X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~mdw/git/exim-config/blobdiff_plain/aa935c91390b4ed5f4cab4e9d55e702bd828edfe..5d2f8b656ffbe668f50ac5354dc0b37210032576:/spam.m4

diff --git a/spam.m4 b/spam.m4
index 350dcd2..c51bc36 100644
--- a/spam.m4
+++ b/spam.m4
@@ -25,57 +25,87 @@ DIVERT(null)
 ###--------------------------------------------------------------------------
 ### Spam filtering.
 
-SECTION(global, policy)m4_dnl
-spamd_address = CONF_spamd_address CONF_spamd_port
-
-SECTION(routers, allspam)m4_dnl
-## If we're verifying an address and the recipient has a `~/.mail/spam-limit'
-## file, then look up the recipient and sender addresses to find a plausible
-## limit and insert it into the `address_data' where the RCPT ACL can find
-## it.  This router always declines, so it doesn't affect the overall outcome
-## of the verification.
-fetch_spam_limit:
+## The Exim documentation tells lies.
+##
+## : *${run{*<_command_>* *<_args_>*}{*<_string1_>*}{*<_string2_>*}}*
+## :     The command and its arguments are first expanded separately, [...]
+##
+## They aren't.  The whole command-and-args are expanded together, and then
+## split at unquoted spaces.  This unpleasant hack sorts out the mess.
+m4_define(<:SHQUOTE:>, <:"${rxquote:$1}":>)
+
+## Utilities for collecting spam limits.
+m4_define(<:SPAMLIMIT_CHECK:>,
+	<:${if match{$1}{\N^-?[0-9]+$\N} {spam_limit=$1} {}}:>)
+
+m4_define(<:SPAMLIMIT_ROUTER:>,
+<:$1:
 	driver = redirect
 	data = :unknown:
 	verify_only = true
-	local_part_suffix = CONF_user_suffix_list
-	local_part_suffix_optional = true
-	check_local_user
-	address_data = \
+	condition = ${if !eq{$acl_c_mode}{submission}}
+	condition = ${extract{spam_limit}{$address_data}{false}{true}}:>)
+
+m4_define(<:SPAMLIMIT_SET:>,
+	<:address_data = \
 		${if def:address_data {$address_data}{}} \
-		${if exists {CONF_userconf_dir/spam-limit} \
-		     {${lookup {$local_part_prefix\
-					$local_part\
-					$local_part_suffix\
-					@$domain/\
-					$sender_address} \
-			       nwildlsearch {CONF_userconf_dir/spam-limit} \
-			       {spam_limit=$value} \
-			       {}}} \
-		     {}} \
-		${if exists {CONF_userconf_dir/spam-limit.userv} \
-		     {${run {timeout 5s -- \
-				userv $local_part exim-spam-limit \
-					$sender_address \
-					$local_part_prefix \
-					$local_part \
-					$local_part_suffix \
-					@$domain} \
-			    {${if match{$value}{\N^[0-9]+$\N} \
-				  {spam_limit=$value} \
-				  {}}} \
-			    {}}} \
-		     {}}
+		m4_ifelse(<:$2:>, <::>, <::>, <:$2 \
+		:>)$1:>)
+
+m4_define(<:SPAMLIMIT_LOOKUP:>,
+	<:condition = ${if exists{$1}}
+	SPAMLIMIT_SET(<:${lookup {$2@$3/$4} nwildlsearch {$1} \
+			       {SPAMLIMIT_CHECK(<:$value:>)}}:>, <:$5:>):>)
+
+m4_define(<:SPAMLIMIT_USERV:>,
+	<:SPAMLIMIT_SET(<:${run {/usr/bin/timeout 5s \
+					/usr/bin/userv CONF_userv_opts \
+					SHQUOTE($1) exim-spam-limit \
+					SHQUOTE($4) \
+					SHQUOTE($2) SHQUOTE(@$3)} \
+				{SPAMLIMIT_CHECK(<:$value:>)}}:>, <:$5:>):>)
+
+m4_define(<:GET_ADDRDATA:>,
+	<:extract{<:$1:>}{${if def:address_data{$address_data}{}}}:>)
+
+SECTION(global, policy)m4_dnl
+spamd_address = CONF_spamd_address CONF_spamd_port
 
 SECTION(acl, rcpt-hooks)m4_dnl
 	## Do per-recipient spam-filter processing.
 	require	 acl = rcpt_spam
 
 SECTION(acl, misc)m4_dnl
-rcpt_spam:
+skip_spam_check:
 
