# -*- perl -*- # # Net::Server::PreForkSimple - Net::Server personality # # Copyright (C) 2001-2022 # # Paul Seamons # # This package may be distributed under the terms of either the # GNU General Public License # or the # Perl Artistic License # # All rights reserved. # ################################################################ package Net::Server::PreForkSimple; use strict; use base qw(Net::Server); use File::Temp qw(tempfile); use Net::Server::SIG qw(register_sig check_sigs); use POSIX qw(WNOHANG EINTR); use Fcntl (); sub net_server_type { __PACKAGE__ } sub options { my $self = shift; my $ref = $self->SUPER::options(@_); my $prop = $self->{'server'}; $ref->{$_} = \$prop->{$_} for qw(max_servers max_requests max_dequeue check_for_dead check_for_dequeue lock_file serialize); $ref->{'sig_passthrough'} = $prop->{'sig_passthrough'} = []; return $ref; } sub post_configure { my $self = shift; my $prop = $self->{'server'}; $self->SUPER::post_configure; ### some default values to check for my $d = { max_servers => 50, # max num of servers to run max_requests => 1000, # num of requests for each child to handle check_for_dead => 30, # how often to see if children are alive }; foreach (keys %$d){ $prop->{$_} = $d->{$_} unless defined($prop->{$_}) && $prop->{$_} =~ /^\d+$/; } $prop->{'ppid'} = $$; } sub post_bind { my $self = shift; my $prop = $self->{'server'}; $self->SUPER::post_bind; if ($prop->{'multi_port'} && $prop->{'serialize'} && $prop->{'serialize'} eq 'none') { $self->log(2, "Passed serialize value of none is incompatible with multiple ports - using default serialize"); delete $prop->{'serialize'}; } if (!$prop->{'serialize'} || $prop->{'serialize'} !~ /^(flock|semaphore|pipe|none)$/i) { $prop->{'serialize'} = ($^O eq 'MSWin32') ? 'pipe' : 'flock'; } $prop->{'serialize'} =~ tr/A-Z/a-z/; if ($prop->{'serialize'} eq 'flock') { $self->log(3, "Setting up serialization via flock"); if (defined $prop->{'lock_file'}) { $prop->{'lock_file_unlink'} = undef; } else { (my $fh, $prop->{'lock_file'}) = tempfile(); # We don't need to keep the file handle open in the parent; # each child opens it separately to avoid sharing the lock close $fh or die "Cannot close lock file $prop->{'lock_file'}: $!"; $prop->{'lock_file_unlink'} = 1; } } elsif ($prop->{'serialize'} eq 'semaphore') { $self->log(3, "Setting up serialization via semaphore"); require IPC::SysV; require IPC::Semaphore; my $s = IPC::Semaphore->new(IPC::SysV::IPC_PRIVATE(), 1, IPC::SysV::S_IRWXU() | IPC::SysV::IPC_CREAT()) or $self->fatal("Semaphore error [$!]"); $s->setall(1) or $self->fatal("Semaphore create error [$!]"); $prop->{'sem'} = $s; } elsif ($prop->{'serialize'} eq 'pipe') { $self->log(3, "Setting up serialization via pipe"); pipe(my $waiting, my $ready); $ready->autoflush(1); $waiting->autoflush(1); $prop->{'_READY'} = $ready; $prop->{'_WAITING'} = $waiting; print $ready "First\n"; } elsif ($prop->{'serialize'} eq 'none') { $self->log(3, "Using no serialization"); } else { $self->fatal("Unknown serialization type \"$prop->{'serialize'}\""); } } sub loop { my $self = shift; my $prop = $self->{'server'}; $prop->{'children'} = {}; if ($ENV{'HUP_CHILDREN'}) { my %children = map {/^(\w+)$/; $1} split(/\s+/, $ENV{'HUP_CHILDREN'}); $children{$_} = {status => $children{$_}, hup => 1} foreach keys %children; $prop->{'children'} = \%children; } $self->log(3, "Beginning prefork ($prop->{'max_servers'} processes)"); $self->run_n_children($prop->{'max_servers'}); $self->run_parent; } sub run_n_children { my ($self, $n) = @_; return if $n <= 0; my $prop = $self->{'server'}; $self->run_n_children_hook; $self->log(3, "Starting \"$n\" children"); for (1 .. $n) { $self->pre_fork_hook; local $!; my $pid = fork; $self->fatal("Bad fork [$!]") if ! defined $pid; if ($pid) { $prop->{'children'}->{$pid}->{'status'} = 'processing'; $self->register_child($pid, 'preforksimple'); } else { $self->run_child; } } } sub run_n_children_hook {} sub run_child { my $self = shift; my $prop = $self->{'server'}; $SIG{'INT'} = $SIG{'TERM'} = $SIG{'QUIT'} = sub { $self->child_finish_hook; exit; }; $SIG{'PIPE'} = 'IGNORE'; $SIG{'CHLD'} = 'DEFAULT'; $SIG{'HUP'} = sub { if (! $prop->{'connected'}) { $self->child_finish_hook; exit; } $prop->{'SigHUPed'} = 1; }; my $needs_lock = ($prop->{'serialize'} eq 'flock') ? 1 : 0; if ($needs_lock) { open($prop->{'lock_fh'}, ">", $prop->{'lock_file'}) or $self->fatal("Couldn't open lock file \"$prop->{'lock_file'}\"[$!]"); } $self->log(4, "Child Preforked ($$)"); delete $prop->{'children'}; $self->child_init_hook; while ($self->accept()) { $prop->{'connected'} = 1; $self->run_client_connection; $prop->{'connected'} = 0; last if $self->done; } $self->child_finish_hook; close($prop->{'lock_fh'}) if $needs_lock && $prop->{'lock_fh'}; $self->log(4, "Child leaving ($prop->{'max_requests'})"); exit; } sub is_prefork { 1 } ### We can only let one process do the selecting at a time ### this override makes sure that nobody else can do it ### while we are. We do this either by opening a lock file ### and getting an exclusive lock (this will block all others ### until we release it) or by using semaphores to block sub accept { my $self = shift; my $prop = $self->{'server'}; if ($prop->{'serialize'} eq 'flock') { while (! flock $prop->{'lock_fh'}, Fcntl::LOCK_EX()) { next if $! == EINTR; $self->fatal("Couldn't get lock on file \"$prop->{'lock_file'}\" [$!]"); } my $v = $self->SUPER::accept(); flock $prop->{'lock_fh'}, Fcntl::LOCK_UN(); return $v; } elsif ($prop->{'serialize'} eq 'semaphore') { $prop->{'sem'}->op(0, -1, IPC::SysV::SEM_UNDO()) or $self->fatal("Semaphore Error [$!]"); my $v = $self->SUPER::accept(); $prop->{'sem'}->op(0, 1, IPC::SysV::SEM_UNDO()) or $self->fatal("Semaphore Error [$!]"); return $v; } elsif ($prop->{'serialize'} eq 'pipe') { my $waiting = $prop->{'_WAITING'}; scalar <$waiting>; # read one line - kernel says who gets it my $v = $self->SUPER::accept(); print { $prop->{'_READY'} } "Next!\n"; return $v; } else { my $v = $self->SUPER::accept(); return $v; } } sub done { my $self = shift; my $prop = $self->{'server'}; $prop->{'done'} = shift if @_; return 1 if $prop->{'done'}; return 1 if $prop->{'requests'} >= $prop->{'max_requests'}; return 1 if $prop->{'SigHUPed'}; if (! kill 0, $prop->{'ppid'}) { $self->log(3, "Parent process gone away. Shutting down"); return 1; } } sub run_parent { my $self=shift; my $prop = $self->{'server'}; $self->log(4, "Parent ready for children."); $prop->{'last_checked_for_dead'} = $prop->{'last_checked_for_dequeue'} = time(); register_sig( PIPE => 'IGNORE', INT => sub { $self->server_close() }, TERM => sub { $self->server_close() }, HUP => sub { $self->sig_hup() }, CHLD => sub { while (defined(my $chld = waitpid(-1, WNOHANG))) { last unless $chld > 0; $self->delete_child($chld); } }, QUIT => sub { $self->{'server'}->{'kind_quit'} = 1; $self->server_close() }, TTIN => sub { $self->{'server'}->{'max_servers'}++; $self->log(3, "Increasing max server count ($self->{'server'}->{'max_servers'})") }, TTOU => sub { $self->{'server'}->{'max_servers'}--; $self->log(3, "Decreasing max server count ($self->{'server'}->{'max_servers'})"); if (defined(my $pid = each %{ $prop->{'children'} })) { $self->delete_child($pid) if ! kill('HUP', $pid); } }, ); $self->register_sig_pass; if ($ENV{'HUP_CHILDREN'}) { while (defined(my $chld = waitpid(-1, WNOHANG))) { last unless $chld > 0; $self->delete_child($chld); } } while (1) { select