Now we only need to tell sendmail what to do with incoming mail, what local hostnames are legitimate and which users, networks and domains may use the SMTP gateway to relay mail not destined locally.
The mailertable is used to define delivery rules. It has a simple syntax, which is described in /usr/share/doc/sendmail/README.cf or /usr/share/sendmail/README depending on your distribution. In a nutshell, each line contains two parts: a destination identifier and an action. The destination identifier matches destination addresses or parts thereof; the action tells sendmail what to do with messages whose destinations match the identifier.
If the identifier begins with a “.”, then all e-mail source addresses ending in whatever follows the dot will match. If it doesn't, then everything following the “@” sign must be identical to the identifier. The e-mail address email@example.com won't match the identifier polkatistas.org but will match .polkatistas.org.
The action takes the form agent:action where agent is either a mailer (defined in sendmail.mc/linux.mc in MAILER() statements) or the built-in agents “local” or “error”. The “local” agent, of course, means the mail should be delivered to a local user, specified after the colon (if nothing follows the colon, the user specified in the message itself will be used).
Below is a mailertable with two different actions:
polkatistas.org smtp:internalmail.polkatistas.org mail.polkatistas.org local:postmaster
In addition to delivery rules, sendmail needs to know which e-mail destinations should be considered synonyms of the local (SMTP gateway's) hostname. These are specified in /etc/mail/local-host-names, one per line:
mail.polkatistas.org weird-al.polkatistas.org 18.104.22.168Finally, we need to define allowed relayers by editing /etc/mail/access. The syntax is simple. Each line contains a source name or address, paired with an action (again, see README.cf or its equivalent on your distribution for details). The action can be RELAY, REJECT, DISCARD, OK or ERROR. In practice, the most useful of these actions is RELAY. Because by default relaying is rejected, REJECT and DISCARD are only useful when defining exceptions to explicit RELAY rules.
Here is a simple access file:
localhost.localdomain RELAY localhost RELAY 127.0.0.1 RELAY 192.168 RELAY
Do you notice the absence of real hostnames in the example above? In this example, the SMTP gateway performs only outbound relays; inbound mail must be addressed to a local e-mail address, and outbound relays must originate from hosts whose IP addresses begin with the octets 192.168 (obviously a noninternet-routable network). I like this technique (using IP addresses) because then I can prevent IP address spoofing with my firewall rules, but I can't prevent forged From: addresses in e-mail (however, your needs may be different of course):
access local-host-names mailertable
SMTP AUTH (in sendmail version 8.10 and later) adds authentication to SMTP transactions, e.g., to determine whether to permit relaying. This is especially useful when systems or users don't run their own MTA but need to send mail, i.e., need to relay outbound mail through a central gateway.
If you're running an SMTP server that relays mail from other domains, you probably want this feature, as it's an important defense against unsolicited commercial e-mail, the perpetrators of which rely heavily on SMTP relays.
There's just one more file that may need tweaking: aliases. This file contains a map of e-mail aliases to user names. Usually an SMTP gateway doesn't need a very granular alias database; to translate entire domains' (or virtual domains') e-mail addresses you're better off using the user database (which, sadly, I don't have space to cover). It's fairly self-explanatory, though, so edit it if you need to.
Now three of the four files we've just discussed, mailertable, access and aliases, actually can't be used by sendmail directly; they must first be converted to databases. The /etc/mail file contains a handy Makefile for this purpose. To use it simply change your working directory to /etc/mail and enter this command:
Make access.db mailertable.db
Note that this won't work for aliases, which has its own utility, newaliases. Run newaliases without any flags to convert your changed /etc/aliases file into a new /etc/aliases.db file automatically.
And that's it for now. There's much I haven't covered, notably the smrsh shell (applicable mainly to local mail delivery, not to gateways). But hopefully I've given you some useful hints and pointers to more complete sources of information. Good luck!