One of the most threatening and irritating problems I deal with as a network administrator is viruses. It seems that the virus threat has increased tenfold over the last couple of years. I have found it increasingly unnerving to see what the newer viruses are capable of doing. It's not even their ability that is most disturbing, it's the ease with which they get into our systems. Recently I spent two hours cleaning the MTX and Navidad viruses off of a user's machine.

The usual first approach that administrators take in virus prevention is to install a desktop virus protection program. That is wise, but it seems to me the most fail-safe way to protect a corporate network from viruses is to prevent their entrance into the system in the first place. By far the most common entry point for viruses, especially macro viruses, is corporate e-mail systems. However, this is usually the most neglected piece in any virus detection solution. The e-mail virus software on the market today is often application-specific, expensive or both (not to mention unreliable). Being a mid-sized business, we did not have it in our budget this year to buy an antivirus suite off the shelf, so I turned to Linux and open source. I did find a couple of interesting projects on the Internet that might meet our needs, but I decided to write my own instead. I wanted our system to be extremely easy to follow and easy to extend without a user having to know C or be a Perl guru. Also, I wanted the system to take advantage of the power of the utilities that every base Linux installation usually has. All of this would ensure that another could administer the system in my absence and that it would be portable.

The basic outline of the system consists of using Bash scripts, metamail, grep, the Obtuse Systems' smtpd product, Samba and a command-line virus scanner. A flowchart-style diagram can be found in Figure 1. The Obtuse Systems' SMTP store and forward package is freely available at www.obtuse.com/smtpd.html. The current version as of this writing is version 2.0. The virus scanner I chose was McAfee Virus Scan for UNIX/Linux, but there are quite a few others to choose from. Some are free and some are not. Do make sure you choose one that sets exit status codes based on what it finds and that is well supported with frequent signature updates.

The system can be set up on an existing Linux firewall or a separate machine, if you do not already have a Linux firewall in place. If you choose to set up a separate machine as the e-mail firewall, it doesn't have to be very powerful. A 200MHz 586 with 32MB of RAM would be plenty. Our network is attached to the Internet via SDSL and is protected by a Mandrake Linux machine running IP masquerade. This design made it easy to set up the system on our current firewall machine. The internal e-mail system used is not important as long as it speaks SMTP or ESMTP. In our case, we use Novell's Groupwise product. All SMTP traffic (port 25) should be redirected from the SMTP port on the firewall to the machine you have set up as your e-mail firewall on the inside (or to the firewall itself in our case). Now let's move on to the actual setup.

Our first step is to set up the filesystem hierarchy. A diagram of this can be found in Figure 2. We will set up our system in a directory named /var/spool/smtpd. If you have an average e-mail volume of over 25,000 e-mails a day, I suggest partitioning a separate hard drive dedicated to the e-mail firewall and mounting it in this directory. The base directory will be /var/spool/smtpd. In this directory we will make five subdirectories named incoming, outgoing, etc, bin and quarantine. To start, become root and issue this command:

mkdir -p /var/spool/smtpd/ {etc,bin,incoming,outgoing,quarantine}

Next, change the permissions on the whole hierarchy to be accessible only by the uucp user, since all the programs will be run as this user. These commands will do the trick:

chown -R uucp.uucp /var/spool/smtpd
chmod -R 700 /var/spool/smtpd

Now we will set up the first component of the system. You will need to grab the smtpd package from the Obtuse Systems' web site mentioned earlier. Switch to the download directory and unpack the tarball by issuing the command:

tar -zxvf smtpd-2.0.tar.gz

Now change to the smtpd-2.0 directory and edit the Makefile to reflect the following changes:

SPOOLDIR = /var/spool/smtpd
SPOOLSUBDIR = incoming
POLL_TIME = 300
PARANOID_SMTP = 1
JUNIPER_SUPPORT = 0
CHECK_IDENT = 0

What we want is for smtpd to store mail in the incoming subdirectory and smtpfwdd to read mail from the outgoing directory. To make this possible we have to insert a line into the file smtpfwdd.c. Insert the following two lines at line number 75:

// Pull mail from the outgoing subdir.
#define SPOOLSUBDIR "outgoing"

Finish up by compiling and installing the package with these commands:

make
make install

Next, we need to populate the /var/spool/smtpd/etc directory with some files to allow smtpd to operate correctly. Copy the resolv.conf file from /etc to the /var/spool/smtpd/etc directory, and then copy the file localtime from /etc to here. You also should copy the antirelay_check_rules_example file from the smtpd-2.0 distribution directory to /var/spool/smtpd/etc and rename it smtpd_check_rules. You can look on the Obtuse Systems' web site for instructions on how to create check rules if you need this. I would at least put in an antirelay rule to start. To have the smtpd program start automatically, place an entry in your /etc/inetd.conf file similar to this:

smtp    stream tcp nowait root /usr/local/sbin/smtpd    smptd

This line should replace any other SMTP entries that may be there. We will need to start smtpfwdd manually from /etc/rc.d/rc.local (or whatever your local rc file is). So go ahead and add an entry like this to your startup file:

### Start the smtpfwdd forwarding daemon
/usr/local/sbin/smtpfwdd

Finally, you will need to shut down any other mail transfer agent (MTA) that might be running. These would include Postfix, sendmail, qmail and the like. On Red Hat systems you can do this simply by running the setup utility, unchecking any MTAs from the system services menu and rebooting. Be aware that some MTAs run as children of other processes, such as Postfix, and simply can't be killed directly. If you can reboot the machine, then go ahead and do that now. If not, then you can issue these commands to get smtpd and smtpfwdd up and running:

kill -HUP inetdpid
/usr/local/sbin/smtpfwdd

where inetdpid is the Process ID for inetd.

Once the smtpd and smtpfwdd dæmons are running, you can test the setup by launching a Telnet session on port 25 (the smtp port) of the e-mail firewall, like so:

telnet email.firewall.com 25

where email.firewall.com is the hostname of your e-mail firewall.

You should get a prompt that says

220 email.firewall.com SMTP ready,
Who are you gonna pretend to be today?

If you get any other prompt, then you probably forgot to shut down a running MTA on the server. A ps -e should bear this out.

Let's look at what we have so far. At this point you should have a machine with the smtpd dæmon running and accepting e-mail. All received e-mail is being stored in the /var/spool/smtpd/incoming directory as simple ASCII text files. To test this out, try issuing these commands:

telnet email.firewall.com 25
helo firewall.com
mail from: joe@firewall.com
rcpt to:
data
This is a test.
.
quit

where you@domain.com is your e-mail address. Be sure to end your message body with a period on a line all by itself. This tells the server you are through sending text.

If all went well, you should now see a text file in the /var/spool/smtpd/incoming directory containing the contents of our SMTP session. Now move the file into the /var/spool/smtpd/outgoing directory. In a few minutes the file should disappear, and you should receive the e-mail in your main mailbox. smtpd saves e-mails as text files named smtpd??????, where ?????? is a randomly generated message ID. smtpfwdd reads these text files and forwards them to the destination server. After a file is successfully forwarded, it is deleted.

So how does the mail get from incoming to outgoing? Here is where our e-mail scanning script will come into play. The script will search through each e-mail file in the incoming directory one by one and look for viral content. If the file doesn't contain a virus, it is moved to the outgoing directory. If it does, then it gets moved to the quarantine directory. It's as simple as that. But before the mail scanning script will work, we need to have a functional piece of virus detection software installed. Let's focus on that now.

