Coro/myhttpd/httpd.pl

use Coro;
use Coro::Semaphore;
use Coro::SemaphoreSet;
use Coro::EV;
use Coro::Socket;
use Coro::Signal;
use Coro::AIO ();

use Fcntl;
use HTTP::Date;
use POSIX ();

use Compress::Zlib ();

use common::sense;

# at least on my machine, this thingy serves files
# quite a bit faster than apache, ;)
# and quite a bit slower than thttpd :(

$SIG{PIPE} = 'IGNORE';

our $accesslog;
our $errorlog;
our @listen_sockets;

our $NOW;
our $HTTP_NOW;

our $ERROR_LOG;
our $ACCESS_LOG;
our $TRANSFER_LOCK = new Coro::SemaphoreSet; # lock to be acquired per ip

our $update_time = EV::periodic 0, 1, undef, sub {
   $NOW = time;
   $HTTP_NOW = time2str $NOW;
};
$update_time->invoke;

if ($ERROR_LOG) {
   use IO::Handle;
   open $errorlog, ">>$ERROR_LOG"
     or die "$ERROR_LOG: $!";
   $errorlog->autoflush(1);
}

if ($ACCESS_LOG) {
   use IO::Handle;
   open $accesslog, ">>$ACCESS_LOG"
     or die "$ACCESS_LOG: $!";
   $accesslog->autoflush(1);
}

sub slog {
   my $level = shift;
   my $format = shift;
   my $NOW = (POSIX::strftime "%Y-%m-%d %H:%M:%S", gmtime $::NOW);
   printf "$NOW: $format\n", @_;
   printf $errorlog "$NOW: $format\n", @_ if $errorlog;
}

our $connections = new Coro::Semaphore $::MAX_CONNECTS || 250;
our $httpevent   = new Coro::Signal;

our $queue_file  = new transferqueue $::MAX_TRANSFERS;
our $queue_index = new transferqueue 10;

our $tbf_top     = new tbf rate => $::TBF_RATE || 100000;

my $unused_bytes = 0;
my $unused_last  = time;

sub unused_bandwidth {
   $unused_bytes += $_[0];
   if ($unused_last < $NOW - 30 && $unused_bytes / ($NOW - $unused_last) > 50000) {
      $unused_last = $NOW;
      $unused_bytes = 0;
      $queue_file->force_wake_next;
      slog 1, "forced filetransfer due to unused bandwidth";
   }
}

sub listen_on {
   my $listen = $_[0];

   push @listen_sockets, $listen;

   # the "main thread"
   async {
      slog 1, "accepting connections";
      while () {
         $connections->down;
         my @conn = $listen->accept;
         #slog 3, "accepted @$connections ".scalar(@pool);

         async_pool {
            eval {
               conn->new (@conn)->handle;
            };
            slog 1, "$@" if $@ && !ref $@;

            $httpevent->broadcast; # only for testing, but doesn't matter much

            $connections->up;
         }
      }
   };
}

my $http_port = new Coro::Socket
        LocalAddr => $::SERVER_HOST,
        LocalPort => $::SERVER_PORT,
        ReuseAddr => 1,
        Listen    => 50,
   or die "unable to start server";

listen_on $http_port;

if ($::SERVER_PORT2) {
   my $http_port = new Coro::Socket
           LocalAddr => $::SERVER_HOST,
           LocalPort => $::SERVER_PORT2,
           ReuseAddr => 1,
           Listen    => 50,
      or die "unable to start server";

   listen_on $http_port;
}

package conn;

use common::sense;

use Socket;
use HTTP::Date;
use Convert::Scalar 'weaken';
use IO::AIO;

IO::AIO::min_parallel $::AIO_PARALLEL;

our $AIO_WATCHER = EV::io IO::AIO::poll_fileno, EV::READ, \&IO::AIO::poll_cb;

our %conn; # $conn{ip}{self} => connobj
our %uri;  # $uri{ip}{uri}{self}
our %blocked;
our %mimetype;

sub read_mimetypes {
   if (open my $fh, "<mime_types") {
      while (<$fh>) {
         if (/^([^#]\S+)\t+(\S+)$/) {
            $mimetype{lc $1} = $2;
         }
      }
   } else {
      print "cannot open mime_types\n";
   }
}

read_mimetypes;

sub new {
   my $class = shift;
   my $fh = shift;
   my $peername = shift;
   my $self = bless { fh => $fh }, $class;
   my (undef, $iaddr) = unpack_sockaddr_in $peername
      or $self->err (500, "unable to decode peername");

   $self->{remote_addr} =
      $self->{remote_id} = inet_ntoa $iaddr;

