Server IP : 85.214.239.14 / Your IP : 3.145.37.211 Web Server : Apache/2.4.62 (Debian) System : Linux h2886529.stratoserver.net 4.9.0 #1 SMP Tue Jan 9 19:45:01 MSK 2024 x86_64 User : www-data ( 33) PHP Version : 7.4.18 Disable Function : pcntl_alarm,pcntl_fork,pcntl_waitpid,pcntl_wait,pcntl_wifexited,pcntl_wifstopped,pcntl_wifsignaled,pcntl_wifcontinued,pcntl_wexitstatus,pcntl_wtermsig,pcntl_wstopsig,pcntl_signal,pcntl_signal_get_handler,pcntl_signal_dispatch,pcntl_get_last_error,pcntl_strerror,pcntl_sigprocmask,pcntl_sigwaitinfo,pcntl_sigtimedwait,pcntl_exec,pcntl_getpriority,pcntl_setpriority,pcntl_async_signals,pcntl_unshare, MySQL : OFF | cURL : OFF | WGET : ON | Perl : ON | Python : ON | Sudo : ON | Pkexec : OFF Directory : /proc/3/root/proc/3/task/3/root/proc/3/task/3/root/usr/share/perl5/Mail/SpamAssassin/Conf/ |
Upload File : |
# <@LICENSE> # Licensed to the Apache Software Foundation (ASF) under one or more # contributor license agreements. See the NOTICE file distributed with # this work for additional information regarding copyright ownership. # The ASF licenses this file to you under the Apache License, Version 2.0 # (the "License"); you may not use this file except in compliance with # the License. You may obtain a copy of the License at: # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # </@LICENSE> =head1 NAME Mail::SpamAssassin::Conf::Parser - parse SpamAssassin configuration =head1 SYNOPSIS (see Mail::SpamAssassin) =head1 DESCRIPTION Mail::SpamAssassin is a module to identify spam using text analysis and several internet-based realtime blocklists. This class is used internally by SpamAssassin to parse its configuration files. Please refer to the C<Mail::SpamAssassin> documentation for public interfaces. =head1 STRUCTURE OF A CONFIG BLOCK This is the structure of a config-setting block. Each is a hashref which may contain these keys: =over 4 =item setting the name of the setting it modifies, e.g. "required_score". this also doubles as the default for 'command' (below). THIS IS REQUIRED. =item command The command string used in the config file for this setting. Optional; 'setting' will be used for the command if this is omitted. =item aliases An [aryref] of other aliases for the same command. optional. =item type The type of this setting: - $CONF_TYPE_NOARGS: must not have any argument, like "clear_headers" - $CONF_TYPE_STRING: string - $CONF_TYPE_NUMERIC: numeric value (float or int) - $CONF_TYPE_BOOL: boolean (0/no or 1/yes) - $CONF_TYPE_TEMPLATE: template, like "report" - $CONF_TYPE_ADDRLIST: list of mail addresses, like "welcomelist_from" - $CONF_TYPE_HASH_KEY_VALUE: hash key/value pair, like "describe" or tflags - $CONF_TYPE_STRINGLIST list of strings, stored as an array - $CONF_TYPE_IPADDRLIST list of IP addresses, stored as an array of SA::NetSet - $CONF_TYPE_DURATION a nonnegative time interval in seconds - a numeric value (float or int), optionally suffixed by a time unit (s, m, h, d, w), seconds are implied if unit is missing If this is set, and a 'code' block does not already exist, a 'code' block is assigned based on the type. In addition, the SpamAssassin test suite will validate that the settings do not 'leak' between users. Note that C<$CONF_TYPE_HASH_KEY_VALUE>-type settings require that the value be non-empty, otherwise they'll produce a warning message. =item code A subroutine to deal with the setting. ONE OF B<code> OR B<type> IS REQUIRED. The arguments passed to the function are C<($self, $key, $value, $line)>, where $key is the setting (*not* the command), $value is the value string, and $line is the entire line. There are two special return values that the B<code> subroutine may return to signal that there is an error in the configuration: C<$Mail::SpamAssassin::Conf::MISSING_REQUIRED_VALUE> -- this setting requires that a value be set, but one was not provided. C<$Mail::SpamAssassin::Conf::INVALID_VALUE> -- this setting requires a value from a set of 'valid' values, but the user provided an invalid one. C<$Mail::SpamAssassin::Conf::INVALID_HEADER_FIELD_NAME> -- this setting requires a syntactically valid header field name, but the user provided an invalid one. Any other values -- including C<undef> -- returned from the subroutine are considered to mean 'success'. It is good practice to set a 'type', if possible, describing how your settings are stored on the Conf object; this allows the SpamAssassin test suite to validate that the settings do not 'leak' between users. =item default The default value for the setting. may be omitted if the default value is a non-scalar type, which should be set in the Conf ctor. note for path types: using "__userstate__" is recommended for defaults, as it allows Mail::SpamAssassin module users who set that configuration setting, to receive the correct values. =item is_priv Set to 1 if this setting requires 'allow_user_rules' when run from spamd. =item is_admin Set to 1 if this setting can only be set in the system-wide config when run from spamd. (All settings can be used by local programs run directly by the user.) =back =cut package Mail::SpamAssassin::Conf::Parser; use Mail::SpamAssassin::Conf; use Mail::SpamAssassin::Constants qw(:sa); use Mail::SpamAssassin::Logger; use Mail::SpamAssassin::Util qw(untaint_var compile_regexp); use Mail::SpamAssassin::NetSet; use strict; use warnings; # use bytes; use re 'taint'; our @ISA = qw(); my $ARITH_EXPRESSION_LEXER = ARITH_EXPRESSION_LEXER; my $META_RULES_MATCHING_RE = META_RULES_MATCHING_RE; ########################################################################### sub new { my $class = shift; $class = ref($class) || $class; my ($conf) = @_; my $self = { 'conf' => $conf }; $self->{command_luts} = { }; bless ($self, $class); $self; } ########################################################################### sub register_commands { my($self, $arrref) = @_; my $conf = $self->{conf}; $self->set_defaults_from_command_list($arrref); $self->build_command_luts($arrref); push(@{$conf->{registered_commands}}, @{$arrref}); } sub set_defaults_from_command_list { my ($self, $arrref) = @_; my $conf = $self->{conf}; foreach my $cmd (@{$arrref}) { # note! exists, not defined -- we want to be able to set # "undef" default values. if (exists($cmd->{default})) { $conf->{$cmd->{setting}} = $cmd->{default}; } } } sub build_command_luts { my ($self, $arrref) = @_; my $conf = $self->{conf}; foreach my $cmd (@{$arrref}) { my $cmdname = $cmd->{command} || $cmd->{setting}; $self->{command_luts}->{$cmdname} = $cmd; if ($cmd->{aliases} && scalar @{$cmd->{aliases}} > 0) { foreach my $name (@{$cmd->{aliases}}) { $self->{command_luts}->{$name} = $cmd; } } } } ########################################################################### sub parse { my ($self, undef, $scoresonly) = @_; # leave $rules in $_[1] my $conf = $self->{conf}; $self->{scoresonly} = $scoresonly; # Language selection: # See http://www.gnu.org/manual/glibc-2.2.5/html_node/Locale-Categories.html # and http://www.gnu.org/manual/glibc-2.2.5/html_node/Using-gettextized-software.html my $lang = $ENV{'LANGUAGE'}; # LANGUAGE has the highest precedence but has a if ($lang) { # special format: The user may specify more than $lang =~ s/:.*$//; # one language here, colon separated. We use the } # first one only (lazy bums we are :o) $lang ||= $ENV{'LC_ALL'}; $lang ||= $ENV{'LC_MESSAGES'}; $lang ||= $ENV{'LANG'}; $lang ||= 'C'; # Nothing set means C/POSIX if ($lang =~ /^(C|POSIX)$/) { $lang = 'en_US'; # Our default language } else { $lang =~ s/[@.+,].*$//; # Strip codeset, modifier/audience, etc. } # (eg. .utf8 or @euro) # get fast-access handles on the command lookup tables my $lut = $self->{command_luts}; my %migrated_keys = map { $_ => 1 } @Mail::SpamAssassin::Conf::MIGRATED_SETTINGS; $self->{currentfile} = '(no file)'; $self->{linenum} = (); my $skip_parsing = 0; my @curfile_stack; my @if_stack; my @conf_lines = split (/\n/, $_[1]); my $line; $self->{if_stack} = \@if_stack; $self->{cond_cache} = { }; $self->{file_scoped_attrs} = { }; my $keepmetadata = $conf->{main}->{keep_config_parsing_metadata}; while (defined ($line = shift @conf_lines)) { local ($1); # bug 3838: prevent random taint flagging of $1 my $parse_error; # undef by default, may be overridden # don't count internal file start/end lines $self->{linenum}{$self->{currentfile}}++ if index($line, 'file ') != 0; if (index($line,'#') > -1) { # bug 5545: used to support testing rules in the ruleqa system if ($keepmetadata && $line =~ /^\#testrules/) { $self->{file_scoped_attrs}->{testrules}++; next; } # bug 6800: let X-Spam-Checker-Version also show what sa-update we are at if (index($line, '# UPD') == 0 && $line =~ /^\# UPDATE version (\d+)$/) { for ($self->{currentfile}) { # just aliasing, not a loop $conf->{update_version}{$_} = $1 if defined $_ && $_ ne '(no file)'; } } $line =~ s/(?<!\\)#.*$//; # remove comments $line =~ s/\\#/#/g; # hash chars are escaped, so unescape them } $line =~ s/^\s+//; # remove leading whitespace $line =~ s/\s+$//; # remove tailing whitespace next unless($line); # skip empty lines # handle i18n if (index($line, 'lang') == 0 && $line =~ s/^lang\s+(\S+)\s+//) { next if $lang !~ /^$1/i; } my($key, $value) = split(/\s+/, $line, 2); $key = lc $key; # convert all dashes in setting name to underscores. $key =~ tr/-/_/; $value = '' unless defined($value); # $key if/elsif blocks sorted by most commonly used if ($key eq 'endif') { if ($value ne '') { $parse_error = "config: '$key' must be standalone"; goto failed_line; } my $lastcond = pop @if_stack; if (!defined $lastcond) { $parse_error = "config: missing starting 'if' for '$key'"; goto failed_line; } $skip_parsing = $lastcond->{skip_parsing}; next; } elsif ($key eq 'ifplugin') { if ($value eq '') { $parse_error = "config: missing '$key' condition"; goto failed_line; } $self->handle_conditional ($key, "plugin ($value)", \@if_stack, \$skip_parsing); next; } elsif ($key eq 'if') { if ($value eq '') { $parse_error = "config: missing '$key' condition"; goto failed_line; } $self->handle_conditional ($key, $value, \@if_stack, \$skip_parsing); next; } elsif ($key eq 'file') { if ($value =~ /^start\s+(.+)$/) { dbg("config: parsing file $1"); push (@curfile_stack, $self->{currentfile}); $self->{currentfile} = $1; next; } elsif ($value =~ /^end\s/) { foreach (@if_stack) { my $msg = "config: unclosed '$_->{type}' found ". "in $self->{currentfile} (line $_->{linenum})"; $self->lint_warn($msg, undef); } $self->{file_scoped_attrs} = { }; @if_stack = (); $skip_parsing = 0; $self->{currentfile} = pop @curfile_stack; next; } else { $parse_error = "config: missing '$key' value"; goto failed_line; } } elsif ($key eq 'include') { if ($value eq '') { $parse_error = "config: missing '$key' value"; goto failed_line; } $value = $self->fix_path_relative_to_current_file($value); my $text = $conf->{main}->read_cf($value, 'included file'); unshift (@conf_lines, "file end $self->{currentfile}", split (/\n/, $text), "file start $self->{currentfile}"); next; } elsif ($key eq 'else') { if ($value ne '') { $parse_error = "config: '$key' must be standalone"; goto failed_line; } # TODO: if/else/else won't get flagged here :( if (!@if_stack) { $parse_error = "config: '$key' missing starting if"; goto failed_line; } # Check if we are blocked anywhere in previous if-stack (Bug 7848) if (grep { $_->{skip_parsing} } @if_stack) { $skip_parsing = 1; } else { $skip_parsing = !$skip_parsing; } next; } # preprocessing? skip all other commands next if $skip_parsing; if ($key eq 'require_version') { if ($value eq '') { $parse_error = "config: missing '$key' value"; goto failed_line; } # if it wasn't replaced during install, assume current version ... next if ($value eq "\@\@VERSION\@\@"); my $ver = $Mail::SpamAssassin::VERSION; # if we want to allow "require_version 3.0" be good for all # "3.0.x" versions: ## make sure it's a numeric value #$value += 0.0; ## convert 3.000000 -> 3.0, stay backward compatible ... #$ver =~ s/^(\d+)\.(\d{1,3}).*$/sprintf "%d.%d", $1, $2/e; #$value =~ s/^(\d+)\.(\d{1,3}).*$/sprintf "%d.