Package: crm114
Version: 20070810-2
Severity: important
I get the very same error than the one reported by Marcus Fritzsch in
#439514 (which apparently is a different bug).
My spam|ham teaching setup was working flawlessly previously (I guess it
appeared in 20070810-2).
I pipe mails to "crmlearn --learnnonspam" or "crmlearn --learnspam" with
crmlearn being:
#!/bin/sh
grep -a -v "^X-CRM114" | \
$HOME/.crm114/mailfilter.crm -u $HOME/.crm114/ $1 --force | \
grep -a "^X-CRM114"
My mailfilter.crm is attached (it is indeed the one from the examples
directory with no change).
Just in case, I recreated the *.css files from scratch.
The error is:
ERROR: mailfilter.crm broke. Here's the error:
ERROR:
/usr/bin/crm: *ERROR*
This file should have learncounts, but doesn't, and the learncount slot
is busy. It's hosed. Time to die.
Sorry, but this program is very sick and probably should be killed off.
This happened at line 716 of file /home/bubulle/.crm114/mailfilter.crm
-- System Information:
Debian Release: lenny/sid
APT prefers unstable
APT policy: (500, 'unstable'), (1, 'experimental')
Architecture: i386 (i686)
Kernel: Linux 2.6.22-1-686 (SMP w/1 CPU core)
Locale: LANG=fr_FR.UTF-8, LC_CTYPE=fr_FR.UTF-8 (charmap=UTF-8)
Shell: /bin/sh linked to /bin/dash
Versions of packages crm114 depends on:
ii libc6 2.6.1-1+b1 GNU C Library: Shared libraries
ii libtre4 0.7.5-1 regexp matching library with appro
Versions of packages crm114 recommends:
ii metamail 2.7-53 implementation of MIME
-- no debconf information
#! /usr/bin/crm
# -- bangline args removed for system portability --
# -( learnspam learnnonspam learnfile stats_only config spamcss nonspamcss
fileprefix force unlearn cache )
#
# A statistical mail sorter with mail-to-yourself commanding
#
# Copyright (C) 2002-2006 William S. Yerazunis; licensed under the
# GNU Public License (GPL) version 2. A copy of this license is included
# in the distribution media, or obtain one from www.fsf.org .
#
# Note to SunOS and FreeBSD users - you MUST remove EVERYTHING on
# the first line of this program from the first "-" to the end of
# the first line (including the "-" sign itself) or you will not get
# what you expect. This is due to "differences of opinion" on how
# a bangline should be dealt with.
#
#
# --->>> Design Philosophy ( do these IN ORDER )
#
# * if --fileprefix is specified, all filenames EXCEPT --config
# are prefixed with that (You need a trailing slash on the prefix
# if it is a directory name.)
# * if --config , grab the config file from the specified place.
# * Load the mailfilter.cf config file from wherever config or
# fileprefix points (--config overrides --fileprefix).
# * If --spamcss is specified, use that as the spam.css file
# * If --nonspamcss is specified, use that as the nonspam.css file
# * If --learnspam, learn as spam and exit.
# * If --learnnonspam, learn as nonspam and exit
# * If --force, force-feed the learning
# * If --learnfile, use :learnfile:.css and :learnfile:text.txt
# * If --stats_only, do normal classification but don't do any
# forwarding, only output the status on stdout and return
# the exit code.
#
# * check for the "command word", if present, execute the command
#
# * check to see if any of the whitelist patterns apply. If so,
# accept the mail to /var/spool/the_user (which is actually to
# be found at /var/spool/mail/:*:_env_USER:
#
# * check to see if any of the blacklist patterns apply. If so,
# flush the mail to the "blacklisted" file.
#
# * check to see if it's commanded to be learned as a spam or a
# nonspam model. If so, learn it to the appropriate .css (Crm
# Sparse Spectra) file
#
# * run the email through the classifier. If the classifier thinks
# it's good, send it to the /var/spool/mail/the_user file, else
# send it to the "doubtful" file.
#
##############################################################
#
# --- uncomment this if you want to include a "forced"
# configuration file ---
# insert mailfilterconfig.crm
#
#
# --- These vars must have a value, or else we'll get errors ----
#
isolate (:classifier_reason:) /no reason yet/
#
isolate (:classify_status:) //
#
isolate (:our_exit_code:) /0/
#
isolate (:stats:) / pR: 0.000000 /
#
isolate (:pr:) / pR: 0.00000/
#
isolate (:subj_text:) / (None) /
#
isolate (:add_extra_stuff:) //
#
isolate (:decision_length:) /4096/
#
isolate (:cache:) // <default>
isolate (:cachedir:) //
isolate (:cacheid:) //
isolate (:msg_hash:) //
isolate <default> (:learnspam:)
isolate <default> (:learnnonspam:)
isolate <default> (:learnfile:)
isolate <default> (:stats_only:)
isolate <default> (:automatic_training:)
#
# Isolate these email addresses, and give them values,
# in case the user doesn't.
isolate (:reject_address:) //
isolate (:fail_priority_mail_to:) //
isolate (:fail_blacklist_mail_to:) //
isolate (:fail_SSM_mail_to:) //
isolate (:log_rejections:) //
#
# this ISOLATE will guarantee that :fileprefix: exists, and keep it's
# prior (commandline) value if it does, and an empty string if it doesnt
isolate (:fileprefix:)
#
# This ISOLATE will guarantee that :force: will exist, and keep the
# commandline value ("SET") , or the null string if the user doesn't
# use --force on the command line.
isolate (:force:)
#
# This ISOLATE will guarantee that :unlearn: will exist, and will keep
# the commandline value ("SET") or the null string if the user doesn't
# use --unlearn on the command line.
isolate (:unlearn:)
