Subversion Repositories

?revision_form?Rev ?revision_input??revision_submit??revision_endform?

Rev 1 | Blame | Compare with Previous | Last modification | View Log | RSS feed

                          GREYLISTING with SA-Exim
                          ------------------------


INTRODUCTION
------------
SA-Exim allows for intelligent greylisting by combining the idea of greylisting
with Spam scores from SpamAssassin

If you don't know what greylisting is, you should probably go read up there:
http://projects.puremagic.com/greylisting/
(note that this implementation works differently than the one described there)

So, SA-Exim isn't just yet another greylisting implementation. By tying it
into SA-Exim, and especially by running SA at SMTP time, you can do the
following things:
- do not bother greylisting people who send messages detected as spam by SA
  (indeed, regular greylisting will accept mail from a spammer if he retries
  or sends it from an open relay)
  SA-Exim will never greylist, or whitelist a sender based on a mail clearly
  marked as spam by SA.

- do not delay mail from people who aren't spamming you (this one is the most
  important feature of SA-Exim greylisting, as it removes the biggest
  disadvantage linkes to greylisting)

- only greylist (and maybe later whitelist) hosts that send you mail with
  a certain SA score.


IMPLEMENTATION
--------------
So how does this all work?
SA comes with a patch for SA 2.x (and a module for SA 3.x) that does the
following things:
- add a greylisting rule which gets run at the very end, and where if
  the score is already higher than a configured value, we do not bother
  greylisting the host. We just return a rule failure, which doesn't
  change the score and lets SA-Exim reject the mail as usual
- if the score is lower than the "surely spam" threshold (shown as 11 in the
  example below), check for a file in 
  /var/spool/sa-exim/messageids/co/nn/ect/ip/envfrom/envto
  - if it's there, check if it was written more than x seconds ago (1800s/30mn
    in the example below)
    - if so, change the status to whitelisted and return true so that SA applies
      the whitelist negative score
    - if not, simply increase counters, host is still greylisted 
  - if the file is not there, create it
- every x time (like 4 hours or two days), remove all greylist entries that
  only saw one mail (i.e. still greylisted, not whitelisted yet).
  This is done with a find cron job
- every y time (like 1 week), remove whitelisted entries so that your filesystem
  doesn't clutter up with hosts you're not going to hear from again in a while


Then, you call the greylisting rule with this (in SA's local.cf)
# reseval is a special eval which only runs after you have the result from
# everything else (lets us not greylist a host that is sending spam, otherwise
# this rule might set a sufficiently negative score that the next spam would
# be allowed in)
# Note the 'key' -> 'value'; syntax. It's a special hack to go through SA's
# config parser. You need to keep that exact syntax
# greylistsecs: how long you greylist a tuplet because whitelisting it
# greylistnullfrom: set to 1 to also greylist mail with a null env from
# greylistfourthbyte: keep the 4 bytes of the connecting host instead of 3
header GREYLIST_ISWHITE reseval:greylisting("( 'dir' => '/var/spool/sa-exim/tuplets'; 'method' => 'dir'; 'greylistsecs' => '1800'; 'dontgreylistthreshold' => 11; 'connectiphdr' => 'X-SA-Exim-Connect-IP'; 'envfromhdr' => 'X-SA-Exim-Mail-From'; 'rcpttohdr' => 'X-SA-Exim-Rcpt-To'; 'greylistnullfrom' => 0; 'greylistfourthbyte' => 0 )")
describe GREYLIST_ISWHITE The incoming server has been whitelisted for this rece
ipient and sender
score GREYLIST_ISWHITE  -1.5

Note that SA greylisting depends on X-SA-Exim-Rcpt-To, so you have to ensure
that SAmaxrcptlistlength is set to a reasonably high value (up to 8000) instead
of the current default of 0 (you can remove the header in exim's system_filter
or a transport if you don't want it to show in user's mails, see "privacy
warning" in README)


Now, in case you aren't confused yet, you get even more knobs to play with :)
If a spammer resends you a spam until it gets whitelisted (or typically, it
gets sent to a relay that resends it to you), even if you are setup to
accept the spam at the point, you don't want to lower the SA score too much
just because the mail was resent to you several times (i.e. a rather negative
score for GREYLIST_ISWHITE). So, you can actually configure SA-Exim to temp
reject messages on a much higher score than usual, if they don't have the
GREYLIST_ISWHITE tag.