%d", $1, $2/e; if ($ver ne $value) { my $msg = "config: configuration file '$self->{currentfile}' requires ". "version $value of SpamAssassin, but this is code version ". "$ver. Maybe you need to use ". "the -C switch, or remove the old config files? ". "Skipping this file."; warn $msg; $self->lint_warn($msg, undef); $skip_parsing = 1; } next; } my $cmd = $lut->{$key}; # we've either fallen through with no match, in which case this # if() will fail, or we have a match. if ($cmd) { if ($self->{scoresonly}) { # reading user config from spamd if ($cmd->{is_priv} && !$conf->{allow_user_rules}) { info("config: not parsing, 'allow_user_rules' is 0: $line"); goto failed_line; } if ($cmd->{is_admin}) { info("config: not parsing, administrator setting: $line"); goto failed_line; } } if (!$cmd->{code}) { if (! $self->setup_default_code_cb($cmd)) { goto failed_line; } } my $ret = &{$cmd->{code}} ($conf, $cmd->{setting}, $value, $line); next if !$ret; if ($ret eq $Mail::SpamAssassin::Conf::INVALID_VALUE) { $parse_error = "config: invalid '$key' value"; goto failed_line; } elsif ($ret eq $Mail::SpamAssassin::Conf::INVALID_HEADER_FIELD_NAME) { $parse_error = "config: invalid header field name"; goto failed_line; } elsif ($ret eq $Mail::SpamAssassin::Conf::MISSING_REQUIRED_VALUE) { $parse_error = "config: missing '$key' value"; goto failed_line; } else { next; } } # last ditch: try to see if the plugins know what to do with it if ($conf->{main}->call_plugins("parse_config", { key => $key, value => $value, line => $line, conf => $conf, user_config => $self->{scoresonly} })) { # a plugin dealt with it successfully. next; } failed_line: my $msg = $parse_error; my $is_error = 1; if (!$msg) { # use a default warning, if a more specific one wasn't output if ($migrated_keys{$key}) { # this key was moved into a plugin; non-fatal for lint $is_error = 0; $msg = "config: failed to parse line, now a plugin"; } else { # a real syntax error; this is fatal for --lint $msg = "config: failed to parse line"; } } if ($self->{currentfile} eq '(no file)') { $msg .= " in $self->{currentfile}: $line"; } else { $msg .= " in $self->{currentfile} ". "(line $self->{linenum}{$self->{currentfile}}): $line"; } $self->lint_warn($msg, undef, $is_error); } delete $self->{if_stack}; delete $self->{cond_cache}; delete $self->{linenum}; $self->lint_check(); $self->fix_tests(); delete $self->{scoresonly}; } sub handle_conditional { my ($self, $key, $value, $if_stack_ref, $skip_parsing_ref) = @_; my $conf = $self->{conf}; # If we have already successfully evaled the $value, # just do what we would do then if (exists $self->{cond_cache}{"$key $value"}) { push (@{$if_stack_ref}, { 'type' => $key, 'conditional' => $value, 'skip_parsing' => $$skip_parsing_ref, 'linenum' => $self->{linenum}{$self->{currentfile}} }); if ($self->{cond_cache}{"$key $value"} == 0) { $$skip_parsing_ref = 1; } return; } my @tokens = ($value =~ /($ARITH_EXPRESSION_LEXER)/og); my $eval = ''; foreach my $token (@tokens) { if ($token eq '(' || $token eq ')' || $token eq '!') { # using tainted subr. argument may taint the whole expression, avoid my $u = untaint_var($token); $eval .= $u . " "; } elsif ($token eq 'plugin') { # replace with a method call $eval .= '$self->cond_clause_plugin_loaded'; } elsif ($token eq 'can') { # replace with a method call $eval .= '$self->cond_clause_can'; } elsif ($token eq 'has') { # replace with a method call $eval .= '$self->cond_clause_has'; } elsif ($token eq 'version') { $eval .= $Mail::SpamAssassin::VERSION." "; } elsif ($token eq 'perl_version') { $eval .= $]." "; } elsif ($token eq 'local_tests_only') { $eval .= '($self->{conf}->{main}->{local_tests_only}?1:0) ' } elsif ($token =~ /^(?:\W{1,5}|[+-]?\d+(?:\.\d+)?)$/) { # using tainted subr. argument may taint the whole expression, avoid my $u = untaint_var($token); $eval .= $u . " "; } elsif ($token =~ /^\w[\w\:]+$/) { # class name # Strictly controlled form: if ($token =~ /^(?:\w+::){0,10}\w+$/) { # trunk Dmarc.pm was renamed to DMARC.pm # (same check also in Conf.pm loadplugin) if ($token eq 'Mail::SpamAssassin::Plugin::Dmarc') { $token = 'Mail::SpamAssassin::Plugin::DMARC'; } # backwards compatible - removed in 4.1 # (same check also in Conf.pm loadplugin) elsif ($token eq 'Mail::SpamAssassin::Plugin::WhiteListSubject') { $token = 'Mail::SpamAssassin::Plugin::WelcomeListSubject'; } my $u = untaint_var($token); $eval .= "'$u'"; } else { my $msg = "config: not allowed value '$token' ". "in $self->{currentfile} (line $self->{linenum}{$self->{currentfile}})"; $self->lint_warn($msg, undef); return; } } else { my $msg = "config: unparseable value '$token' ". "in $self->{currentfile} (line $self->{linenum}{$self->{currentfile}})"; $self->lint_warn($msg, undef); return; } } push (@{$if_stack_ref}, { 'type' => $key, 'conditional' => $value, 'skip_parsing' => $$skip_parsing_ref, 'linenum' => $self->{linenum}{$self->{currentfile}} }); if (eval $eval) { $self->{cond_cache}{"$key $value"} = 1; # leave $skip_parsing as-is; we may not be parsing anyway in this block. # in other words, support nested 'if's and 'require_version's } else { if ($@) { my $msg = "config: error parsing conditional ". "in $self->{currentfile} (line $self->{linenum}{$self->{currentfile}}): $eval ($@)"; warn $msg; $self->lint_warn($msg, undef, 0); # not fatal? } $self->{cond_cache}{"$key $value"} = 0; $$skip_parsing_ref = 1; } } # functions supported in the "if" eval: sub cond_clause_plugin_loaded { return 1 if $_[1] eq 'Mail::SpamAssassin::Plugin::RaciallyCharged'; # removed in 4.1 return $_[0]->{conf}->{plugins_loaded}->{$_[1]}; } sub cond_clause_can { my ($self, $method) = @_; if ($self->{currentfile} =~ q!\buser_prefs$! ) { warn "config: 'if can $method' not available in user_prefs"; return 0 } $self->cond_clause_can_or_has('can', $method); } sub cond_clause_has { my ($self, $method) = @_; $self->cond_clause_can_or_has('has', $method); } sub cond_clause_can_or_has { my ($self, $fn_name, $method) = @_; local($1,$2); if (!defined $method) { my $msg = "config: bad 'if' line, no argument to $fn_name() ". "in $self->{currentfile} (line $self->{linenum}{$self->{currentfile}})"; $self->lint_warn($msg, undef); } elsif ($method =~ /^(.