System Administration

Automated Mail Purging for SMTP Mail

Mr. Keller gives us three scripts for cleaning out old mail files automatically.

By Michael S. Keller

If your Linux system actively handles electronic mail, especially for several users, you may discover that mail consumes a lot of space where it resides, usually in one of /var/mail (System V-compliant systems), /usr/mail (BSD), /usr/spool/mail or /var/spool/mail (where my Debian GNU/Linux system stores it). The collection of Bash scripts presented here provides a way to reduce the space used by mail files by purging old messages.

Background

For several months I worked as a contractor at a company that used a Unix system to process all electronic mail for its employees. Several hundred users retrieved their mail from a single host that served as both Simple Mail Transfer Protocol (SMTP) server (for general mail transport) and Post Office Protocol (POP3) server (for clients to retrieve mail without logging into the host). While it had a home-grown mail purge system, users could inadvertently defeat the system, allowing files to grow without end.

The existing mail purge system ran as a nightly cron job. It would determine the date 60 days before and construct a string in SMTP mail header format. It would then loop through all the user mail files in /var/mail using the grep command to find that string and note the lines on which it appeared. For each file that contained the string, it would use the tail command to discard everything prior to the line containing the target date string.

If messages arrived undamaged, stayed intact, always appeared in date sequence and cron called this job daily, this method would usually work. However, sometimes cron doesn't run a job, messages do not always arrive in date order and other software run against mail files might reorder messages. Because of the existing solution's all-or-nothing keep-or-discard method, it had the following problems:

It was possible to lose newer messages if an older date appeared after a newer one.
If a message body had an un-escaped string that matched the search string, part of a message could be lost.
If older messages appeared out of order, they might remain longer than desired, since the keep/discard decision was based on simply finding a string, rather than examining each message's date.
It depended on finding the exact date string instead of making a numeric comparison that could discard messages older than the target date. A message containing invalid headers and lying at the end of the message file might remain, even if it had aged beyond the retention period.
A further effect is that some mail readers might not handle damaged messages gracefully.

A Solution

To find a more reliable solution, I sought an existing free program that would address my needs. I asked my peers, including those on Internet mailing lists, but got no useful response. A search of Internet Unix archives revealed nothing, either. Thus, I was left to create a solution.

To ensure acceptable handling of messages, I settled on these requirements:

Each message must undergo individual examination to keep or discard it based on its creation date.
A reliable way to determine where messages began and ended was needed.
Because I discovered that date formats vary a bit (mail-handling programs only loosely follow the rules), I needed to convert each message date to a simple number and use that number for the keep/discard choice.

I discovered that formail, part of the procmail mail-handling package, can split an SMTP mail file into individual messages and repair damaged headers in one pass. This ability enables individual examination of each message to decide whether to keep or discard it. Since each message would undergo separate examination, date order would not be important.

How It Works

The first script, mailrm.sh (see Listing 1), requires one command-line argument, naming the number of days of messages to preserve. When run, it sets needed variables and starts checking each mail file in $MAILDIR. It uses formail, located in $FORMAIL, to verify, repair and split each mail file into individual messages.

Each message is then examined by the mailage.sh script (see Listing 2) to determine whether to keep it. First, it checks the message header's ``From'' line for the date, moving fields as necessary. (If formail has to repair a message date, the resulting date doesn't have a time zone in it.) Then it compares the message date with the value computed for today minus the number of days to retain messages. If the message is newer, the script concatenates the message onto STDOUT, saving it to a temporary file. If the message is older, the script exits with no output.

Afterward, if the new output file has a different number of lines from the original, it is moved into the original file's place, its ownership is reset and its permissions are restricted. If the original mail file is now zero length, mailrm.sh removes it. If a user has been removed from a system leaving his mail behind, this script deletes his mail file after all the messages in it have expired.

The third script, maildate.sh (see Listing 3), returns the integer number of days since 1900 of an input date in the form ``MMM DD YYYY''. The returned integer is useful for calculating the difference between two dates.

Installation

Copy the scripts to your favorite directory. I used /usr/local/bin. Edit mailrm.sh and mailage.sh to reflect your mail directory and the location of formail. Then just run the script with the number of days to retain messages as its argument. For example, to purge messages older than 60 days:

/usr/local/bin/mailrm.sh 60

What Could Be Done Better

Is this mail purge solution perfect? No, it does nothing to lock mail files, which could pose a problem if a user's mail client polls frequently or this job runs during busy hours. Some possible steps to address this could include stopping sendmail while the scripts run, preventing the POP3 server from running and tightening permissions on the mail directory to prevent access from non-superusers. Since it would normally run in the early-morning hours via cron, the probability of collision would be low.

Also, this solution relies on external utilities that may not function as expected. formail might not properly handle all the mail headers, though I haven't encountered problems yet. cat might not like some characters that could appear in messages, resulting in lost message text. I've had few problems with cat, but your experience may be different.

The date conversion logic in maildate.sh is simplistic. It's not accurate for the year 1900, and the leap year calculation will not work correctly after the year 2099. However, it works well for calculating the difference between two dates, and it's reasonably fast.

Since I use three scripts to get around parameter-passing limitations in Bash, this package runs more slowly than it might (because of having to fork processes repeatedly). Recoding the scripts into a single Perl script might help--my Perl skills are too limited for this project.

Conclusion

If you have a need for automated mail purging, these scripts can help you reach your goal. At least, they may give you ideas for your own solution. If you create a more elegant solution, I'd like to hear about it.

Resources

Michael S. Keller is a Technical Analyst with Paranet, Inc., a nation-wide network services provider owned by Sprint. He has used computers for twenty years and Unix variants for seven. You may reach him at mskeller@paranet.com. Paranet's virtual home is at http://www.paranet.com/.

All listings referred to in this article are available by anonymous download in the file ftp://ftp.ssc.com/lj/listings/issue47/2118.tgz.