   $self->{time} = $::NOW;

   weaken ($Coro::current->{conn} = $self);

   ++$::conns;
   $::maxconns = $::conns if $::conns > $::maxconns;

   $self
}

sub DESTROY {
   my $self = shift;

   --$::conns;
}

sub prune_cache {
   my $hash = $_[0];

   for (keys %$hash) {
      if (ref $hash->{$_} eq HASH::) {
         prune_cache($hash->{$_});
         unless (scalar keys %{$hash->{$_}}) {
            delete $hash->{$_};
         }
      }
   }
}

sub prune_caches {
   prune_cache \%conn;
   prune_cache \%uri;

   for (keys %blocked) {
      delete $blocked{$_} unless $blocked{$_}[0] > $::NOW;
   }
}

our $PRUNE_WATCHER = EV::timer 60, 60, \&prune_caches;

sub slog {
   my $self = shift;
   main::slog($_[0], "$self->{remote_id}> $_[1]");
}

sub response {
   my ($self, $code, $msg, $hdr, $content) = @_;
   my $res = "HTTP/1.1 $code $msg\015\012";
   my $GZ = "";

   if (exists $hdr->{Connection}) {
      if ($hdr->{Connection} =~ /close/) {
         $self->{h}{connection} = "close"
      }
   } else {
      if ($self->{version} < 1.1) {
         if ($self->{h}{connection} =~ /keep-alive/i) {
            $hdr->{Connection} = "Keep-Alive";
         } else {
            $self->{h}{connection} = "close"
         }
      }
   }

   if ($self->{method} ne "HEAD"
       && $self->{h}{"accept-encoding"} =~ /\bgzip\b/
       && 400 < length $content
       && $hdr->{"Content-Length"} == length $content
       && !exists $hdr->{"Content-Encoding"}
   ) {
      my $orig = length $content;
      $hdr->{"Content-Encoding"} = "gzip";
      $content = Compress::Zlib::memGzip(\$content);
      $hdr->{"Content-Length"} = length $content;
      $GZ = sprintf "GZ%02d", 100 - 100*((length $content) / $orig);
   }

   $res .= "Date: $HTTP_NOW\015\012";
   $res .= "Server: $::NAME\015\012";

   while (my ($h, $v) = each %$hdr) {
      $res .= "$h: $v\015\012"
   }
   $res .= "\015\012";

   $res .= $content if defined $content and $self->{method} ne "HEAD";

   my $log = (POSIX::strftime "%Y-%m-%d %H:%M:%S", gmtime $::NOW).
             " $self->{remote_id} \"$self->{uri}\" $code ".$hdr->{"Content-Length"}.$GZ.
             " \"$self->{h}{referer}\"\n";

   print $::accesslog $log if $::accesslog;
   print STDERR $log;

   $tbf_top->request(length $res, 1e6);
   $self->{written} += print {$self->{fh}} $res;
}

sub err {
   my $self = shift;
   my ($code, $msg, $hdr, $content) = @_;

   unless (defined $content) {
      $content = "$code $msg\n";
      $hdr->{"Content-Type"} = "text/plain";
      $hdr->{"Content-Length"} = length $content;
   }
   $hdr->{"Connection"} = "close";

   $self->response ($code, $msg, $hdr, $content);

   die bless {}, err::
}

sub handle {
   my $self = shift;
   my $fh = $self->{fh};

   my $host;

   $fh->timeout($::REQ_TIMEOUT);
   while () {
      $self->{reqs}++;

      # read request and parse first line
      my $req = $fh->readline("\015\012\015\012");

      unless (defined $req) {
         if (exists $self->{version}) {
            last;
         } else {
            $self->err(408, "request timeout");
         }
      }

      $self->{h} = {};

      $fh->timeout($::RES_TIMEOUT);

      $req =~ /^(?:\015\012)?
                (GET|HEAD) \040+
                ([^\040]+) \040+
                HTTP\/([0-9]+\.[0-9]+)
                \015\012/gx
         or $self->err(405, "method not allowed", { Allow => "GET,HEAD" });

      $self->{method} = $1;
      $self->{uri} = $2;
      $self->{version} = $3;

      $3 =~ /^1\./
         or $self->err(506, "http protocol version $3 not supported");

      # parse headers
      {
         my (%hdr, $h, $v);

         $hdr{lc $1} .= ",$2"
            while $req =~ /\G
                  ([^:\000-\040]+):
                  [\011\040]*
                  ((?: [^\015\012]+ | \015\012[\011\040] )*)
                  \015\012
               /gxc;

         $req =~ /\G\015\012$/
            or $self->err(400, "bad request");