*)::([^:]+)$/) { no strict "refs"; my($module, $meth) = ($1, $2); return 1 if $module->can($meth) && ( $fn_name eq 'has' || &{$method}() ); } else { my $msg = "config: bad 'if' line, cannot find '::' in $fn_name($method) ". "in $self->{currentfile} (line $self->{linenum}{$self->{currentfile}})"; $self->lint_warn($msg, undef); } return; } # Let's do some linting here ... # This is called from _parse(), BTW, so we can check for $conf->{tests} # easily before finish_parsing() is called and deletes it. # sub lint_check { my ($self) = @_; my $conf = $self->{conf}; if ($conf->{lint_rules}) { # Check for description and score issues in lint fashion while ( my $k = each %{$conf->{descriptions}} ) { if (!exists $conf->{tests}->{$k}) { dbg("config: description exists for non-existent rule $k"); } } while ( my($sk) = each %{$conf->{scores}} ) { if (!exists $conf->{tests}->{$sk}) { # bug 5514: not a lint warning any more dbg("config: score set for non-existent rule $sk"); } } } } # Iterate through tests and check/fix things sub fix_tests { my ($self) = @_; my $conf = $self->{conf}; my $would_log_dbg = would_log('dbg'); while ( my $k = each %{$conf->{tests}} ) { # we should set a default score for all valid rules... Do this here # instead of add_test because mostly 'score' occurs after the rule is # specified, so why set the scores to default, then set them again at # 'score'? # if ( ! exists $conf->{scores}->{$k} ) { # T_ rules (in a testing probationary period) get low, low scores my $set_score = index($k, 'T_') == 0 ? 0.01 : 1.0; $set_score = -$set_score if ( ($conf->{tflags}->{$k}||'') =~ /\bnice\b/ ); for my $index (0..3) { $conf->{scoreset}->[$index]->{$k} = $set_score; } } # loop through all the tests and if we are missing a description with debug # set, throw a note except for testing T_ or meta __ rules. if ($would_log_dbg && $k !~ m/^(?:T_|__)/i) { if ( ! exists $conf->{descriptions}->{$k} ) { dbg("config: no description set for rule $k"); } } } } ########################################################################### sub setup_default_code_cb { my ($self, $cmd) = @_; my $type = $cmd->{type}; if ($type == $Mail::SpamAssassin::Conf::CONF_TYPE_STRING) { $cmd->{code} = \&set_string_value; } elsif ($type == $Mail::SpamAssassin::Conf::CONF_TYPE_BOOL) { $cmd->{code} = \&set_bool_value; } elsif ($type == $Mail::SpamAssassin::Conf::CONF_TYPE_NUMERIC) { $cmd->{code} = \&set_numeric_value; } elsif ($type == $Mail::SpamAssassin::Conf::CONF_TYPE_HASH_KEY_VALUE) { $cmd->{code} = \&set_hash_key_value; } elsif ($type == $Mail::SpamAssassin::Conf::CONF_TYPE_ADDRLIST) { $cmd->{code} = \&set_addrlist_value; } elsif ($type == $Mail::SpamAssassin::Conf::CONF_TYPE_TEMPLATE) { $cmd->{code} = \&set_template_append; } elsif ($type == $Mail::SpamAssassin::Conf::CONF_TYPE_NOARGS) { $cmd->{code} = \&set_no_value; } elsif ($type == $Mail::SpamAssassin::Conf::CONF_TYPE_STRINGLIST) { $cmd->{code} = \&set_string_list; } elsif ($type == $Mail::SpamAssassin::Conf::CONF_TYPE_IPADDRLIST) { $cmd->{code} = \&set_ipaddr_list; } elsif ($type == $Mail::SpamAssassin::Conf::CONF_TYPE_DURATION) { $cmd->{code} = \&set_duration_value; } else { warn "config: unknown conf type $type!"; return 0; } return 1; } sub set_no_value { my ($conf, $key, $value, $line) = @_; unless (!defined $value || $value eq '') { return $Mail::SpamAssassin::Conf::INVALID_VALUE; } } sub set_numeric_value { my ($conf, $key, $value, $line) = @_; unless (defined $value && $value !~ /^$/) { return $Mail::SpamAssassin::Conf::MISSING_REQUIRED_VALUE; } unless ($value =~ /^ [+-]? \d+ (?: \. \d* )? \z/sx) { return $Mail::SpamAssassin::Conf::INVALID_VALUE; } # it is safe to untaint now that we know the syntax is a valid number $conf->{$key} = untaint_var($value) + 0; } sub set_duration_value { my ($conf, $key, $value, $line) = @_; local ($1,$2); unless (defined $value && $value !~ /^$/) { return $Mail::SpamAssassin::Conf::MISSING_REQUIRED_VALUE; } unless ($value =~ /^( \+? \d+ (?: \. \d* )? ) (?: \s* ([smhdw]))? \z/sxi) { return $Mail::SpamAssassin::Conf::INVALID_VALUE; } $value = $1; $value *= { s => 1, m => 60, h => 3600, d => 24*3600, w => 7*24*3600 }->{lc $2} if defined $2; # it is safe to untaint now that we know the syntax is a valid time interval $conf->{$key} = untaint_var($value) + 0; } sub set_bool_value { my ($conf, $key, $value, $line) = @_; unless (defined $value && $value !~ /^$/) { return $Mail::SpamAssassin::Conf::MISSING_REQUIRED_VALUE; } # bug 4462: allow yes/1 and no/0 for boolean values $value = lc $value; if ($value eq 'yes' || $value eq '1') { $value = 1; } elsif ($value eq 'no' || $value eq '0') { $value = 0; } else { return $Mail::SpamAssassin::Conf::INVALID_VALUE; } $conf->{$key} = $value; } sub set_string_value { my ($conf, $key, $value, $line) = @_; unless (defined $value && $value !~ /^$/) { return $Mail::SpamAssassin::Conf::MISSING_REQUIRED_VALUE; } $conf->{$key} = $value; # keep tainted } sub set_string_list { my ($conf, $key, $value, $line) = @_; unless (defined $value && $value !~ /^$/) { return $Mail::SpamAssassin::Conf::MISSING_REQUIRED_VALUE; } push(@{$conf->{$key}}, split(/\s+/, $value)); } sub set_ipaddr_list { my ($conf, $key, $value, $line) = @_; unless (defined $value && $value !~ /^$/) { return $Mail::SpamAssassin::Conf::MISSING_REQUIRED_VALUE; } foreach my $net (split(/\s+/, $value)) { $conf->{$key}->add_cidr($net); } $conf->{$key.'_configured'} = 1; } sub set_hash_key_value { my ($conf, $key, $value, $line) = @_; my($k,$v) = split(/\s+/, $value, 2); unless (defined $v && $v ne '') { return $Mail::SpamAssassin::Conf::MISSING_REQUIRED_VALUE; } $conf->{$key}->{$k} = $v; # keep tainted } sub set_addrlist_value { my ($conf, $key, $value, $line) = @_; unless (defined $value && $value !~ /^$/) { return $Mail::SpamAssassin::Conf::MISSING_REQUIRED_VALUE; } $conf->{parser}->add_to_addrlist ($key, split(/\s+/, $value)); # keep tainted } sub remove_addrlist_value { my ($conf, $key, $value, $line) = @_; unless (defined $value && $value !