# # Courier::Filter # A purely Perl-based filter framework for the Courier MTA. # # (C) 2003-2008 Julian Mehnle # $Id: Filter.pm 210 2008-03-21 19:30:31Z julian $ # ############################################################################### =head1 NAME Courier::Filter - Purely Perl-based mail filter framework for the Courier MTA =cut package Courier::Filter; =head1 VERSION 0.200 =cut use version; our $VERSION = qv('0.200'); use warnings; use strict; #use threads; #BEGIN { require threads if ... } use Error ':try'; use IO::Handle; use IO::Socket::UNIX; use IO::Select; use Courier::Config; use Courier::Message; use Courier::Error; use constant TRUE => (0 == 0); use constant FALSE => not TRUE; =head1 SYNOPSIS use Courier::Filter; use Courier::Filter::Logger::Moo; use Courier::Filter::Module::Foo; use Courier::Filter::Module::Bar; my $filter = Courier::Filter->new( mandatory => 1, logger => Courier::Filter::Logger::Moo->new( ... ), modules => [ Courier::Filter::Module::Foo->new( ... ), Courier::Filter::Module::Bar->new( ... ) ], testing => 0, debugging => 0 ); my $exit_code = $filter->run() || 0; exit($exit_code); =head1 DESCRIPTION B framework, see L.> The B class is the heart of the Courier::Filter framework. To drive a B filter process, create a B object, passing the filter modules and loggers you want to use to the constructor, and call the C method. Courier::Filter will then take care of creating the courierfilter socket in the right place in a safe manner, listening for connections from Courier, asking filter modules for consideration of messages, notifying Courier of whether messages should be accepted or rejected, logging message rejections, catching and logging errors, and finally removing the socket when being terminated by Courier. =cut # Implementation: ############################################################################### =head2 Constructor The following constructor is provided: =over =item B: returns I; throws I, Perl exceptions Creates a new C object. Also creates the courierfilter socket in the right place in a safe manner. %options is a list of key/value pairs representing any of the following options: =over =item B The name of the filter process. Used to build the socket name. Defaults to the base name of the process (C<$0>). =item B A boolean value controlling whether the filter process should act as a mandatory courierfilter. If B, users will not be able to bypass the filter modules in this filter process from their individual B filters. Technically, this controls whether the courierfilter socket will be created in the C (B) or the C (B) directory in Courier's run-time state directory (see L). Defaults to B. =begin comment =item B The number of worker threads that should be employed for considering messages. If B or B<0>, no multi-threading will be used and everything (socket connection handling, message consideration, and logging) will be done in a single thread. The unthreaded mode is is the most memory-conserving mode and is generally the best choice for single-processor systems. Otherwise, i.e. if set to B, a pool of an according number of worker threads will be employed for considering messages. The worker threads are created during the startup of Courier::Filter and process messages that are dispatched to them by the main thread. Logging is also done in a single separate thread. The multi-threaded mode requires a certain amount of memory for each thread but is a viable way to distribute work on multi-processor systems. No more than two threads per processor are recommended. Defaults to B. =end comment =item B A B object that will be used for logging message rejections and error messages. You may override this for individual filter modules for which you do not want the global logger to be used. If no logger is specified, logging is disabled. =item B I. A so-called B structure. A module group is a reference to an array that may contain filter module objects (i.e. instances of sub-classes of B), as well as other module groups. Thus, a module group is essentially a tree structure with filter modules as its leaves. When considering messages, Courier::Filter walks the tree in a recursive-descent, depth-first order, asking every filter module for consideration of the message's acceptability. For instance, given the following filter module group: [$m1, $m2, [$m3, [$m4, $m5]], $m6] Courier::Filter queries the filter modules in ascending order from 1 to 6. The acceptability result returned by each module determines how Courier::Filter proceeds with considering the current message: =over =item * If a module states an B, Courier::Filter aborts the consideration process and rejects the message. =item * If a module states an B, Courier::Filter just proceeds to the next module in turn. =item * If a module states an B, Courier::Filter skips the rest of the current module group and proceeds to the next item in the superordinate module group, assuming the whole group to be an implicit accept. =back For instance, take the nested filter module group from above: [$m1, $m2, [$m3, [$m4, $m5]], $m6] | | '---g3---'| | | '----group 2----' | '------------group 1-------------' Let's assume Courier::Filter queries the filter module $m3. If $m3 states an B, the consideration process is aborted and the current message is rejected. If $m3 states an B, Courier::Filter proceeds to $m4. If $m3 states an B, the rest of group 2 (including all of group 3) is skipped and the acceptability result of group 2 is assumed an implicit accept, so