What we need is a command-line based virus scanner. I chose McAfee Virus Scan for UNIX/Linux, so that is what I describe here. The McAfee product has a intuitive list of exit status codes, so it is easy to integrate into shell scripts. You can get this product by going to www.nai.com and using the e-store. After you install the product you will need to make sure it is executable by uucp. To do this, go to the /usr/local/uvscan directory and execute this command:

chmod -R 755 *

Now test it out and make sure it works. Become uucp and issue this command:

/usr/local/uvscan/uvscan --version

If the test is successful, then we can move on to updating the virus signature files. The signature files (or definition files as they are sometimes called) are the lifeblood of any virus scanner. Therefore, we will want to automate the process of updating them on a frequent basis. It is important to check for these updates at least biweekly, since new versions usually are put out at least three times a month. To do this we must first create a Bash script called sigupdate and place it in the /var/spool/smtpd/bin directory. As always, make sure the script is owned and executable by the uucp user. The contents of the script are shown in Listing 1 and should be easy to follow. The sigupdate file should be set to run once a week, preferably at a low-volume time of day. We will do this later by inserting a crontab entry for it. You also will need to create a file in the uucp user's home directory called .netrc. Edit this file to look as follows:

machine ftp.nai.com
login anonymous
password admin@domain.com
macdef init
cd pub/antivirus/datfiles/4.x
bin
prompt
mget dat-*.tar
close
bye

Listing 1. sigupdate Script

The .netrc file controls sessions for predefined FTP hosts. This is the best way to make our FTP session automatic. You can reference the FTP man page for help with the netrc syntax. Go ahead and run sigupdate as soon as it, and .netrc, are ready. After the update finishes, execute this command:

/usr/local/uvscan/uvscan --version

Look at the virus data file creation date; it should be recent, usually no more than a month old. If it is not, you will need to check that sigupdate is running properly from the command line.

Now let's move on to the main mail scanning script. We will name it scanmail, and it will be stored in the /var/spool/smtpd/bin directory. This script will do all of the direct manipulation of the e-mail text files created by smtpd. Create the file and make it owned and executable by uucp. You will find the complete script in Listing 2 [available at ftp.linuxjournal.com/pub/lj/listings/issue92/4882.tgz]. It is heavily commented so I will just give you an overview here.

Starting on line 19, scanmail first switches to the incoming directory and stores the e-mail filenames in an array variable. It then loops through each filename and uses grep to search for specific patterns within the files. Two passes are made through each e-mail. The first pass greps for attachments that are obviously bad. The patterns for this pass are stored in a file named matches.bad, which we will create later. If grep hits a match, that file is quarantined and an e-mail is sent to the administrator containing a timestamp, the name of the file, who it came from and who was to receive it.

If no match was made, a second pass is made on the e-mail. This time grep uses a file named matches.doc to search for attachments that could contain macro viruses or embedded viruses. If a match is made this time, the attachment is decoded into a dynamically generated temporary directory using the metamail program. The temporary directory is created by using the name of the e-mail with a “_d” concatenated onto the end. The contents of the temp directory are then scanned with our command-line virus scanner. If a virus is detected here (by checking the exit status of the scanner), scanmail quarantines the e-mail and the attachment(s) and sends the administrator an e-mail alert. A courteous e-mail also is sent to the sender saying that they may need to check their system for viruses and giving the name of the virus detected.

At this point, if no viruses have been found, the e-mail is moved over to the outgoing directory where it is delivered to the internal e-mail server by smtpfwdd. smtpfwdd scans the outgoing directory for mail to deliver every five minutes.

The next thing to do is set up our filename lists that scanmail will use to search for suspicious attachments. scanmail uses the grep tool to search through files. We will take advantage of the -f switch to make grep pull a list of patterns from a specific text file. The layout of the text file is simple, only one pattern per line. grep will then match any pattern listed in the file. Switch to the /var/spool/smtpd/etc directory and create two files called matches.bad and matches.doc. In the matches.bad file we list all the filename patterns that we definitely don't want to come into our network without being inspected by the administrator.

The matches.doc file, on the other hand, should contain filename patterns for documents that may contain embedded viruses, such as Word documents or spreadsheets. When creating these files, use the form filename=.*\.exe for each line. This is so you will not get false alarms because of random strings in the mime encoding that happen to match what grep is scouting. Also, be sure not to have any blank lines in the file because grep will interpret a blank line as a pattern to be searched for and will match all e-mails. Vim is a good editor to use for creating these files since you easily can see any blank lines. You can find the contents of the files I use in Listing 3.