~ /^$/) { return $Mail::SpamAssassin::Conf::MISSING_REQUIRED_VALUE; } $conf->{parser}->remove_from_addrlist ($key, split(/\s+/, $value)); } sub set_template_append { my ($conf, $key, $value, $line) = @_; if ( $value =~ /^"(.*?)"$/ ) { $value = $1; } $conf->{$key} .= $value."\n"; # keep tainted } sub set_template_clear { my ($conf, $key, $value, $line) = @_; unless (!defined $value || $value eq '') { return $Mail::SpamAssassin::Conf::INVALID_VALUE; } $conf->{$key} = ''; } ########################################################################### sub finish_parsing { my ($self, $isuserconf) = @_; my $conf = $self->{conf}; # note: this function is called once for system-wide configuration # with $isuserconf set to 0, then again for user conf with $isuserconf set to 1. if (!$isuserconf) { $conf->{main}->call_plugins("finish_parsing_start", { conf => $conf }); } else { $conf->{main}->call_plugins("user_conf_parsing_start", { conf => $conf }); } # compile meta rules $self->compile_meta_rules(); $self->fix_priorities(); $self->fix_tflags(); dbg("config: finish parsing"); while (my ($name, $text) = each %{$conf->{tests}}) { my $type = $conf->{test_types}->{$name}; # Adjust priority -100 for net rules instead of default 0 my $priority = $conf->{priority}->{$name} ? $conf->{priority}->{$name} : ($conf->{tflags}->{$name}||'') =~ /\bnet\b/ ? -100 : 0; $conf->{priorities}->{$priority}++; # eval type handling if (($type & 1) == 1) { if (my ($function, $args) = ($text =~ /^(\w+)\((.*?)\)$/)) { my $argsref = $self->pack_eval_args($args); if (!defined $argsref) { $self->lint_warn("syntax error for eval function $name: $text"); next; } # Validate type my $expected_type = $conf->{eval_plugins_types}->{$function}; if (defined $expected_type && $expected_type != $type) { # Allow both body and rawbody if expecting body if (!($expected_type == $Mail::SpamAssassin::Conf::TYPE_BODY_EVALS && $type == $Mail::SpamAssassin::Conf::TYPE_RAWBODY_EVALS)) { my $estr = $Mail::SpamAssassin::Conf::TYPE_AS_STRING{$expected_type}; $self->lint_warn("wrong rule type defined for $name, expected '$estr'"); next; } } if ($type == $Mail::SpamAssassin::Conf::TYPE_BODY_EVALS) { $conf->{body_evals}->{$priority}->{$name} = [ $function, [@$argsref] ]; } elsif ($type == $Mail::SpamAssassin::Conf::TYPE_HEAD_EVALS) { $conf->{head_evals}->{$priority}->{$name} = [ $function, [@$argsref] ]; } elsif ($type == $Mail::SpamAssassin::Conf::TYPE_RBL_EVALS) { # We don't do priorities for $Mail::SpamAssassin::Conf::TYPE_RBL_EVALS # we also use the arrayref instead of the packed string $conf->{rbl_evals}->{$name} = [ $function, [@$argsref] ]; } elsif ($type == $Mail::SpamAssassin::Conf::TYPE_RAWBODY_EVALS) { $conf->{rawbody_evals}->{$priority}->{$name} = [ $function, [@$argsref] ]; } elsif ($type == $Mail::SpamAssassin::Conf::TYPE_FULL_EVALS) { $conf->{full_evals}->{$priority}->{$name} = [ $function, [@$argsref] ]; } #elsif ($type == $Mail::SpamAssassin::Conf::TYPE_URI_EVALS) { # $conf->{uri_evals}->{$priority}->{$name} = [ $function, [@$argsref] ]; #} else { $self->lint_warn("unknown type $type for $name: $text", $name); next; } } else { $self->lint_warn("syntax error for eval function $name: $text", $name); next; } } # non-eval tests else { if ($type == $Mail::SpamAssassin::Conf::TYPE_BODY_TESTS) { $conf->{body_tests}->{$priority}->{$name} = $text; } elsif ($type == $Mail::SpamAssassin::Conf::TYPE_HEAD_TESTS) { $conf->{head_tests}->{$priority}->{$name} = $text; } elsif ($type == $Mail::SpamAssassin::Conf::TYPE_META_TESTS) { # Handled by compile_meta_rules() } elsif ($type == $Mail::SpamAssassin::Conf::TYPE_URI_TESTS) { $conf->{uri_tests}->{$priority}->{$name} = $text; } elsif ($type == $Mail::SpamAssassin::Conf::TYPE_RAWBODY_TESTS) { $conf->{rawbody_tests}->{$priority}->{$name} = $text; } elsif ($type == $Mail::SpamAssassin::Conf::TYPE_FULL_TESTS) { $conf->{full_tests}->{$priority}->{$name} = $text; } elsif ($type == $Mail::SpamAssassin::Conf::TYPE_EMPTY_TESTS) { } else { $self->lint_warn("unknown type $type for $name: $text", $name); next; } } } $self->lint_trusted_networks(); if (!$isuserconf) { $conf->{main}->call_plugins("finish_parsing_end", { conf => $conf }); } else { $conf->{main}->call_plugins("user_conf_parsing_end", { conf => $conf }); } $conf->found_any_rules(); # before we might delete {tests} if (!$conf->{allow_user_rules}) { # free up stuff we no longer need delete $conf->{tests}; delete $conf->{priority}; #test_types are needed - see bug 5503 #delete $conf->{test_types}; } } # Returns all rulenames matching glob (FOO_*) sub expand_ruleglob { my ($self, $ruleglob, $rulename) = @_; my $expanded; if (exists $self->{ruleglob_cache}{$ruleglob}) { $expanded = $self->{ruleglob_cache}{$ruleglob}; } else { my $reglob = $ruleglob; $reglob =~ s/\?/./g; $reglob =~ s/\*/.*?/g; # Glob rules, but do not match ourselves.. my @rules = grep {/^${reglob}$/ && $_ ne $rulename} keys %{$self->{conf}->{scores}}; if (@rules) { $expanded = join('+', sort @rules); } else { $expanded = '0'; } } my $logstr = $expanded eq '0' ? 'no matches' : $expanded; dbg("rules: meta $rulename rules_matching($ruleglob) expanded: $logstr"); $self->{ruleglob_cache}{$ruleglob} = $expanded; return " ($expanded) "; } sub compile_meta_rules { my ($self) = @_; my (%meta, %meta_deps, %rule_deps); my $conf = $self->{conf}; foreach my $name (keys %{$conf->{tests}}) { next unless $conf->{test_types}->{$name} == $Mail::SpamAssassin::Conf::TYPE_META_TESTS; my $rule = $conf->{tests}->{$name}; # Expand meta rules_matching() before lexing $rule =~ s/${META_RULES_MATCHING_RE}/$self->expand_ruleglob($1,$name)/ge; # Lex the rule into tokens using a rather simple RE method ... my @tokens = ($rule =~ /$ARITH_EXPRESSION_LEXER/og); # Set the rule blank to start $meta{$name} = ''; # List dependencies that are meta tests in the same priority band $meta_deps{$name} = [ ]; # List all rule dependencies $rule_deps{$name} = [ ]; # Go through each token in the meta rule foreach my $token (@tokens) { # operator (triage, already validated by is_meta_valid) if ($token !~ tr/+&|()!<>=//c) { $meta{$name} .