undef, undef, undef, 10; if (check_sigs()){ last if $prop->{'_HUP'}; } $self->idle_loop_hook(); # periodically make sure children are alive my $time = time(); if ($time - $prop->{'last_checked_for_dead'} > $prop->{'check_for_dead'}) { $prop->{'last_checked_for_dead'} = $time; foreach (keys %{ $prop->{'children'} }) { kill(0,$_) or $self->delete_child($_); } } # make sure we always have max_servers my $total_n = 0; my $total_d = 0; foreach (values %{ $prop->{'children'} }){ if( $_->{'status'} eq 'dequeue' ){ $total_d ++; }else{ $total_n ++; } } if( $prop->{'max_servers'} > $total_n ){ $self->run_n_children( $prop->{'max_servers'} - $total_n ); } # periodically check to see if we should clear the queue if( defined $prop->{'check_for_dequeue'} ){ if( $time - $prop->{'last_checked_for_dequeue'} > $prop->{'check_for_dequeue'} ){ $prop->{'last_checked_for_dequeue'} = $time; if( defined($prop->{'max_dequeue'}) && $total_d < $prop->{'max_dequeue'} ){ $self->run_dequeue(); } } } } } sub idle_loop_hook {} sub close_children { my $self = shift; $self->SUPER::close_children(@_); check_sigs(); # since we have captured signals - make sure we handle them register_sig(PIPE => 'DEFAULT', INT => 'DEFAULT', TERM => 'DEFAULT', QUIT => 'DEFAULT', HUP => 'DEFAULT', CHLD => 'DEFAULT', TTIN => 'DEFAULT', TTOU => 'DEFAULT', ); } 1; __END__ =head1 NAME Net::Server::PreForkSimple - Net::Server personality =head1 SYNOPSIS use base qw(Net::Server::PreForkSimple); sub process_request { #...code... } __PACKAGE__->run(); =head1 DESCRIPTION Please read the pod on Net::Server first. This module is a personality, or extension, or sub class, of the Net::Server module. This personality binds to one or more ports and then forks C child processes. The server will make sure that at any given time there are always C available to receive a client request. Each of these children will process up to C client connections. This type is good for a heavily hit site that can keep C processes dedicated to the serving. (Multi port accept defaults to using flock to serialize the children). At this time, it does not appear that this module will pass tests on Win32 systems. Any ideas or patches for making the tests pass would be welcome. =head1 SAMPLE CODE Please see the sample listed in Net::Server. =head1 COMMAND LINE ARGUMENTS In addition to the command line arguments of the Net::Server base class, Net::Server::PreFork contains several other configurable parameters. Key Value Default max_servers \d+ 50 max_requests \d+ 1000 serialize (flock|semaphore |pipe|none) undef # serialize defaults to flock on multi_port or on Solaris lock_file "filename" File::Temp->new check_for_dead \d+ 30 max_dequeue \d+ undef check_for_dequeue \d+ undef =over 4 =item max_servers The maximum number of child servers to start and maintain. This does not apply to dequeue processes. =item max_requests The number of client connections to receive before a child terminates. =item serialize Determines whether the server serializes child connections. Options are undef, flock, semaphore, pipe, or none. Default is undef. On multi_port servers or on servers running on Solaris, the default is flock. The flock option uses blocking exclusive flock on the file specified in I (see below). The semaphore option uses IPC::Semaphore (thanks to Bennett Todd) for giving some sample code. The pipe option reads on a pipe to choose the next. the flock option should be the most bulletproof while the pipe option should be the most portable. (Flock is able to relinquish the block if the process dies between accept on the socket and reading of the client connection - semaphore and pipe do not). An option of none will not perform any serialization. If "none" is passed