Courier::Filter proceeds to $m6. If no B has occured when Courier::Filter reaches the end of the main module group, or a module in the main group states an B, the message is accepted. Using nested groups of filter modules with normal or inverse polarity, it should be possible to implement sufficiently complex filtering policies to satisfy very most needs. =item B A boolean value controlling whether the I filter process should I apply any filtering to trusted messages. For details on how the trusted status is determined, see the description of the C property in Courier::Message. In most MTA configurations, this option can be used to white-list so-called outbound messages. Defaults to B. =item B A boolean value controlling whether the I filter process should run in "testing" mode. In testing mode, planned message rejections will be logged as usual, but no messages will actually be rejected. Defaults to B. NOTE: You may also enable testing mode on individual filter module objects, see L. Enabling testing mode globally is not the same as individually enabling testing mode on all filter modules, though. When global testing mode is enabled, Courier::Filter only ignores the I result, but still follows the rules of the normal consideration process, e.g. aborting as soon as a filter module states an B, etc. When an individual filter module is in testing mode, its I result is ignored, and the consideration process is continued with the next filter module. So individually enabling testing mode on all filter modules allows you to thoroughly test the correctness and performance of all installed filter modules, or even to gather stochastically indepent statistics on the hit/miss rates of your filter modules. =item B A boolean value controlling whether extra debugging information should be logged by Courier::Filter. Defaults to B. You need to enable debugging mode for filter modules separately. =for comment TODO: Filter modules' debugging mode should really default to Courier::Filter's global debugging mode. =back =cut sub new { my ($class, %options) = @_; $0 =~ m{([^/]+)$}; my $name = $options{name} || $1; my $mandatory = defined($options{mandatory}) ? $options{mandatory} : TRUE; my $threads = $options{threads}; my $logger = $options{logger}; my $modules = [ @{$options{modules}} ] || []; my $trusting = $options{trusting}; my $testing = $options{testing}; my $debugging = $options{debugging}; my $socket_dir = Courier::Config->runtime_dir . '/' . ( $mandatory ? 'allfilters' : 'filters' ); my $socket_dir_unused = Courier::Config->runtime_dir . '/' . ( !$mandatory ? 'allfilters' : 'filters' ); my $socket_prename = "$socket_dir/.$name"; my $socket_name = "$socket_dir/$name"; my $socket_prename_unused = "$socket_dir_unused/.$name"; my $socket_name_unused = "$socket_dir_unused/$name"; if (-e $socket_name) { -S $socket_name or throw Courier::Error("$socket_name already exists but is not a socket"); # Try to connect to socket to see if it is alive or # if it is left over from a crashed Courier::Filter: my $test_socket = IO::Socket::UNIX->new( Peer => $socket_name ); not defined($test_socket) or throw Courier::Error("Live socket $socket_name found -- is Courier::Filter already running?"); # Socket exists but is dead. Remove it: unlink($socket_name); } unlink($socket_prename); unlink($socket_prename_unused); unlink($socket_name_unused); my $socket = IO::Socket::UNIX->new( Local => $socket_prename, Listen => SOMAXCONN ) or throw Courier::Error("Unable to create socket $socket_prename"); rename($socket_prename, $socket_name) or unlink($socket_prename), throw Courier::Error("Unable to rename socket $socket_prename to $socket_name"); chmod(0660, $socket_name) or throw Courier::Error("Unable to chmod socket $socket_name"); IO::Handle->new_from_fd(3, '>')->close(); my $filter = { name => $name, mandatory => $mandatory, threads => $threads, logger => $logger, modules => $modules, trusting => $trusting, testing => $testing, debugging => $debugging, socket => $socket, socket_name => $socket_name, terminate => FALSE }; return bless($filter, $class); } =back =begin comment =head2 Destructor The following destructor is provided: =over =item B Removes the courierfilter socket when the B object is destroyed. There is no need to call this explicitly. =end comment =cut sub destroy { my ($filter) = @_; return if not $filter->{terminate}; $filter->{'socket'}->close(); unlink($filter->{socket_name}); # Dissolve worker thread pool: # foreach my $thread (threads->list) { # $thread->join() # if $thread->tid and $thread != threads->self; # } return; } =begin comment =back =end comment =head2 Instance methods The following instance methods are provided: =over =item B: throws I, Perl exceptions Runs the Courier::Filter. Listens for connections from Courier on the courierfilter socket, asks the configured filter modules for consideration of messages, notifies Courier of whether messages should be accepted or rejected, and logs message rejections. When Courier requests termination of the courierfilter, removes the socket and returns. =cut sub run { my ($filter) = @_; my $class = ref($filter); my $socket = $filter->{'socket'}; my $select = IO::Select->new(\*STDIN, $socket); while (not $filter->{terminate}) { # Wait for incoming connection requests # or EOF from STDIN: ######################################## my @ready_handles = $select->can_read(); foreach my $handle (@ready_handles) { if ($handle == $socket) { # Incoming connection