= "$token "; } # rule-like check for local_tests_only elsif ($token eq 'local_tests_only') { $meta{$name} .= '($_[0]->{main}->{local_tests_only}||0) '; } # ... rulename? elsif ($token =~ IS_RULENAME) { # Will end up later in a compiled sub called from do_meta_tests: # $_[0] = $pms # $_[1] = $h ($pms->{tests_already_hit}), $meta{$name} .= "(\$_[1]->{'$token'}||0) "; if (!exists $conf->{test_types}->{$token}) { dbg("rules: meta test $name has undefined dependency '$token'"); push @{$rule_deps{$name}}, $token; next; } if ($conf->{scores}->{$token} == 0) { # bug 5040: net rules in a non-net scoreset # there are some cases where this is expected; don't warn # in those cases. unless ((($conf->get_score_set()) & 1) == 0 && ($conf->{tflags}->{$token}||'') =~ /\bnet\b/) { dbg("rules: meta test $name has dependency '$token' with a zero score"); } } # If the token is another meta rule, add it as a dependency if ($conf->{test_types}->{$token} == $Mail::SpamAssassin::Conf::TYPE_META_TESTS) { push @{$meta_deps{$name}}, $token; } # Record all dependencies push @{$rule_deps{$name}}, $token; } # ... number or operator (already validated by is_meta_valid) else { $meta{$name} .= "$token "; } } } # Sort by length of dependencies list. It's more likely we'll get # the dependencies worked out this way. my @metas = sort { @{$meta_deps{$a}} <=> @{$meta_deps{$b}} } keys %meta; my $count; do { $count = $#metas; my %metas = map { $_ => 1 } @metas; # keep a small cache for fast lookups # Go through each meta rule we haven't done yet for (my $i = 0 ; $i <= $#metas ; $i++) { next if (grep( $metas{$_}, @{ $meta_deps{ $metas[$i] } })); splice @metas, $i--, 1; # remove this rule from our list } } while ($#metas != $count && $#metas > -1); # run until we can't go anymore # If there are any rules left, we can't solve the dependencies so complain my %unsolved_metas = map { $_ => 1 } @metas; # keep a small cache for fast lookups foreach my $rulename_t (@metas) { my $msg = "rules: excluding meta test $rulename_t, unsolved meta dependencies: ". join(", ", grep($unsolved_metas{$_}, @{ $meta_deps{$rulename_t} })); $self->lint_warn($msg); } foreach my $name (keys %meta) { if ($unsolved_metas{$name}) { $conf->{meta_tests}->{$name} = sub { 0 }; $rule_deps{$name} = [ ]; } if ($meta{$name} eq '( ) ') { # Bug 8061: # meta FOOBAR () considered a rule declaration to support rule_hits API or # other dynamic rules, only evaluated at finish_meta_rules unless got_hit. # Other style metas without dependencies will be evaluated immediately. $meta{$name} = '0'; # Evaluating () would result in undef } elsif (@{$rule_deps{$name}}) { $conf->{meta_dependencies}->{$name} = $rule_deps{$name}; foreach my $deprule (@{$rule_deps{$name}}) { $conf->{meta_deprules}->{$deprule}->{$name} = 1; } } else { $conf->{meta_nodeps}->{$name} = 1; } # Compile meta sub eval '$conf->{meta_tests}->{$name} = sub { '.$meta{$name}.'};'; # Paranoid check die "rules: meta compilation failed for $name: '$meta{$name}': $@" if ($@); } } sub fix_priorities { my ($self) = @_; my $conf = $self->{conf}; return unless $conf->{meta_dependencies}; # order requirement my $pri = $conf->{priority}; my $tflags = $conf->{tflags}; # sort into priority order, lowest first -- this way we ensure that if we # rearrange the pri of a rule early on, we cannot accidentally increase its # priority later. foreach my $rule (sort { $pri->{$a} <=> $pri->{$b} } keys %{$pri}) { # we only need to worry about meta rules -- they are the # only type of rules which depend on other rules my $deps = $conf->{meta_dependencies}->{$rule}; next unless (defined $deps); my $basepri = $pri->{$rule}; foreach my $dep (@$deps) { my $deppri = $pri->{$dep}; if (defined $deppri && $deppri > $basepri) { if ($basepri < -100 && ($tflags->{$dep}||'') =~ /\bnet\b/) { dbg("rules: $rule (pri $basepri) requires $dep (pri $deppri): fixed to -100 (net rule)"); $pri->{$dep} = -100; $conf->{priorities}->{-100}++; } else { dbg("rules: $rule (pri $basepri) requires $dep (pri $deppri): fixed"); $pri->{$dep} = $basepri; } } } } } sub fix_tflags { my ($self) = @_; my $conf = $self->{conf}; my $tflags = $conf->{tflags}; # Inherit net tflags from dependencies while (my($rulename,$deps) = each %{$conf->{meta_dependencies}}) { my $tfl = $tflags->{$rulename}||''; next if $tfl =~ /\bnet\b/; foreach my $deprule (@$deps) { if (($tflags->{$deprule}||'') =~ /\bnet\b/) { dbg("rules: meta $rulename inherits tflag net, depends on $deprule"); $tflags->{$rulename} = $tfl eq '' ? 'net' : "$tfl net"; last; } } } } # Deprecated function sub pack_eval_method { warn "deprecated function pack_eval_method() used\n"; return ('',undef); } sub pack_eval_args { my ($self, $args) = @_; return [] if $args =~ /^\s+$/; # bug 4419: Parse quoted strings, unquoted alphanumerics/floats, # unquoted IPv4 and IPv6 addresses, and unquoted common domain names. # s// is used so that we can determine whether or not we successfully # parsed ALL arguments. my @args; local($1,$2,$3); while ($args =~ s/^\s* (?: (['"]) (.*?) \1 | ( [\d\.:A-Za-z-]+? ) ) \s* (?: , \s* | $ )//x) { # DO NOT UNTAINT THESE ARGS # The eval function that handles these should do that as necessary, # we have no idea what acceptable arguments look like here. push @args, defined $2 ? $2 : $3; } if ($args ne '') { return undef; ## no critic (ProhibitExplicitReturnUndef) } return \@args; } ########################################################################### sub lint_trusted_networks { my ($self) = @_; my $conf = $self->{conf}; # validate trusted_networks and internal_networks, bug 4760. # check that all internal_networks are listed in trusted_networks # too. do the same for msa_networks, but check msa_networks against # internal_networks if trusted_networks aren't defined my ($nt, $matching_against); if ($conf->{trusted_networks_configured}) { $nt = $conf->{trusted_networks}; $matching_against = 'trusted_networks'; } elsif ($conf->{internal_networks_configured}) { $nt = $conf->{internal_networks}; $matching_against = 'internal_networks'; } else { return; } foreach my $net_type ('internal_networks', 'msa_networks') { next unless $conf->{"${net_type}_configured"}; next if $net_type eq $matching_against; my $replace_nets; my @valid_net_list; my $net_list = $conf->{$net_type}; foreach my $net (@{$net_list->{nets}}) { # don't check to see if an excluded network is included - that's senseless if (!