In other words, let's say you have this in sa-exim.conf:
SApermreject: 11.0
SAtempreject: 3.0
SAgreylistraisetempreject: 6.5

If a mail comes in at less than 3.0, the SA patch/module remembers the sending
server's connecting IP, the env from, and the rcpt to(s), and whitelists those.
(those will be referred to as tuplets, one for each rcpt to)

If the score is between 3.0 and 11.0,
- if at least one of the tuplets is already whitelisted, SA applies the -1.5
  score and yields an end score below 9.5, Now, at the same time, SAtempreject
  is temporarily raised by 6.5, so everything under 9.5 is accepted, which 
  basically means that the mail goes through.
- if none of the tuplets are whitelisted, they get greylisted
- if they are greylisted, they can get upgraded to whitelisted status if the
  sending server has been trying for long enough (1800secs in the example given
  above). At this point the same thing happens as in case #1 and the mail is
  accepted

If a tuplet that is going to be whitelisted or greylisted, already is, SA
updates counters to let you run reports and anything else you want, like
deciding if or when you'd like to expire the entry

If by now you wonder why you would want to both decrease the SA score and
increase the maximum score you'll accept mails on, the reason is as follows:
You probably don't want to lower the SA score by 8 just because the tuplet
is whitelisted (not only does it mess with the SA scoring of messages that used
to be flagged as spam, but a spam with a score of 13.5 would then be lowered
to 5.5, be temprejected, and be close to the accept range).
Instead, giving an SA score of -1.5, a message with 13.5 becomes 12.0 and still
gets rejected right away. You also do not overly (and artifically) lower the
score of a message, just for SA-Exim's sake

If you so wish, you can also give the SA rule a score of -0.1, and only
dynamically raise the tempreject score for messages that are whitelisted.


SCORE SETUP
-----------
It makes little sense to have
SAtempreject + SAgreylistraisetempreject + SA GREYLIST_ISWHITE > SApermreject
as there is little point to raise SAtempreject if the message that's
whitelisted still gets refused by the SApermreject score

As to whether you want to put more points into SA GREYLIST_ISWHITE or
SAgreylistraisetempreject, this is your call, but as a general rule, you only
want to change the SA score in a way that makes sense for spam scoring,
as it similarly affects the score of all messages, whether SA-Exim sees them
in the non spam range, tempreject range, or "this is spam that I would never
let in" range.


FILE SETUP
----------
Make very sure that uid nobody can traverse /var/spool/sa-exim and
create tuplets writeable by nobody (or whoever you run SA as)

Then, setup a cron job to delete tuplets that are older than 14 days for
whitelisted entries, and 2 days for greylisted entries (or whatever
values you fancy).
Note that because this implementation does not systematically force the senders to resend you mail, unless they sent something that looks too much like spam,
you will typically see few whitelisted entries, and those will either be 
potential spam that was actually resent to you at least 30mn after the
initial copy (or whatever value you setup in "header GREYLIST_ISWHITE"), or
people who sent you several Emails (where the second Email will just happen to
trigger a whitelisting).


FILE SETUP
----------
You should install greylistclean.cron in /etc/cron.d/ on your system to
call greylistclean and clean up greylisted entries and whitelisted entries
that haven't been used in a while.
You can optionally modify it to tweak the cleanup times.
Note that you need to tweak greylistclean.cron to match the user spamd runs
as if you aren't using the recommended --username=nobody


SA PATCH (SA 2.x)
-----------------
For all this to work, you also need to patch SA with SA-greylist.diff
from the source tar (or /usr/share/doc/sa-exim*/ for a precompiled package).
This patch never made it to the main SA 2.x branch as the developers had mostly
switched to 3.x where you can use plugins.
If you still use SA 3.x, you can go to /usr/share/perl5/Mail (or wherever
appropriate on your system), and run 
patch -p0 -s < /path/to/sa-exim/SA-greylisting.diff
Note that while the patch works, it will not be maintained anymore since 
it is deprecated for the SA 3.x plugin