and there are multiple ports then a the default serialization will be used instead of "none." =item lock_file Filename to use in flock serialized accept in order to serialize the accept sequence between the children. This will default to a generated temporary filename. If default value is used the lock_file will be removed when the server closes. =item check_for_dead Seconds to wait before checking to see if a child died without letting the parent know. =item max_dequeue The maximum number of dequeue processes to start. If a value of zero or undef is given, no dequeue processes will be started. The number of running dequeue processes will be checked by the check_for_dead variable. =item check_for_dequeue Seconds to wait before forking off a dequeue process. The run_dequeue hook must be defined when using this setting. It is intended to use the dequeue process to take care of items such as mail queues. If a value of undef is given, no dequeue processes will be started. =back =head1 CONFIGURATION FILE C allows for the use of a configuration file to read in server parameters. The format of this conf file is simple key value pairs. Comments and white space are ignored. #-------------- file test.conf -------------- ### server information max_servers 80 max_requests 1000 ### user and group to become user somebody group everybody ### logging ? log_file /var/log/server.log log_level 3 pid_file /tmp/server.pid ### access control allow .+\.(net|com) allow domain\.com deny a.+ ### background the process? background 1 ### ports to bind host 127.0.0.1 port localhost:20204 port 20205 ### reverse lookups ? # reverse_lookups on #-------------- file test.conf -------------- =head1 PROCESS FLOW Process flow follows Net::Server until the loop phase. At this point C are forked and wait for connections. When a child accepts a connection, finishes processing a client, or exits, it relays that information to the parent, which keeps track and makes sure there are always C running. =head1 HOOKS The PreForkSimple server has the following hooks in addition to the hooks provided by the Net::Server base class. See L =over 4 =item C<$self-Erun_n_children_hook()> This hook occurs at the top of run_n_children which is called each time the server goes to start more child processes. This gives the parent to do a little of its own accounting (as desired). Idea for this hook came from James FitzGibbon. =item C<$self-Echild_init_hook()> This hook takes place immediately after the child process forks from the parent and before the child begins accepting connections. It is intended for any additional chrooting or other security measures. It is suggested that all perl modules be used by this point, so that the most shared memory possible is used. =item C<$self-Echild_finish_hook()> This hook takes place immediately before the child tells the parent that it is exiting. It is intended for saving out logged information or other general cleanup. =item C<$self-Erun_dequeue()> This hook only gets called in conjunction with the check_for_dequeue setting. =item C<$self-Eidle_loop_hook()> This hook is called in every pass through the main process wait loop. =back =head1 HOT DEPLOY Since version 2.000, the PreForkSimple server has accepted the TTIN and TTOU signals. When a TTIN is received, the max_servers is increased by 1. If a TTOU signal is received the max_servers is decreased by 1. This allows for adjusting the number of handling processes without having to restart the server. =head1 BUGS Tests don't seem to work on Win32. Any ideas or patches would be welcome. =head1 TO DO See L =head1 AUTHOR Paul T. Seamons paul@seamons.com =head1 THANKS See L =head1 SEE ALSO Please see also L, L, L, L, L L L L =cut