$net->{exclude} && !$nt->contains_net($net)) { my $msg = "$matching_against doesn't contain $net_type entry '". ($net->{as_string})."'"; $self->lint_warn($msg, undef); # complain $replace_nets = 1; # and omit it from the new internal set } else { push @valid_net_list, $net; } } if ($replace_nets) { # something was invalid. replace the old nets list with a fixed version # (which may be empty) $net_list->{nets} = \@valid_net_list; } } } ########################################################################### sub add_test { my ($self, $name, $text, $type) = @_; my $conf = $self->{conf}; # Don't allow invalid names ... if ($name !~ IS_RULENAME) { $self->lint_warn("config: error: rule '$name' has invalid characters ". "(not Alphanumeric + Underscore + starting with a non-digit)\n", $name); return; } # Also set a hard limit for ALL rules (rule names longer than 40 # characters throw warnings). Check this separately from the above # pattern to avoid vague error messages. if (length $name > 100) { $self->lint_warn("config: error: rule '$name' is too long ". "(recommended maximum length is 22 characters)\n", $name); return; } # Warn about, but use, long rule names during --lint if ($conf->{lint_rules}) { if (length($name) > 40 && $name !~ /^__/ && $name !~ /^T_/) { $self->lint_warn("config: warning: rule name '$name' is over 40 chars ". "(recommended maximum length is 22 characters)\n", $name); } } # parameter to compile_regexp() my $ignore_amre = $self->{conf}->{lint_rules} || $self->{conf}->{ignore_always_matching_regexps}; # all of these rule types are regexps if ($type == $Mail::SpamAssassin::Conf::TYPE_BODY_TESTS || $type == $Mail::SpamAssassin::Conf::TYPE_URI_TESTS || $type == $Mail::SpamAssassin::Conf::TYPE_RAWBODY_TESTS || $type == $Mail::SpamAssassin::Conf::TYPE_FULL_TESTS) { $self->parse_captures($name, \$text); my ($rec, $err) = compile_regexp($text, 1, $ignore_amre); if (!$rec) { $self->lint_warn("config: invalid regexp for $name '$text': $err", $name); return; } $conf->{test_qrs}->{$name} = $rec; } elsif ($type == $Mail::SpamAssassin::Conf::TYPE_HEAD_TESTS) { # If redefining header test, clear out opt hashes so they don't leak to # the new test. There are separate hashes for options as it saves lots # of memory (exists, neg, if-unset are rarely used). if (exists $conf->{tests}->{$name}) { delete $conf->{test_opt_exists}->{$name}; delete $conf->{test_opt_unset}->{$name}; delete $conf->{test_opt_neg}->{$name}; } local($1,$2,$3); # RFC 5322 section 3.6.8, ftext printable US-ASCII chars not including ":" # no re "strict"; # since perl 5.21.8: Ranges of ASCII printables... if ($text =~ /^exists:(.*)/) { my $hdr = $1; # $hdr used in eval text, validate carefully if ($hdr !~ /^[\w.-]+:?$/) { $self->lint_warn("config: invalid head test $name header: $hdr"); return; } $hdr =~ s/:$//; $conf->{test_opt_header}->{$name} = $hdr; $conf->{test_opt_exists}->{$name} = 1; } else { # $hdr used in eval text, validate carefully # check :addr etc header options if ($text !~ /^([\w.-]+(?:\:|(?:\:[a-z]+){1,2})?)\s*([=!]~)\s*(.+)$/) { $self->lint_warn("config: invalid head test $name: $text"); return; } my ($hdr, $op, $pat) = ($1, $2, $3); $hdr =~ s/:$//; if ($hdr =~ /:(?!(?:raw|addr|name|host|domain|ip|revip|first|last)\b)/i) { $self->lint_warn("config: invalid header modifier for $name: $hdr", $name); return; } if ($pat =~ s/\s+\[if-unset:\s+(.+)\]$//) { $conf->{test_opt_unset}->{$name} = $1; } $self->parse_captures($name, \$pat); my ($rec, $err) = compile_regexp($pat, 1, $ignore_amre); if (!$rec) { $self->lint_warn("config: invalid regexp for $name '$pat': $err", $name); return; } $conf->{test_qrs}->{$name} = $rec; $conf->{test_opt_header}->{$name} = $hdr; $conf->{test_opt_neg}->{$name} = 1 if $op eq '!~'; } } elsif ($type == $Mail::SpamAssassin::Conf::TYPE_META_TESTS) { if ($self->is_meta_valid($name, $text)) { # Untaint now once and not repeatedly later $text = untaint_var($text); } else { return; } } elsif (($type & 1) == 1) { # *_EVALS # create eval_to_rule mappings if (my ($function) = ($text =~ m/(.*?)\s*\(.*?\)\s*$/)) { push @{$conf->{eval_to_rule}->{$function}}, $name; } } $conf->{tests}->{$name} = $text; $conf->{test_types}->{$name} = $type; if ($name =~ /^AUTOLEARNTEST/) { dbg("config: auto-learn: $name has type $type = $conf->{test_types}->{$name} during add_test\n"); } $conf->{priority}->{$name} ||= 0; if ($conf->{main}->{keep_config_parsing_metadata}) { # {source_file} eats lots of memory and is unused unless # keep_config_parsing_metadata is set (ruleqa stuff) $conf->{source_file}->{$name} = $self->{currentfile}; $conf->{if_stack}->{$name} = $self->get_if_stack_as_string(); if ($self->{file_scoped_attrs}->{testrules}) { $conf->{testrules}->{$name} = 1; # used in build/mkupdates/listpromotable } } # if we found this rule in a user_prefs file, it's a user rule -- note that # we may need to recompile the rule code for this type (if they've already # been compiled, e.g. in spamd). # # Note: the want_rebuild_for_type 'flag' is actually a counter; it is decremented # after each scan. This ensures that we always recompile at least once more; # once to *define* the rule, and once afterwards to *undefine* the rule in the # compiled ruleset again. # # If two consecutive scans use user rules, that's ok -- the second one will # reset the counter, and we'll still recompile just once afterwards to undefine # the rule again. # if ($self->{scoresonly}) { $conf->{want_rebuild_for_type}->{$type} = 2; $conf->{user_defined_rules}->{$name} = 1; } } sub add_regression_test { my ($self, $name, $ok_or_fail, $string) = @_; my $conf = $self->{conf}; if ($conf->{regression_tests}->{$name}) { push @{$conf->{regression_tests}->{$name}}, [$ok_or_fail, $string]; } else { # initialize the array, and create one element $conf->{regression_tests}->{$name} = [ [$ok_or_fail, $string] ]; } } sub is_meta_valid { my ($self, $name, $rule) = @_; # $meta is a degenerate translation of the rule, replacing all variables (i.e. rule names) with 0. my $meta = ''; # Paranoid check (Bug #7557) if ($rule =~ /(?:\:\:|->|[\$\@\%\;\{\}])/) { warn("config: invalid meta $name rule: $rule\n"); return 0; } # Process expandable functions before lexing $rule =~ s/${META_RULES_MATCHING_RE}/ 0 /g; # Lex the rule into tokens using a rather simple RE method ... my @tokens = ($rule =~ /($ARITH_EXPRESSION_LEXER)/og); if (length($name) == 1) { for (@tokens) { print "$name $_\n " or die "Error writing token: $!"; } } # Go through each token in the meta rule foreach my $token (@tokens) { # If the token is a syntactically legal rule name, make it zero if ($token =~ IS_RULENAME) { $meta .