#
# now, :clf: is the classify & learn flags; note that we have two
# separate flags here in a bizarre chain. The reason is that :unlearn:
# can have the value "SET", whereas :rft: needs "refute"
isolate (:clf:) //
#
# and someplace to catch mailtrainer if we need it.
isolate (:mailtrainer_output:) //
#
#####################################################################
#
# This is the code to read the per-user configuration. Note
# that because this happens during the run, it will _override_
# any comand line arguments that get set.
{
isolate (:option_txt:)
isolate (:ev:)
isolate (:verbose_startup:)
isolate (:config:)
#
# Part 1 - read in the options/configuration file
#
{
{
match [:config:] /.+/
input [:*:config:] (:option_txt:)
}
alius
{
# read in the standard mail filter configuration file.
input [:*:fileprefix:mailfilter.cf] (:option_txt:)
}
}
#
#
# reset loop for matching to start of :option_txt:
match [:option_txt:] //
# and loop till there are no more options.
{
# find a line that looks like a parameter setting...
match < fromend nomultiline > (:line: :name: :value:) \
[:option_txt:] /^[ ]*(:[[:graph:]]+:)[ \t]+\/(.*)\//
{
# don't execute the assign if there's
# a # at the start of the line.
match <absent> [:name:] /^\x23/
{
# Verbose startup?
match [:verbose_startup:] /SET/
output / :*:name:\n :*:value:\n/
}
isolate (:*:name:) /:*:value:/
}
liaf
}
}
#
# Now, a tricky bit - we need to add "unlearn" to the :clf:
# if it was in the line params - but we have to append, not replace,
# because :clf: (CLassfier Flags) also contains the classifier we use.
#
{
match [:unlearn:] /SET/
alter (:clf:) /:*:clf: refute/
}
#
#
# Do a quick check- has the password been changed or not? If it's
# still the default, put in something that will be well-nigh unguessable
# (esp. since it will contain recieved headers that the sender cannot
# see nor control.)
{
match [:spw:] /DEFAULT-PASSWORD/
# yes, it's the same as default. So we scramble it just so
# nobody can hack in
hash (:spw:) /:*:_env_string::*:_dw:/
}
#############################################################
#
# Set up the addresses that we might need to mail to
#
isolate (:reject_address:) /:*:general_fails_to:/
{
match [:fail_priority_mail_to:] <absent> /[[:graph:]]/
alter (:fail_priority_mail_to:) /:*:general_fails_to:/
}
{
match [:fail_blacklist_mail_to:] <absent> /[[:graph:]]/
alter (:fail_blacklist_mail_to:) /:*:general_fails_to:/
}
{
match [:fail_SSM_mail_to:] <absent> /[[:graph:]]/
alter (:fail_SSM_mail_to:) /:*:general_fails_to:/
}
###############################################################
# Does the user want us to log all incoming mail? This is handy for
# testing and auditing purposes.
{
match [:log_to_allmail.txt:] /yes/
output [:*:fileprefix:allmail.txt] <append> /:*:_dw:/
}
###############################################################
#
# Is text cacheing turned on?
{
match [:text_cache:] /./
{
### If the text_cache dir isn't there, create it
# and it's subdirectories.
#
isolate (:tmp:) //
syscall () (:tmp:) /ls :*:text_cache: 2>&1 /
match [:tmp:] <absent> /texts/
syscall () () /mkdir -p :*:text_cache: /
syscall () () /mkdir -p :*:text_cache:\/texts /
syscall () () /mkdir :*:text_cache:\/prob_good /
syscall () () /mkdir :*:text_cache:\/prob_spam /
syscall () () /mkdir :*:text_cache:\/known_good /
syscall () () /mkdir :*:text_cache:\/known_spam /
syscall () () /mkdir :*:text_cache:\/empty /
}
#
# Yes, text cacheing is on. Save the input as is in the cache.
#
isolate (:system_time:) //
syscall () (:system_time:) /date +%Y%m%d_%H%M%S_%N /
match [:system_time:] (:: :cacheid:) /([[:graph:]]+)..../
hash (:msg_hash:) /:*:_dw:/
alter (:cacheid:) /sfid-:*:cacheid:_:*:msg_hash:/
# As long as this isn't a "learn" run, nor a "stats-only" run,
# we should save the text of this message in the text cache directory.
# Note to self: eventually this should also work with the
# command <password> spam nonspam stuff
{
match <absent> [:stats_only:] /SET/
match <absent> [:learnspam:] /SET/
match <absent> [:learnnonspam:] /SET/
output [:*:text_cache:\/texts\/:*:cacheid:] /:*:_dw:/
}
}
##############################################################
#
# Grab the text that we're going to actually work with.
#
# We copy this into m_text - the "mutilated text". It
# will become an annotated _copy_ of the incoming text,
# with whatever changes we think will help us classify better.
#
# We clip m_text to be the first :decision_length: characters of
# the incoming mail.
#
match (:m_text:) [:_dw: 0 :*:decision_length:] /.*/
isolate (:m_text:)