SA PLUGIN (SA 3.x)
------------------
Newer versions of SpamAssassin support plugins, so there is no need to 
patch SA anymore, you can just install the Greylisting.pm module on your
system and get SA to use it
This is how you call the module in SA 3.x (i.e. put this in your
/etc/spamassassin/local.cf)

# Note the 'key' -> 'value'; syntax. It's a special hack to go through SA's
# config parser. You need to keep that exact syntax
# greylistsecs: how long you greylist a tuplet because whitelisting it
# greylistnullfrom: set to 1 to also greylist mail with a null env from
# greylistfourthbyte: keep the 4 bytes of the connecting host instead of 3
loadplugin Greylisting /usr/share/perl5/Mail/SpamAssassin/Plugin/Greylisting.pm
header GREYLIST_ISWHITE eval:greylisting("( 'dir' => '/var/spool/sa-exim/tuplets'; 'method' => 'dir'; 'greylistsecs' => '1800'; 'dontgreylistthreshold' => 11; 'connectiphdr' => 'X-SA-Exim-Connect-IP'; 'envfromhdr' => 'X-SA-Exim-Mail-From'; 'rcpttohdr' => 'X-SA-Exim-Rcpt-To'; 'greylistnullfrom' => 1; 'greylistfourthbyte' => 0 )")
describe GREYLIST_ISWHITE The incoming server has been whitelisted for this recipient and sender
score GREYLIST_ISWHITE  -1.5
# Run SpamAssassin last, after all other rules.
# (lets us not greylist a host that is sending spam, otherwise this rule might
# set a sufficiently negative score that the next spam would be allowed in)
priority GREYLIST_ISWHITE 99999


SA-EXIM NEW BEHAVIOR CONCERNS
-----------------------------
What greylisting changes as far as spam accepting or rejection is concerned:
Once a tuplet has been whitelisted, spam from that host is more likely
to be accepted until the tuplet expires. In the case of a mailing list,
unless you run a find / rm based on the creation time and not the last
modified time, you will then be a bit more likely to accept spam from
that list.
If this turns out to not be acceptable in your case, there isn't a whole
lot you can do about this, except deleting greylist entries for the host
from cron before they get promoted to whitelist.

What you can do on top of the existing greylisting code:
Parse the SA-Exim logs and if you get spam from an IP, you can decide
to delete greylist entries in /var/spool/sa-exim/tuplets/IP or just
/var/spool/sa-exim/tuplets/IP/envfrom
This may not may not be a good thing if you receive the occasional spam
from a mailing list as you'll then re-delay mail for that list, but then
again, it will also remove whitelisting for a host that spammed you once
with an Email that managed to get under the SA scoring radar


GREYLISTING AND MXES
--------------------
Depending on your configuration, you may have realized that SA-Exim doesn't
play very well with secondary MXes for your domain if they don't run SA-Exim
too (for instance, you'd send a tempreject on spam and clog up your
secondary, or maybe even teergrube it if you forgot to add your MX's IP
in the do not teergrube list.
For greylisting, it's even more simple:
If your secondary MXes aren't running SA-Exim with greylisting, then
greylisting's efficiency will be greatly reduced as most spammers will send
their spams to your secondary MXes which will accept the mail for you,
even if it's sent only once, and then your MXes will resend the spam to you
until you accept it (rendering greylisting useless)

Now, if your secondaries are running greylisting too, most mail will flow
through with no delay whatsoever. However, in the worst case scenario, a mail
that isn't spam, but triggers greylisting because its score is high enough to
generate a tempreject, could be delayed up to twice the whitelisting time
if it were to go to your secondary MX first (assuming your primary is
unreachable or temporarily overloaded), and then be resent to your primary
MX, which would trigger a second greylisting delay
FIXME: implement a whitelist of sending IPs so that greylisting returns
whitelisted right away


SECURITY
--------
The greylisting function works around the SA parser by sending all the options
as a hash inside a string. In turn, greylisting evals the said string.
This is a security problem if you allow your users to run custom rules and it
gives them access to run spamassassin as a user different from their own, or
in a way that they otherwise wouldn't be able to.
Do not run greylisting if this a problem for you (in the default SA/SA-Exim
setup, this shouldn't be a concern since it doesn't even parse users' config
files)