= "0 "; } # if it is a (decimal) number or a string of 1 or 2 punctuation # characters (i.e. operators) tack it onto the degenerate rule elsif ($token =~ /^(\d+(?:\.\d+)?|[[:punct:]]{1,2})\z/s) { $meta .= "$token "; } # Skip anything unknown (Bug #7557) else { $self->lint_warn("config: invalid meta $name token: $token", $name); return 0; } } $meta = untaint_var($meta); # was carefully checked my $evalstr = 'my $x = '.$meta.'; 1;'; if (eval $evalstr) { return 1; } my $err = $@ ne '' ? $@ : "errno=$!"; chomp $err; $err =~ s/\s+(?:at|near)\b.*//s; $err =~ s/Illegal division by zero/division by zero possible/i; $self->lint_warn("config: invalid expression for rule $name: \"$rule\": $err\n", $name); return 0; } sub parse_captures { my ($self, $name, $re) = @_; # Check for named regex capture templates if (index($$re, '%{') >= 0) { local($1); # Replace %{FOO} with %\{FOO\} so compile_regexp doesn't fail with unescaped left brace while ($$re =~ s/(?<!\\)\%\{([A-Z][A-Z0-9]*(?:_[A-Z0-9]+)*(?:\([^\)\}]*\))?)\}/%\\{$1\\}/g) { dbg("config: found named capture for rule $name: $1"); $self->{conf}->{capture_template_rules}->{$name}->{$1} = 1; } } # Make rules with captures run before anything else if ($$re =~ /\(\?P?[<'][A-Z]/) { dbg("config: adjusting regex capture rule $name priority to -10000"); $self->{conf}->{priority}->{$name} = -10000; $self->{conf}->{capture_rules}->{$name} = 1; } } # Deprecated functions, leave just in case.. sub is_delimited_regexp_valid { my ($self, $rule, $re) = @_; warn "deprecated is_delimited_regexp_valid() called, use compile_regexp()\n"; my ($rec, $err) = compile_regexp($re, 1, 1); return $rec; } sub is_regexp_valid { my ($self, $rule, $re) = @_; warn "deprecated is_regexp_valid() called, use compile_regexp()\n"; my ($rec, $err) = compile_regexp($re, 1, 1); return $rec; } sub is_always_matching_regexp { warn "deprecated is_always_matching_regexp() called\n"; return; } ########################################################################### sub add_to_addrlist { my ($self, $singlelist, @addrs) = @_; my $conf = $self->{conf}; foreach my $addr (@addrs) { $addr = lc $addr; my $re = $addr; $re =~ s/[\000\\\(]/_/gs; # paranoia $re =~ s/([^\*\?_a-zA-Z0-9])/\\$1/g; # escape any possible metachars $re =~ tr/?/./; # "?" -> "." $re =~ s/\*+/\.\*/g; # "*" -> "any string" my ($rec, $err) = compile_regexp("^${re}\$", 0); if (!$rec) { warn "could not compile $singlelist '$addr': $err"; return; } $conf->{$singlelist}->{$addr} = $rec; } } sub add_to_addrlist_rcvd { my ($self, $listname, $addr, $domain) = @_; my $conf = $self->{conf}; $domain = lc $domain; $addr = lc $addr; if ($conf->{$listname}->{$addr}) { push @{$conf->{$listname}->{$addr}{domain}}, $domain; } else { my $re = $addr; $re =~ s/[\000\\\(]/_/gs; # paranoia $re =~ s/([^\*\?_a-zA-Z0-9])/\\$1/g; # escape any possible metachars $re =~ tr/?/./; # "?" -> "." $re =~ s/\*+/\.\*/g; # "*" -> "any string" my ($rec, $err) = compile_regexp("^${re}\$", 0); if (!$rec) { warn "could not compile $listname '$addr': $err"; return; } $conf->{$listname}->{$addr}{re} = $rec; $conf->{$listname}->{$addr}{domain} = [ $domain ]; } } sub remove_from_addrlist { my ($self, $singlelist, @addrs) = @_; my $conf = $self->{conf}; foreach my $addr (@addrs) { delete($conf->{$singlelist}->{lc $addr}); } } sub remove_from_addrlist_rcvd { my ($self, $listname, @addrs) = @_; my $conf = $self->{conf}; foreach my $addr (@addrs) { delete($conf->{$listname}->{lc $addr}); } } sub add_to_addrlist_dkim { add_to_addrlist_rcvd(@_); } sub remove_from_addrlist_dkim { my ($self, $listname, $addr, $domain) = @_; my $conf = $self->{conf}; my $conf_lname = $conf->{$listname}; $addr = lc $addr; if ($conf_lname->{$addr}) { $domain = lc $domain; my $domains_listref = $conf_lname->{$addr}{domain}; # removing $domain from the list my @replacement = grep { lc $_ ne $domain } @$domains_listref; if (!@replacement) { # nothing left, remove the entire addr entry delete($conf_lname->{$addr}); } elsif (@replacement != @$domains_listref) { # anything changed? $conf_lname->{$addr}{domain} = \@replacement; } } } ########################################################################### sub fix_path_relative_to_current_file { my ($self, $path) = @_; # the path may be specified as "~/foo", so deal with that $path = $self->{conf}->{main}->sed_path($path); if (!File::Spec->file_name_is_absolute ($path)) { my ($vol, $dirs, $file) = File::Spec->splitpath ($self->{currentfile}); $path = File::Spec->catpath ($vol, $dirs, $path); dbg("config: fixed relative path: $path"); } return $path; } ########################################################################### sub lint_warn { my ($self, $msg, $rule, $iserror) = @_; if (!defined $iserror) { $iserror = 1; } if ($self->{conf}->{main}->{lint_callback}) { $self->{conf}->{main}->{lint_callback}->( msg => $msg, rule => $rule, iserror => $iserror ); } elsif ($self->{conf}->{lint_rules}) { warn $msg."\n"; } else { info($msg); } if ($iserror) { $self->{conf}->{errors}++; } } ########################################################################### sub get_if_stack_as_string { my ($self) = @_; return join ' ', map { $_->{conditional} } @{$self->{if_stack}}; } ########################################################################### 1;