#
# :b_text: is the text with base64's expanded.
isolate (:b_text:) /:*:m_text:/
#
# :i_text: is the text with Hypertextus Interruptus removed.
isolate (:i_text:) /:*:m_text:/
#
#
# To start with, the commanded text is assumed to be the entire input.
# THEN
# If there's a command followed by text, we save the text so we can
# put that, and _only_ that, into the .txt corpi.
{
isolate (:cmd_txt:) /:*:_dw:/
match (:: :cmd_txt:) [:_dw:] /command :*:spw: [^\n]*\n(.*)/
}
#
#
# do we do any expansions?
{
# expansion 1: - do we perform base64 expansions?
{
{
match [:do_base64:] /yes/
{
# yes, expand base64's if there are any
#
# Note: some spams don't even bother to use
# a 'Content-Transfer-Encoding' marker,
# and even fewer use Content-Type: text/whatever
# so we have to sort of wing it, when to expand
# what _might_ be base64 and when to ignore it.
# For now, if it says it's a base64, it gets
# expanded, no matter what the type. Maybe
# someday someone will put in a lockout for
# things like .jpg files, .doc files, etc.
#
isolate (:exp_text:)
match [:b_text:] <nocase> (:a: :h: :b:) \
/(Content-Transfer-Encoding): base64(.*)/
match (:c:) [:b:] \
/([a-zA-Z0-9+=!\/]+:*:_nl:){2,200}/
#
syscall (:*:c:) (:exp_text:) /:*:mime_decoder: /
# and stuff the result back into b_text for
# classification right in context.
alter (:c:) /:*:exp_text:/
# and mark this piece of mime as "prior".
alter (:h:) /Content-Transfer-Prior-Encoding/
# repeat till no more Mime base64 encodings
liaf
}
}
alius
{
# if no base64 expansions enabled, empty out :b_text:
#
alter (:b_text:) //
}
}
#
# If we had expansions, bust the html contents out of them, otherwise
# ignore b_text as it's redundant
{
{
match [:b_text:] <nocase> /Content-Transfer-Prior-Encoding/
alter (:i_text:) /:*:b_text:/
}
alius
{
# if :b_text: _didn't_ have a base64, it's useless
alter (:b_text:) //
}
}
# expansion 2 : do we bust HTML comments ( a.k.a.
# hypertextus interruptus) out?
{
match [:undo_interruptus:] /yes/
isolate (:commentbin:) //
{
match [:i_text:] (:comment:) /<!--([^-]|-[^-]|--[^>])*-->/
alter (:commentbin:) /:*:commentbin: :*:comment:/
alter (:comment:) //
liaf
}
# if we had at least 80 characters worth of comments, then
# it's worth using the decommented text, else not.
# (this my personal judgement call)
{
{
match [:commentbin:] /(.){80,}/
}
alius
{
alter (:i_text:) //
}
}
}
}
# and reassemble the mucked-over text into the :m_text: var, always
# with the base64's expanded, then a second decommented copy
#
{
isolate (:m_text:) \
/:*:m_text: :*:_nl: :*:b_text: :*:_nl: :*:i_text: :*:_nl:/
}
#########################################################
#
# Do we want to do any rewrites before running?
#
{
match [:rewrites_enabled:] /yes/
isolate (:rewrites:)
input (:rewrites:) [:*:fileprefix:rewrites.mfp]
# reset matching on rewrites to start of string - if no string, no more
# processing of rewrites !!
match [:rewrites:] //
#
#
{
# Grab the next regex; turn the one-per-line patterns into a
# regex and a replacement string.
# First, do the line-spanning regexes.
match <fromend nomultiline> (:ch: :fr: :to:) [:rewrites:] /(.+)>-->(.*)/
# see if the "fr" regex matches anywhere
{
match [:m_text:] (:place:) /:*:fr:/
# Yep, it matched... alter it and do it again
#
alter (:place:) /:*:to:/
liaf
}
# Nope, didn't match... grab the next regex and try again,
liaf
}
#
# reset back to the start of the rewrites.
#
match [:rewrites:] //
#
# and do it again for non-line-spanners
{
# Go through and do it again, except this time do it for
# the non-line-spanning regexes.
match <fromend nomultiline> (:ch: :fr: :to:) [:rewrites:] /(.+)>->(.*)/
# see if the "fr" regex matches anywhere
{
match [:m_text:] <nomultiline> (:place:) /:*:fr:/
# Yep, it matched... alter it and do it again
#
alter (:place:) /:*:to:/
liaf
}
# Nope, didn't match... grab the next regex and try again,
liaf
}
} # done with rewrites.
###################################################################
#
# Command Dispatch processing starts here
#
# ---------do we have a --learnspam or --learnnonspam command line key?
#
match (:text:) [:m_text:] /.*/
isolate (:c:) //
isolate ( :spamcss: :nonspamcss: )
{
match <absent> [:spamcss:] /./
alter (:spamcss:) /spam.css/
}
{
match <absent> [:nonspamcss:] /./
alter (:nonspamcss:) /nonspam.css/
}
{
match [:learnspam:] /SET/
goto /:learnspamhere:/
}
{
match [:learnnonspam:] /SET/
goto /:learnnonspamhere:/
}
{
match (:trash: :file:) [:learnfile:] /(.+)/
goto /:learntofilehere:/
}
#
#
#------------ Are we enabled for "inoculations" via email?
#
{
match [:inoculations_enabled:] /yes/
#
# see if we have an inoculation header.
#
match <nomultiline> [:m_text:] \
/Inoculation-Sender: ([[:graph:]]+)/ (:x: :sender:)
match <nomultiline> [:m_text:] \
/Inoculation-Type: ([[:graph:]]+)/ (:x: :type:)
match <nomultiline> [:m_text:] \
/Inoculation-Authentication: (.*)$/ (:x: :auth:)
#
# See if the sender is in our list, and if so, what is their secret?
#
isolate (:inoculation_passwd:) //
input (:inoculation_passwd:) [:*:fileprefix:inoc_passwd.txt]
match [:inoculation_passwd:] <nomultiline> \
/:*:sender: :*:type: (.*)/ (:x: :secret:)
#
# We now have the shared secret, calculate the checksum we should have
#
# grab the body...
match /\n\n(.*)/ (:x: :body:)
#
# and calculate the hash.
isolate (:md5out:)
syscall (:*:secret::*:_nl::*:body:) (:md5out:) /md5sum/
match [:md5out:] /([[:graph:]]+)/ (:x: :md5out:)
