I used Michael Peek's Perl script, then decided to reimplement its general algorithm without using Perl in order to decrease memory expense. You can choose between two variants: one requires to recompile Exim (4.51 or newer) but is more efficient, another described here works with any Exim instance (Perl support isn't needed) but is slightly slower (though much faster than Perl).

Michael wrote:

We're a small organization, and I'm a lazy man. I don't really want to set up and maintain a database server. Especially since for us a greylist only involves keeping track of a few hundred KB of data at any one time.

It certainly makes sense. A filesystem can be considered as a database of sorts, using it for keeping greylisting data requires much less memory and maintenance. My implementation also keeps the data in files in a directory, but I place the data into filenames, not file content - that speeds up searching. Also, with my modifications of the algorithm a (long) update at every call is unnecessary.

• Basic knowledge of your Exim configuration file.
• Basic knowledge of Unix commands.

You need to know where binaries of some Unix commands live and where Exim spool directory is. Run commands:

whereis find touch; exim -be '$spool_directory'

In my case (FreeBSD) the output is:

find: /usr/bin/find /usr/share/man/man1/find.1.gz /usr/src/usr.bin/find
touch: /usr/bin/touch /usr/share/man/man1/touch.1.gz /usr/src/usr.bin/touch
/var/spool/exim

So, in my case both find and touch binaries are in /usr/bin, Exim spool directory is /var/spool/exim

In Exim spool directory create a subdirectory greylist with same owner and permissions as other subdirectories there. An example:

root@lena:/root# cd /var/spool/exim/
root@lena:/var/spool/exim# ls -l
total 56
drwxr-x---  2 mailnull  mail    512 May 17 17:23 db
drwxr-x---  2 mailnull  mail    512 May 24 21:59 input
drwxr-x---  2 mailnull  mail    512 May 24 21:44 msglog
drwxr-x---  2 mailnull  mail    512 May 24 21:59 scan
root@lena:/var/spool/exim# mkdir greylist
root@lena:/var/spool/exim# chown mailnull:mail greylist
root@lena:/var/spool/exim# chmod 750 greylist

Create a cron job like this (one line):

*/30 * * * * /usr/bin/find /var/spool/exim/greylist -cmin +363 -type f -delete

Here 363 is time in minutes: after the first attempt (defer), letters from the /24 block with the envelope-from and envelope-to addresses are deferred for 3 minutes, then allowed for a time between 6 hours and 6.5 hours.

Make a backup copy of your Exim configuration file.

If you used the Perl script for greylisting and don't use Perl for anything else then delete (or comment out) the perl_startup line.

For simplest usage, in your RCPT acl check choose a place (somewhere below accepting authenticated users) to put something like this (correct directory names if they differ in your operating system):

warn set acl_m_greyfile = /var/spool/exim/greylist/${length_255:\
  ${sg{$sender_host_address}{\N\.\d+$\N}{}},\
  ${sg{$sender_address,$local_part@$domain}{\N[^\w.,=@-]\N}{} }}

defer log_message = greylisted
  condition = ${if exists{$acl_m_greyfile}\
  {${if >{${eval:$tod_epoch-\
  ${extract{mtime}{${stat:$acl_m_greyfile}} }}\
  }{180}{0}{1}}\
  }{${if eq{${run{/usr/bin/touch $acl_m_greyfile} }}{}{1}{1} }} }
  message = Deferred: Temporary error, please try again later

Here the first sg does the same as $cidr_mask=24 in Michael's script, 180 is defer timeout in seconds.

If you use Exim 4.63 or older then change $acl_m_greyfile to for example $acl_m9.

I greylist only suspicious connections and use various whitelists in order to minimize delays and false positives. You can use snippets from my Exim configuration file for developing your Exim configuration.

You can test greylisting on one email address (receiving spam) before employing it for all mail, for that insert a condition between log_message and the condition with exists, like this:

condition = ${if eq{$local_part@$domain}\
                   {someaddress@your.domain}}

Check syntax of your updated Exim configuration file with exim -bV, restart exim daemon (using kill -HUP) and watch the log files for possible errors.

The data is kept in names of files in the greylist directory. Its size varies, but doesn't grow forever. My mailserver endures about 3 thousand spam attempts per day, at the time of this writing the directory is 25 Kbytes long and contains 350 files. No noticeable delays were observed in case of 60000 files (FreeBSD 7). 350000 files proved too much (too slow) - in case of so heavy load a database copes better. Quantity of files is limited also by quantity of free inodes on the partition with the greylist directory (one inode per file). Check quantity of free inodes with df -i command. For example, under FreeBSD a 3 GB /var partition has 400000 free inodes by default.

Lena