-	## If the client is trusted, don't bother with any of this.
+	## If the client is trusted, or this is a new submission, don't
+	## bother with any of this.  We will have verified the sender
+	## fairly aggressively before granting this level of trust.
 	accept	 hosts = +trusted
+	accept	 condition = ${if eq{$acl_c_mode}{submission}}
+
+	## If all domains have disabled spam checking then don't check.
+	accept	!condition = $acl_c_spam_check_domain
+
+	## Otherwise we should check.
+	deny
+
+rcpt_spam:
+
+	## If this is a virtual domain, and it says `spam-check=no', then we
+	## shouldn't check spam.  But we can't check domains at DATA time, so
+	## instead we must track whether all recipients have disabled
+	## checking.
+	warn	!domains = ${if exists{CONF_sysconf_dir/domains.conf} \
+			 {partial0-lsearch; CONF_sysconf_dir/domains.conf} \
+			 {}}
+		 set acl_c_spam_check_domain = true
+	warn	!condition = $acl_c_spam_check_domain
+		 condition = DOMKV(spam-check, {${expand:$value}}{true})
+		 set acl_c_spam_check_domain = true
+
+	## See if we should do this check.
+	accept	 acl = skip_spam_check
 
 	## Always accept mail to `postmaster'.  Currently this is not
 	## negotiable; maybe a tweak can be added to `domains.conf' if
@@ -87,12 +117,17 @@ rcpt_spam:
 	## during recipient verification.  (This just saves duplicating this
 	## enormous expression.)
 	warn	 set acl_m_this_spam_limit = \
-			${sg {${extract {spam_limit} \
-					{${if def:address_data \
-					      {$address_data}{}}} \
-					{$value}{nil}}} \
+			${sg {${GET_ADDRDATA(spam_limit){$value}{nil}}} \
 			     {^(|.*\\D.*)\$}{CONF_spam_max}}
 
+	warn	 condition = ${GET_ADDRDATA(user){true}{false}}
+		 set acl_m_spam_users = \
+			${if def:acl_m_spam_users {$acl_m_spam_users::}{}}\
+			${GET_ADDRDATA(user) \
+				{$value=${sg{$local_part@$domain}\
+					    {([!:])}{!\$1}}} \
+				fail}
+
 	## If there's a spam limit already established, and it's different
 	## from this user's limit, then the sender will have to try this user
 	## again later.
@@ -111,27 +146,23 @@ rcpt_spam:
 	## All done.
 	accept
 
-SECTION(acl, data-spam)m4_dnl
+SECTION(acl, data-hooks)m4_dnl
 	## Do spam checking.
 	require	 acl = data_spam
 
 SECTION(acl, misc)m4_dnl
 data_spam:
 
-	## If the client is trusted, don't bother with any of this.
-	accept	 hosts = +trusted
+	## See if we should do this check.
+	accept	 acl = skip_spam_check
 
 	## Check header validity.
 	require	 verify = header_syntax
 
 	## Check the message for spam, comparing to the configured limit.
-	deny	 spam = exim:true
-		 message = Tinned meat product detected ($spam_score)
-		 condition = ${if >{$spam_score_int}{$acl_m_spam_limit} \
-				  {true}{false}}
+	warn	 spam = exim:true
 
-	## Insert headers from the spam check now that we've decided to
-	## accept the message.
+	## Format some reporting stuff.
 	warn
 
 		 ## Convert the limit (currently 10x fixed point) into a
@@ -178,14 +209,28 @@ data_spam:
 		 ## Undo the escaping.
 		 set acl_m_spam_tests = ${sg{$acl_m_spam_tests}{!(.)}{\$1}}
 
-		 ## Insert the headers.
-		 add_header = X-SpamAssassin-Score: \
+	## If we've decided to reject, then leave a dropping in the log file
+	## so that users can analyse rejections for incoming messages, and
+	## tell the sender to get knotted.
+	deny	 message = Tinned meat product detected ($spam_score)
+		 log_message = Spam rejection \
+			score=$spam_score \
+			limit=$acl_m_spam_limit_presentation \
+			tests=$acl_m_spam_tests \
+			users=$acl_m_spam_users
+		 condition = ${if >{$spam_score_int}{$acl_m_spam_limit} \
+				  {true}{false}}
+
+	## Insert headers from the spam check now that we've decided to
+	## accept the message.
+	warn
+		 ADD_HEADER(<:X-CONF_header_token-SpamAssassin-Score: \
 			$spam_score/$acl_m_spam_limit_presentation \
-			($spam_bar)
-		 add_header = X-SpamAssassin-Status: \
+			($spam_bar):>)
+		 ADD_HEADER(<:X-CONF_header_token-SpamAssassin-Status: \
 			score=$spam_score, \
 			limit=$acl_m_spam_limit_presentation, \n\t\
-			tests=$acl_m_spam_tests
+			tests=$acl_m_spam_tests:>)
 
 	## We're good.
 	accept