#
# does this hash match with the given hash?
match [:auth:] /:*:md5out:/
#
# Yes, it matched. It's a real innoculation.
#
# grab the text we want to actually learn (this is the payload)
match [:m_text:] (:x: :text:) /\n\n(.*)/
#
# and learn it appropriately
{
match [:type:] /nonspam/
goto /:learnnonspamhere:/
}
alius
{
match [:type:] /spam/
goto /:learnspamhere:/
}
}
#
# -------------check for the COMMAND WORD ----------
#
{
#
# grab the password as :pw:, and any arg(s) as :c:
#
match <nomultiline> (:z: :pw: :c: ) [:_dw:] /^command ([[:graph:]]+) (.*)/
#
# check the password. If it's invalid, FAIL out of this bracket set
# and just treat this as ordinary (non-command) mail.
match [:pw:] /:*:spw:/
{
# was it a command to add something to the whitelist?
match <nomultiline> (:q: :a:) [:c:] /whitelist (.*)/
output [:*:fileprefix:whitelist.mfp] <append> /:*:a::*:_nl:/
alter (:z:) /*** :*:z: *** :*:_nl:Whitelist command executed! :*:_nl:/
accept
exit /:*:accepted_mail_exit_code:/
}
{
# was it a command to add something to the blacklist?
match <nomultiline> (:q: :a:) [:c:] /blacklist (.*)/
output [:*:fileprefix:blacklist.mfp] <append> /:*:a::*:_nl:/
alter (:z:) /*** :*:z: *** :*:_nl:Blacklist command executed! :*:_nl:/
accept
exit /:*:accepted_mail_exit_code:/
}
#
# Did the user specify command "force"?
#
{
match <nomultiline> [:c:] /force/
#
# yep, so we set the "force" on.
alter (:force:) /SET/
}
#
# Did the user specify command "unlearn"?
#
{
match <nomultiline> [:c:] /unlearn/
#
# yep, so we set the "force" on.
alter (:unlearn:) /SET/
}
#
# Now, if :unlearn: is set, by either bashline or command, we
# set the :clf: flag to be "refute". Otherwise, we set it to
# be what it was before.
#
{
match [:unlearn:] /SET/
alter (:clf:) /:*:clf: refute/
}
#
# Now, the big mahonka. Learn as nonspam, or as spam
# (note the three subpaths - one each for non-forced, forced, and
# non-forced error messages)
#
{
# was it a command to learn something as nonspam?
match [:c:] /nonspam/
match (:z: :text:) [:m_text:] /:*:_nl:command [[:graph:]]+ nonspam(.*)/
# and learn it as nonspam
:learnnonspamhere:
{
# Are we supposed to use the cached version?
{
match [:c:] /cache/
alter (:cache:) /SET/
}
match [:cache:] /SET/ # can also be set on command line
# yes - so we use mailtrainer.crm to do the training
{
match (:: :fileid:) /X-CRM114-CacheID: ([[:graph:]]+)/
# check- does the cached file exist?
syscall () (:tmp:) /ls :*:text_cache:\/texts\/:*:fileid:/
match [:tmp:] /:*:fileid:/
# yes, it exists - go on with the learning method
# and remember this file on a permanent basis
syscall /ln :*:text_cache:\/texts\/:*:fileid:
:*:text_cache:\/known_good\/:*:fileid: /
# output / \n DOING: crm mailtrainer.crm
--good=:*:text_cache:\/texts\/:*:fileid: --spam=:*:text_cache:\/empty\/ \n / ()
(:mailtrainer_output:)
syscall /crm mailtrainer.crm --fileprefix :*:fileprefix:
--good=:*:text_cache:\/texts\/:*:fileid: --spam=:*:text_cache:\/empty\/ / ()
(:mailtrainer_output:)
# output /mailtrainer output: ':*:mailtrainer_output:'\n/
# and remove it from the prob_* directories, as
# now it's known
syscall /rm -rf :*:text_cache:\/prob_spam\/:*:fileid:/
syscall /rm -rf :*:text_cache:\/prob_good\/:*:fileid:/
#
# now it's trained; put in a marker in the headers
call /:mungmail_delete:/ [X-CRM114-Status: ]
call /:mungmail_delete:/ [X-CRM114-Unsure: ]
call /:mungmail_add:/ [X-CRM114-Action: LEARNED AND CACHED
NONSPAM ]
# Insert the training report into the msgbody if desired
{
match [:add_mailtrainer_report:] /yes/
match (:: :firstline:) /.*(.)/
match (:: :firstline:) /\n\n()/
alter (:firstline:) / :*:mailtrainer_output:\n-----\n/
}
accept
exit /:*:accepted_mail_exit_code:/
}
{
# No, it didn't exist. Add an error message header.
call /:mungmail_add:/ [X-CRM114-ERROR: No cached text with
that cacheID, so nothing done!]
accept
exit /:*:accepted_mail_exit_code:/
}
}
{ # No cacheing, so we learn "natively"
#
# Verify that we need to learn this first (TOE strategy)
classify <:*:clf:> [:text:] /:*:lcr:/ \
(:*:fileprefix::*:nonspamcss: :*:fileprefix::*:spamcss: ) \
(:classify_status:)
match [:classify_status:] <nomultiline> \
/^#0.* pR: ([-. 0-9]+)/ (:: :pr:)
eval /:@: :*:pr: < :*:thick_threshold: :/
output [:*:fileprefix:nonspamtext.txt] \
<append> /\n\n:*:cmd_txt:\n/
#
# write out the pre-mutilation text, with newlines
#
learn <:*:clf:> (:*:fileprefix::*:nonspamcss:) [:text:]
/:*:lcr:/
call /:mungmail_add:/ [X-CRM114-Action: LEARNED NONSPAM]
call /:mungmail_unique:/ [X-CRM114-Status: Good (Learn)]
accept
exit /:*:accepted_mail_exit_code:/
accept
exit /:*:accepted_mail_exit_code:/
}
alius
{
#
# Did the user specify "--force" on the command line?
match [:force:] /SET/
output [:*:fileprefix:nonspamtext.txt] <append>
/\n\n:*:cmd_txt:\n/
#
# write out the pre-mutilation text, with newlines
#
learn < :*:clf: > (:*:fileprefix::*:nonspamcss:) [:text:]
/:*:lcr:/
# syscall (:*:_dw:) (:_dw:) \
# /formail -A "X-CRM114-Action: LEARNED NONSPAM (FORCED)
:*:clf:"/
call /:mungmail_add:/ [X-CRM114-Action: LEARNED NONSPAM
(FORCED)]
call /:mungmail_unique:/ [X-CRM114-Status: Good (Learn)]
accept
exit /:*:accepted_mail_exit_code:/
}
alius
{
call /:mungmail_add:/ [X-CRM114-Action: LEARN AS NONSPAM
UNNECESSARY- ALREADY CLASSIFIED CORRECTLY - NO ACTION TAKEN]
accept
exit /:*:accepted_mail_exit_code:/
}
}
{
# was it a command to learn something as spam?
match [:c:] /spam/
match (:z: :text:) [:m_text:] /:*:_nl:command [[:graph:]]+ spam(.*)/
# and learn it as spam
:learnspamhere:
{
# Are we supposed to use the cached version?
{
match [:c:] /cache/
alter (:cache:) /SET/
}
match [:cache:] /SET/ # can also be set on command line
# check- does the cached file exist?
{
# yes - so we use mailtrainer.crm to do the training
match (:: :fileid:) /X-CRM114-CacheID: ([[:graph:]]+)/
syscall () (:tmp:) /ls :*:text_cache:\/texts\/:*:fileid:/
match [:tmp:] /:*:fileid:/
# remember this file on a permanent basis
syscall /ln :*:text_cache:\/texts\/:*:fileid:
:*:text_cache:\/known_spam\/:*:fileid: /
#output /DOING: crm mailtrainer.crm
--spam=:*:text_cache:\/known_spam\/:*:fileid: --good=:*:text_cache:\/empty\/ /
() (:mailtrainer_output:)
syscall /crm mailtrainer.crm --fileprefix :*:fileprefix:
--spam=:*:text_cache:\/known_spam\/:*:fileid: --good=:*:text_cache:\/empty\/ /
() (:mailtrainer_output:)
#output /mailtrainer output: ':*:mailtrainer_output:'\n/
# and remove it from the prob_* directories, as
# now it's known
syscall /rm -rf :*:text_cache:\/prob_spam\/:*:fileid:/
syscall /rm -rf :*:text_cache:\/prob_good\/:*:fileid:/
#
# now it's trained; put in a marker in the headers
call /:mungmail_delete:/ [X-CRM114-Status: ]
call /:mungmail_delete:/ [X-CRM114-Unsure: ]
call /:mungmail_add:/ [X-CRM114-Action: LEARNED AND CACHED
SPAM]
# Insert the training report in the msgbody, if desired
{
match [:add_mailtrainer_report:] /yes/
match (:: :firstline:) /.*(.)/
match (:: :firstline:) /\n\n()/
alter (:firstline:) / :*:mailtrainer_output:\n ------ \n/
}
accept
exit /:*:accepted_mail_exit_code:/
}
alius
{
call /:mungmail_add:/ [X-CRM114-ERROR: No cached text with that
cacheID, so nothing done! ]
accept
exit /:*:accepted_mail_exit_code:/
}
}
{
# Not cached...
#
# Verify that we need to learn this first (TOE strategy)
classify <:*:clf:> [:text:] /:*:lcr:/ \
(:*:fileprefix::*:spamcss: :*:fileprefix::*:nonspamcss: )\
(:classify_status:)
match [:classify_status:] <nomultiline> \
/^#0.* pR: ([-. 0-9]+)/ (:: :pr:)
eval /:@: :*:pr: < :*:thick_threshold: : /
#
# write out the pre-mutilation text, with newlines
#
output [:*:fileprefix:spamtext.txt] <append> /\n\n:*:cmd_txt: \n/
learn < :*:clf:> (:*:fileprefix::*:spamcss:) [:text:] /:*:lcr:/
call /:mungmail_add:/ [X-CRM114-Action: LEARNED SPAM]
call /:mungmail_unique:/ [X-CRM114-Status: Good (Spam Learn)]
accept
exit /:*:accepted_mail_exit_code:/
}
alius
{
# Did the user specify "--force" on the command line?
match [:force:] /SET/
#
# write out the pre-mutilation text, with newlines
#
output [:*:fileprefix:spamtext.txt] <append> /\n\n:*:cmd_txt:
\n/
learn < :*:clf:> (:*:fileprefix::*:spamcss:) [:text:] /:*:lcr:/
call /:mungmail_add:/ [X-CRM114-Action: LEARNED SPAM (FORCED)]
call /:mungmail_unique:/ [X-CRM114-Status: Good (Spam Learn)]
accept
exit /:*:accepted_mail_exit_code:/
}
alius
{
call /:mungmail_add:/ [X-CRM114-Action: LEARN AS SPAM UNNECESSARY-
ALREADY CLASSIFIED CORRECTLY - NO ACTION TAKEN]
call /:mungmail_unique:/ [X-CRM114-Status: Good (Spam Learn)]
accept
exit /:*:accepted_mail_exit_code:/
}
}
{
# was it a command to learn something as an arbitrary type?
# Note: the files this generates don't get used for anything unless
# you use --spamcss and --nonspamcss in your own scripts.
#
# Note: these "learns" are a-priori "force", since we don't know
# what other .css files we should compare this text to.
#
match [:c:] /learn/
match (:z: :learnfile: :text:) [:m_text:] /:*:_nl:command [[:graph:]]+
learn ([[:graph:]]+)(.*)/
# and learn it
:learntofilehere:
output [:*:fileprefix::*:learnfile:text.txt] <append> /:*:text:/
learn < :*:clf:> (:*:fileprefix::*:learnfile:.css) [:text:] /:*:lcr:/
# syscall (:*:_dw:) (:_dw:) /formail -A "X-CRM114-Action: LEARNED
:*:file: :*:clf:"/
call /:mungmail_add:/ [X-CRM114-Action: LEARNED :*:file:]
call /:mungmail_unique:/ [X-CRM114-Status: Good (Learn)]
accept
exit /:*:accepted_mail_exit_code:/
}
}
#
#
# George's Circuit Breaker - if the mail has already been processed
# by CRM114, then send it directly to output, without further
# processing.
#
# WE DON'T RISK THIS ANY MORE - WITH ~ A MILLION USERS, WE'RE NOW
# A TARGET FOR SPAMMERS TO USE THIS HACK.
#{
# match /X-CRM114/
# alter (:classifier_reason:) \
# / This mail seems to have already been processed by CRM114. /
# alter (:stats:) / pR: 999.99 /
# goto /:looks_good:/
#}
# none of the above - classify this incoming mail instead.
# first according to priority action list,
# then according to whitelist,
# then according to blacklist,
# then according to the CRM sparse spectral classifier.
#
# check it against the priority action list- this list is
# of the form of a + or -, then a pattern. + means accept,
# - means reject. These are executed in order (which is
# different from whitelist or blacklist in that they occur
# in order given, not whitelist-then-blacklist. The priority
# action list is tried before whitelist or blacklist.
#
isolate (:priolist:)
input (:priolist:) [:*:fileprefix:priolist.mfp]