request. $filter->handle_connection($socket); # threads->new(\&handle_connection, $filter, $socket)->detach(); } elsif ($handle == \*STDIN and STDIN->eof()) { # STDIN got closed. $filter->{terminate} = TRUE; } else { # Received data from unknown handle or from STDIN. # This shouldn't happen. throw Courier::Error("Received data from unknown handle or from STDIN"); } } } return; } =begin comment =item B: returns I, I; throws Perl exceptions Handles a single incoming connection to the courierfilter socket. Reads the message file name and zero or more control file names from the connection. Asks filter modules for consideration of the message's acceptability, and notifies Courier of whether the message should be accepted or rejected. Also returns the SMTP status response I and I given to Courier. =end comment =cut sub handle_connection { my ($filter, $socket) = @_; my $class = ref($filter); my $connection = $socket->accept(); my $message_file_name; my @control_file_names; while (my $file_name = <$connection>) { chomp($file_name); last unless $file_name; # Normalize file name: $file_name = Courier::Config->runtime_dir . '/tmp/' . $file_name if $file_name !~ m(^/); if (not defined($message_file_name)) { $message_file_name = $file_name; } else { push(@control_file_names, $file_name); } } return if not defined($message_file_name); my $message = Courier::Message->new( file_name => $message_file_name, control_file_names => \@control_file_names, filter => $filter ); my ($result, $code); # BEGIN XXX #STDERR->print( # "DEBUG: authenticated_user = '" . # ($message->authenticated_user || '(undef)') . # "'\n" #); # END XXX ($result, $code) = $filter->consult_modules($filter->modules, $message) if $filter->testing or not ($filter->trusting and $message->trusted); ($result, $code) = (undef, undef) if $filter->testing or ($filter->trusting and $message->trusted); if ($result) { $code ||= 550; } else { $result = 'Ok'; $code ||= 200; } my @lines = split(/\n/, $result); my $last_line = pop(@lines); $connection->print("$code-$_\n") foreach @lines; $connection->print("$code $last_line\n"); $connection->close(); return ($result, $code); } =begin comment =item B: returns I, I Walks the given modules group structure in a recursive-descent, depth-first order, and asks every filter module for consideration of the given message's acceptability. Returns the group's acceptability result. =end comment =cut sub consult_modules { my ($filter, $modules, $message) = @_; ref($modules) eq 'ARRAY' or throw Courier::Error('Invalid modules group structure, array-ref expected'); foreach my $module (@$modules) { my ($result, @code); if (UNIVERSAL::isa($module, 'Courier::Filter::Module')) { # Single module, make it consider the message: next if $module->trusting and $message->trusted; # ...except when the module trusts the message. my $logger = $module->logger || $filter->logger; ($result, @code) = eval { $module->consider($message) }; if ($@) { $logger->log_error(ref($module) . ': ' . $@) if $logger; ($result, @code) = ('Mail filters temporarily unavailable.', 432); } # Log rejection: $logger->log_rejected_message($message, $result) if $result and $logger; # Ignore result if module is in testing mode: next if $module->testing; } else { # Something else, try to interpret it as a modules group: ($result, @code) = $filter->consult_modules($module, $message); } return $result ? ($result, @code) : undef if defined($result); } return undef; } =item B: returns I Returns the name of the filter process, as set through the constructor's C option. =cut sub name { my ($filter) = @_; # Read-only! return $filter->{name}; } =item B: returns I Returns a boolean value indicating whether the filter process is a mandatory courierfilter, as set through the constructor's C option. =cut sub mandatory { my ($filter) = @_; # Read-only! return $filter->{mandatory}; } =begin comment =item B: returns I Returns the number of filter threads, as set through the constructor's C option. =end comment =cut sub threads { my ($filter) = @_; # Read-only! return $filter->{threads}; } =item B: returns I =item B: returns I If C<$logger> is specified, installs a new global logger. Returns the (newly) configured global logger. =cut sub logger { my ($filter, @logger) = @_; $filter->{logger} = $logger[0] if @logger; return $filter->{logger}; } =item B: returns I =item B: returns I If C<\@modules> is specified, installs a new filter module group structure. Returns the (newly) configured filter modules group structure. =cut sub modules { my ($filter, @modules) = @_; $filter->{modules} = $modules[0] if @modules; return $filter->{modules}; } =item B: returns I Returns a boolean value indicating the trusting mode, as set through the constructor's C option. =cut sub trusting { my ($filter) = @_; # Read-only! return $filter->{trusting}; } =item B: returns I Returns a boolean value indicating the global testing mode, as set through the constructor's C option. =cut sub testing { my ($filter) = @_; # Read-only! return $filter->{testing}; } =item B: returns I =item B: returns I If C<$debugging> is specified, sets the global debugging mode. Returns a boolean value indicating the (newly) configured global debugging mode. =cut sub debugging { my ($filter, @debugging) = @_; $filter->{debugging} = $debugging[0] if @debugging; return $filter->{debugging}; } =back =cut BEGIN { no warnings 'once'; *DESTROY = \&destroy; } =head1 SEE ALSO L, L, L, L For AVAILABILITY, SUPPORT, and LICENSE information, see L. =head1 REFERENCES =over =item The B interface L =back =head1 AUTHOR Julian Mehnle =cut TRUE;