Listing 3. grep Pattern Files

The other script we will use is called daysumm and resides in /var/spool/smtpd/bin. Create this file as shown in Listing 4. daysumm sends a report of the day's e-mail activity to the administrator. It shows how many e-mails were received that day, how many viruses were detected and which viruses they were. This file should be set up in cron to run at 11:59 P.M. each evening. The daysumm script relies on the virus.$date and email.$date files in /var/spool/smtpd/etc. These are text files that are created dynamically, updated by scanmail and are date-dependent. For this reason, daysumm must be run before midnight or the timestamp will change and the wrong files will be read.

Listing 4. daysumm Script

If you will notice, each time we send an e-mail from the firewall itself, such as from a script, we follow it up with the command sendmail -q. The -q switch tells sendmail to start up, check for any outgoing mail, send it and then exit. It effectively flushes all local outbound sendmail queues. This is necessary since there is no MTA running on the machine anymore. The smtpd package is not an MTA but a store-and-forward agent only. It can be thought of as a dedicated mail relay program. Without this command, you would never get any mail that originated on the e-mail firewall itself.

We should now automate the whole process via the cron dæmon. We will do this by way of the uucp user's personal crontab file. Make sure you are logged on as uucp and issue the command crontab -e. This will open up uucp's cron table for editing. Make the following entries for the scanmail, daysumm and sigupdate scripts:

MAILTO=""
# Run the scanmail script every two minutes
*/2 * * * * /var/spool/smtpd/bin/scanmail
# Run the daysumm script every night at 11:59PM
59 23 * * * /var/spool/smtpd/bin/daysumm
# Run sigupdate every Thursday at 4:00PM
0 16 * * 4 /var/spool/smtpd/bin/sigupdate

You can adjust the runtimes as necessary to suit your needs. The main consideration as to how often to run scanmail is your average daily e-mail volume. If you have a daily volume of over 10,000, I would set the interval to every two minutes and set smtpfwdd to run every five minutes. This is so you don't send huge batches of e-mail to your internal server all at once. If you have a volume of around 1,000 a day or less, then scanmail could be run every ten minutes and smtpfwdd could scan every 20 minutes. Also, don't leave out the MAILTO="" entry at the top of the crontab. This is so crond will not send the uucp user any mail about completed cron jobs. E-mail every two minutes will add up quickly, and uucp never checks its own mail.

I also have a Samba share set up to allow me access to the /var/spool/smtpd hierarchy from my Windows machine. This is helpful to administrators that mainly use Windows. It keeps me from having to open up an ssh session each time I need to check a virus warning message. You can do that by adding an entry such as this to your /etc/smb.conf file:

[mail-gate]
   Comment = Email firewall directories
   Path = /var/spool/smtpd
   Valid users =
   Admin users =
   Browseable = No
   Read Only = No

where you is your Samba user name.

The final step is to divert all incoming SMTP connections from your firewall to your new scanning server. If your firewall uses IP masquerade, this is an easy task. Simply execute the command:

ipmasqadm portfw -a -P tcp -L firewall 25 -R destination 25

where firewall is the address of the firewall and destination is the address of your new scanning server.

If you have decided to run the e-mail firewall directly on your existing firewall, then you don't need to make any changes. If you have another firewall product, you will need to read the documentation to figure out how to do port redirection. It should not be too much trouble, but you may need to contact the maker of your product to find out. Be sure to put this redirection command in your startup scripts so that they will persist through rebooting.

If everything went well, you should now have an operational e-mail-virus firewall. Try sending yourself some test messages from a free web e-mail account to make sure everything is in working order. I periodically send myself macro-virus infected documents to make sure the system is still functioning correctly. I can think of many ways to expand this basic system to do other things. Especially promising is the daysumm script, which could be enhanced greatly. I currently am working on a CGI script that will report current statistics, such as average daily volume, in an intranet page. That is only one of a hundred ways this system could help you manage and safeguard your e-mail system with little or no impact on the company budget. If anyone comes up with some cool ways to expand the system, please let me know. I'd love to hear about it.