# reset matching on :priolist: to the start of the string
match [:priolist:] //
#
#
{
#... Grab the next regexturn the one-per-line patterns into a regex
match <fromend nomultiline> (:w: :pm: :pat:) [:priolist:] /(.)(.+)/
#... see if this regex matches the incoming mail
{
match <nomultiline> (:reason:) /:*:pat:/
# Yep, it matched... branch based on pm
#
{
match [:pm:] /[+]/
# put in a little tag saying why prio-listed
alter (:classifier_reason:) /** ACCEPT: CRM114 Priority
Whitelisted by: :*:reason: **:*:_nl:/
alter (:stats:) / pR: 999.99 /
goto /:looks_good:/
}
# No, we didn't have a +, was it a '-'?
{
match [:pm:] /[-]/
alter (:classifier_reason:) /** REJECT: CRM114 Priority
Blacklisted by: :*:reason: **:*:_nl:/
alter (:reject_address:) /:*:fail_priority_mail_to:/
{
match [:log_rejections:] /yes/
output [:*:fileprefix:rejected_by_blacklist.txt]
<append> /:*:_dw:/
}
alter (:stats:) / pR: -999.99 /
goto /:looks_bad:/
}
}
# Nope, didn't match as a priority... grab the next regex
liaf
}
#
#
# check it against the whitelist... load the whitelist...
{
isolate (:whitelist:)
input (:whitelist:) [:*:fileprefix:whitelist.mfp]
# reset matching on whitelist to start of string
match [:whitelist:] //
}
#
#
{
#... Grab the next regexturn the one-per-line patterns into a regex
match <fromend nomultiline> (:waste: :whregex:) [:whitelist:] /(.+)/
#... see if this regex matches the incoming mail
{
match <nomultiline> (:reason:) /:*:whregex:/
# Yep, it matched... whitelist this email
#
# put in a little tag saying why whitelisted:
alter (:classifier_reason:) /** ACCEPT: CRM114 Whitelisted by:
:*:reason: **:*:_nl:/
alter (:_dw:) /:*:_dw:\n\n ** CRM114 Whitelisted by: :*:reason:
**:*:_nl:/
alter (:stats:) / pR: 999.99 /
goto /:looks_good:/
}
# Nope, didn't match... grab the next regex and try again,
liaf
}
#
# No joy, maybe we should blacklist it.
#
# check it against the blacklist
{
isolate (:blacklist:)
input (:blacklist:) [:*:fileprefix:blacklist.mfp]
# reset matching on blacklist to start of string
match [:blacklist:] //
}
#
{
#... Grab the next regexturn the one-per-line patterns into a regex
match <fromend nomultiline> (:waste: :blregex:) [:blacklist:] /(.+)/
#... see if this regex matches the incoming mail
{
match <nomultiline> (:reason:) /:*:blregex:/
# Yep, it matched... blacklist this email
#
# put in a little tag saying why blacklisted
alter (:classifier_reason:) /** REJECT: CRM114 Blacklisted by:
:*:reason: ** :*:_nl:/
alter (:reject_address:) /:*:fail_blacklist_mail_to:/
{
match [:log_rejections:] /yes/
output [:*:fileprefix:rejected_by_blacklist.txt]
<append> /:*:_dw:/
}
alter (:stats:) / pR: -999.99 /
goto /:looks_bad:/
}
# Nope, didn't match... grab the next regex and try again
liaf
}
#
#
#
# End of blacklist processing.
#
#
# All else has failed- we now run our CLASSIFY algorithm
# to make our best guess.
#
#
{
# Run the CSS classifier against the "expanded" text -
# if it classifies as SPAM
# then reject it as SPAM.
#
{
classify <:*:clf:> ( :*:fileprefix::*:nonspamcss: |
:*:fileprefix::*:spamcss: ) ( :stats: ) [:m_text:] /:*:lcr:/
}
# Now we grab the pR and if it's greater than the minus
# threshold, we send it to "good". Otherwise, it goes to bad.
{
match <nomultiline> [:stats:] (:d: :pval:) /pR: (.*)/
eval /:@: :*:pval: > ( 0.0 - :*:thick_threshold: ) : /
alter (:classifier_reason:) /** ACCEPT: CRM114 PASS :*:clf:
Matcher ** :*:_nl::*:stats:/
goto /:looks_good:/
}
alter (:classifier_reason:) /** REJECT: CRM114 FAIL :*:clf: Matcher **
:*:_nl::*:stats:/
alter (:reject_address:) /:*:fail_SSM_mail_to:/
{
match [:log_rejections:] /yes/
output [:*:fileprefix:rejected_by_css.txt] <append> /:*:_dw:/
}
goto /:looks_bad:/
}
#
#
# Final wrap-up routines - dispose of the mail as appropriate.
#
{
:looks_bad:
# is this a :stats_only: run (i.e. for CAMRAM)
{
match [:stats_only:] /SET/
match <nomultiline> [:stats:] (:d: :pval:) /pR: (.*)/
output /:*:pval: :*:_nl:/
alter (:our_exit_code:) /:*:rejected_mail_exit_code:/
goto /:finish_up:/
}
# not stats_only.... we're doing major output.
# save unprocessed text by symlink in the text cache if needed.
{
match [:text_cache:] /./
syscall () () /ln :*:text_cache:\/texts\/:*:cacheid:
:*:text_cache:\/prob_spam\/:*:cacheid:/
}
# and write out the long-form message too.
{
{
match [:add_headers:] /yes/
{
match <nomultiline> [:stats:] (:pr:) /pR: .*$/
}
call /:mungmail_add:/ [X-CRM114-Version: :*:_crm_version:
MF-:*:_pgm_hash: ]
call /:mungmail_unique:/ [X-CRM114-CacheID: :*:cacheid: ]
call /:mungmail_unique:/ [X-CRM114-Status: SPAM ( :*:pr: )]
}
#
# Now, get the Subject: line. If none, make one.
{
{
match (:subject_line: :subj_text:) <nocase nomultiline> \
/^Subject: (.*)/
}
alius
{
match (:end_of_headers:) /\n\n/
alter (:end_of_headers:) /\nSubject: (none)\n\n/
match (:subject_line: :subj_text:) <nomultiline>
/^Subject: (.*)/ }
}
{
#
# If we are re-sending this, we want to de-fang the
# subject, otherwise we don't.
match [:reject_address:] /[a-zA-Z0-9]/
# Paolo P. suggests this alteration to avoid subversion
# by enclosing an alternate target in "marks". We always
# have to do this.
{
match (:dq:) [:subj_text:] /\$/
alter (:dq:) /USD/
liaf
}
{
match (:dq:) [:subj_text:] /[^-a-zA-Z0-9!., ]/
alter (:dq:) //
liaf
}
#
# We isolate subj_text here, so if later syscalls move
# things, the subject text used in "mail" is still OK.
isolate (:subj_text:)
}
#
# If the user asked for a spam-flagging string, put the flagging
# string into the subject.
#
{
match [:spam_flag_subject_string:] /./
alter (:subj_text:) \
/:*:spam_flag_subject_string: :*:subj_text:/
}
{
match [:add_extra_stuff:] /text/
# get rid of any first-column 'From's as they are message
breaks!
# this isn't necessary if we're mailing to someplace else...
{
match (:f:) <nomultiline> [:m_text:] /^From/
alter (:f:) />:*:f:/
liaf
}
alter (:_dw:) /:*:_dw:-=-Extra Stuff-=-\n\n:*:m_text: -0-0-0-
:*:_nl:/
}
{
match [:add_extra_stuff:] /attachment/
# get rid of any first-column 'From's as they are message
breaks!
# this isn't necessary if we're mailing to someplace else...
{
match (:f:) <nomultiline> [:m_text:] /^From/
alter (:f:) / :*:f:/
liaf
}
isolate (:content_type:) //
# syscall (:*:_dw:) (:content_type:) /formail -X "Content-Type"/
call /:mungmail_extract:/ [Content-type] (:content_type:)
isolate (:content_transfer_encoding:) //
# syscall (:*:_dw:) (:content_transfer_encoding:) /formail -X
"Content-Transfer-Encoding"/
call /:mungmail_extract:/ [Content-Transfer-Encoding]
(:content_transfer_encoding:)
# syscall (:*:_dw:) (:_dw:) /formail -A "MIME-Version: 1.0"/
# syscall (:*:_dw:) (:_dw:) /formail -A "Content-Type:
multipart\/mixed\;
boundary=Attachment_Quote_Boundary_1234567890\n--Attachment_Quote_Boundary_1234567890\n:*:content_type::*:content_transfer_encoding:"/
call /:mungmail_add:/ ["Content-Type: multipart\/mixed\;
boundary=Attachment_Quote_Boundary_1234567890\n--Attachment_Quote_Boundary_1234567890\n:*:content_type::*:content_transfer_encoding:]
alter (:_dw:) /:*:_dw::*:_nl:\
--Attachment_Quote_Boundary_1234567890 :*:_nl:\
Content-Type: text\/plain :*:_nl:\
Content-Transfer-Encoding: quoted-printable \n\n\n:*:m_text:\
\n--Attachment_Quote_Boundary_1234567890--\n/
}
#
#
# Decide if we forward or if we just output it.
{
{
# if this match succeeds, we should forward-to-an-address?
# Yes, but only if we _have_ a forward-to address.
match [:reject_address:] /[a-zA-Z0-9]/
{
# -- put the classifier reason in as the first thing!
match [:add_verbose_stats:] /yes/
alter (:_dw:) /:*:_nl: :*:classifier_reason::*:_nl: :*:_dw: /
}
syscall (:*:_dw:) /mail :*:reject_address: -s ':*:subj_text:'/
}
alius
{
{
# -- put the classifier reason in at the end of the headers
match [:add_verbose_stats:] /yes/
match (:start_of_data:) /\n\n/
alter (:start_of_data:) /\n\n :*:classifier_reason: \n /
}
accept
}
}
}
alter (:our_exit_code:) /:*:rejected_mail_exit_code:/
goto /:finish_up:/
}
#
# and here's where we accept something as good email.
{
:looks_good:
# is this a :stats_only: run (i.e. for CAMRAM)
{
match [:stats_only:] /SET/
match <nomultiline> [:stats:] (:d: :pval:) /pR: (.*)/
output /:*:pval: :*:_nl:/
alter (:our_exit_code:) /:*:accepted_mail_exit_code:/
goto /:finish_up:/
}
# Not stats-only; do the full output thing.
# save unprocessed txt by symlink in the text cache if needed.
{
match [:text_cache:] /./
syscall () () /ln :*:text_cache:\/texts\/:*:cacheid:
:*:text_cache:\/prob_good\/:*:cacheid:/
}
# and generate up a pretty mail-out report.
{
match [:add_verbose_stats:] /yes/
alter (:_dw:) /:*:_dw: :*:_nl: :*:classifier_reason: :*:_nl:/
}
{
match [:add_headers:] /yes/
{
match <nomultiline> [:stats:] (:pr:) /pR: .*$/
}
# syscall (:*:_dw:) (:_dw:) /formail -A "X-CRM114-Version:
:*:_crm_version: MF-:*:_pgm_hash: " -A "X-CRM114-Status: Good ( :*:pr: \)"/
call /:mungmail_add:/ [X-CRM114-Version: :*:_crm_version:
MF-:*:_pgm_hash: [:*:pr:]]
call /:mungmail_unique:/ [X-CRM114-CacheID: :*:cacheid: ]
call /:mungmail_unique:/ [X-CRM114-Status: Good ( :*:pr: )]
{
# Maybe we need to tag it as unsure? Note that since mail
# that scores > -thresh (but still < 0)goes out the "good" pipe,
# some "spammy" email might come through here.
match <nomultiline> [:stats:] (:d: :pval:) /pR: (.*)/
eval /:@: :*:pval: < :*:thick_threshold: :/
call /:mungmail_unique:/ [X-CRM114-Status: UNSURE (:*:pval:)
This message is 'unsure'; please train it! ]
}
}
{
match [:add_extra_stuff:] /text/
# get rid of any first-column 'From's as they are message breaks!
# this isn't necessary if we're mailing to someplace else...
{
match (:f:) <nomultiline> [:m_text:] /^From/
alter (:f:) / :*:f:/
liaf
}
alter (:_dw:) /:*:_dw:-=-Extra Stuff-=-\n\n :*:m_text: -0-0-0- \n/
}
{
match [:add_extra_stuff:] /attachment/
# get rid of any first-column 'From's as they are message breaks!
# this isn't necessary if we're mailing to someplace else...
{
match (:f:) <nomultiline> [:m_text:] /^From/
alter (:f:) / :*:f:/
liaf
}
isolate (:content_type:) //
# syscall (:*:_dw:) (:content_type:) /formail -X "Content-Type"/
call /:mungmail_extract:/ (:content_type:) [Content-Type]
isolate (:content_transfer_encoding:) //
# syscall (:*:_dw:) (:content_transfer_encoding:) /formail -X
"Content-Transfer-Encoding"/
call /:mungmail_extract:/ (:content_transfer_encoding:)
[Content-Transfer-Encoding:]
# syscall (:*:_dw:) (:_dw:) /formail -A "MIME-Version: 1.0"/
call /:mungmail_add:/ [MIME-Version: 1.0]
# syscall (:*:_dw:) (:_dw:) /formail -A "Content-Type:
multipart\/mixed\;
boundary=Attachment_Quote_Boundary_1234567890\n--Attachment_Quote_Boundary_1234567890\n:*:content_type::*:content_transfer_encoding:"/
# syscall (:*:_dw:) (:_dw:) /formail -U "Content-Type"/
call /:mungmail_unique:/ [Content-Type:]
alter (:_dw:) /:*:_dw::*:_nl:\
--Attachment_Quote_Boundary_1234567890 :*:_nl:\
Content-Type: text\/plain :*:_nl:\
Content-Transfer-Encoding: quoted-printable \n\n\n:*:m_text:\
\n--Attachment_Quote_Boundary_1234567890--\n/
}
accept
alter (:our_exit_code:) /:*:accepted_mail_exit_code:/
goto /:finish_up:/
}
#
# Here's where we finish up processing in all the paths.
#
:finish_up:
{
# ---- should we consider automatic training?
match [:automatic_training:] /yes/
# bounce out if we've already auto-trained this email
match <absent> /AUTO-TRAINED/
isolate (:msghash:)
hash (:msghash:) /:*:_dw:/
# pick one in 16- here, if the second-to-last digit is a 0
match [:msghash:] /......0./
#
# out put autotraining...
# Yep... we should use this for autotraining
# do we auto-train on acceptance?
{
match [:classifier_reason:] /ACCEPT/
# it wasn't spam... autotrain it "nonspam"
output [:*:fileprefix:nonspamtext.txt] <append> /:*:text:/
learn <microgroom> (:*:fileprefix::*:nonspamcss:) [:m_text:]
/:*:lcr:/
goto /:autotrain_finish:/
}
# or do we autotran on rejection
{
match [:classifier_reason:] /REJECT/
# it was spam... autotrain it "spam"
output [:*:fileprefix:spamtext.txt] <append> /:*:text:/
learn <microgroom> (:*:fileprefix::*:spamcss:) [:m_text:]
/:*:lcr:/
goto /:autotrain_finish:/
}
:autotrain_finish:
{
{
match [:autotrain_address:] /../
# syscall (:*:classifier_reason: :*:_nl: :*:_dw:) /mail -s
"AUTO-TRAINED email - please check" :*:autotrain_address:/
}
alius
{
# there was no autotrain address, so we just accept it.
match (:subj:) /Subject:/
alter (:subj:) /Subject: AUTO_TRAINED- please check! .../
accept
}
}
}
:exit_here:
exit /:*:our_exit_code:/
###################################################
#
#
# This is Mungmail - these are the replacement routines for
# formail(), to remove dependency on formail() being in every
# distribution
#
#
# Add a new header
:mungmail_add: (:new_header:)
{
# Grab the current headers
call /:mungmail_grab_current_headers:/
alter (:current_headers:) /:*:current_headers::*:new_header:\n/
return
}
#
# extract a header (first of them found)
#
:mungmail_extract: (:header_name:)
{
# Extract the header with the given field type, and
# return that. Note that we add the colon here; don't
# put it into the desired_header string.
#
call /:mungmail_grab_current_headers:/
{
match [:current_headers:] <nocase> (:: :desired_header:) \
/(?:^|\n)(:*:header_name: *: ([^\n]|\n[[:space:]])*)/
return /:*:desired_header:/
}
return //
}
#
# delete all current headers of this type, insert ours instead.
#
:mungmail_delete: (:new_header:)
{
call /:mungmail_grab_current_headers:/
{
match (:new_header_type:) [:new_header:] /[[:graph:]]+/
}
#
# a LIAF-loop to delete any header (including continuations) that
# has a type that matches the new_header_type.
{
match [:current_headers:] (:kill_this_line:) \
/:*:new_header_type: ([^\n]|\n[[:space:]])*\n/
alter (:kill_this_line:) //
liaf
}
return
}
# delete all current headers of this type, insert ours instead.
#
:mungmail_unique: (:new_header:)
{
call /:mungmail_grab_current_headers:/
{
match (:new_header_type:) [:new_header:] /[[:graph:]]+/
}
call /:mungmail_delete:/ [:*:new_header:]
call /:mungmail_add:/ [:*:new_header:]
return
}
#
# Helper routine to get the current mail headers of :_dw:
#
:mungmail_grab_current_headers:
{
{
# Grab everything before the first \n\n
match (:: :current_headers:) /(([^\n]+\n)+)\n/
# output /-A-->:*:current_headers:<---\n/
return
}
# if we got here, it wasn't a real message (void body, and/or no
# doubled newline) but it might still have useful text anyway.
# Is there a final newline?
{
match (:current_headers:) /^.*\n$/
# output /-B-->:*:current_headers:<---\n/
return
}
# if we got to here, then there wasn't even a final newline.
# That's a violation of RFC, we'll add it.
{
alter (:_dw:) /:*:_dw:\n/
match (:current_headers:) /.+/
# output /-C-->:*:current_headers:<---\n/
return
}
fault / Couldn't manage to find the headers, though I tried really hard\n/
}
trap (:broken_program_message:) /.*/
{
accept
output /:*:_nl: Aw, crud. mailfilter.crm broke. Here's the error:
:*:_nl:/
output /:*:broken_program_message:/
output [stderr] /:*:_nl: ERROR: mailfilter.crm broke. Here's the
error: :*:_nl:/
output [stderr] /ERROR: :*:broken_program_message:/
}
exit /:*:program_fault_exit_code:/