         $self->{h}{$h} = substr $v, 1
            while ($h, $v) = each %hdr;
      }

      # remote id should be unique per user
      my $id = $self->{remote_addr};

      if (exists $self->{h}{"client-ip"}) {
         $id .= "[".$self->{h}{"client-ip"}."]";
      } elsif (exists $self->{h}{"x-forwarded-for"}) {
         $id .= "[".$self->{h}{"x-forwarded-for"}."]";
      }

      $self->{remote_id} = $id;

      weaken (local $conn{$id}{$self*1} = $self);

      if ($blocked{$id}) {
         $self->err_blocked
            if $blocked{$id}[0] > $::NOW;

         delete $blocked{$id};
      }

      # find out server name and port
      if ($self->{uri} =~ s/^http:\/\/([^\/?#]*)//i) {
         $host = $1;
      } else {
         $host = $self->{h}{host};
      }

      if (defined $host) {
         $self->{server_port} = $host =~ s/:([0-9]+)$// ? $1 : 80;
      } else {
         ($self->{server_port}, $host)
            = unpack_sockaddr_in $self->{fh}->sockname
               or $self->err(500, "unable to get socket name");
         $host = inet_ntoa $host;
      }

      $self->{server_name} = $host;

      weaken (local $uri{$id}{$self->{uri}}{$self*1} = $self);

      eval {
         $self->map_uri;
         $self->respond;
      };

      die if $@ && !ref $@;

      last if $self->{h}{connection} =~ /close/i;

      $httpevent->broadcast;

      $fh->timeout($::PER_TIMEOUT);
   }
}

sub block {
   my $self = shift;

   $blocked{$self->{remote_id}} = [$::NOW + $_[0], $_[1]];
   $self->slog(2, "blocked ip $self->{remote_id}");
   $self->err_blocked;
}

# uri => path mapping
sub map_uri {
   my $self = shift;
   my $host = $self->{server_name};
   my $uri = $self->{uri};

   $host =~ /[\/\\]/
      and $self->err(400, "bad request");

   # some massaging, also makes it more secure
   $uri =~ s/%([0-9a-fA-F][0-9a-fA-F])/chr hex $1/ge;
   $uri =~ s%//+%/%g;
   $uri =~ s%/\.(?=/|$)%%g;
   1 while $uri =~ s%/[^/]+/\.\.(?=/|$)%%;

   $uri =~ m%^/?\.\.(?=/|$)%
      and $self->err(400, "bad request");

   $self->{name} = $uri;

   # now do the path mapping
   $self->{path} = "$::DOCROOT/$host$uri";

   $self->access_check;
}

sub _cgi {
   my $self = shift;
   my $path = shift;
   my $fh;

   # no two-way xxx supported
   if (0 == fork) {
      open STDOUT, ">&".fileno($self->{fh});
      if (chdir $::DOCROOT) {
         $ENV{SERVER_SOFTWARE} = "thttpd-myhttpd"; # we are thttpd-alike
         $ENV{HTTP_HOST}       = $self->{server_name};
         $ENV{HTTP_PORT}       = $self->{server_port};
         $ENV{SCRIPT_NAME}     = $self->{name};
         exec $path;
      }
      Coro::State::_exit(0);
   } else {
      die;
   }
}

sub server_hostport {
   $_[0]{server_port} == 80
      ? $_[0]{server_name}
      : "$_[0]{server_name}:$_[0]{server_port}";
}

sub respond {
   my $self = shift;
   my $path = $self->{path};

   if ($self->{name} =~ s%^/internal/([^/]+)%%) {
      if ($::internal{$1}) {
         $::internal{$1}->($self);
      } else {
         $self->err (404, "not found");
      }
   } else {

      Coro::AIO::aio_stat $path
         and $self->err (404, "not found");

      $self->{stat} = [stat _];

      # idiotic netscape sends idiotic headers AGAIN
      my $ims = $self->{h}{"if-modified-since"} =~ /^([^;]+)/
                ? str2time $1 : 0;

      if (-d _ && -r _) {
         # directory
         if ($path !~ /\/$/) {
            # create a redirect to get the trailing "/"
            # we don't try to avoid the :80
            $self->err (301, "moved permanently", { Location =>  "http://".$self->server_hostport."$self->{uri}/" });
         } else {
            $ims < $self->{stat}[9]
               or $self->err (304, "not modified");

            if (-r "$path/index.html") {
               # replace directory "size" by index.html filesize
               $self->{stat} = [stat ($self->{path} .= "/index.html")];
               $self->handle_file ($queue_index, $tbf_top);
            } else {
               $self->handle_dir;
            }
         }
      } elsif (-f _ && -r _) {
         -x _ and $self->err (403, "forbidden");

         if (keys %{$conn{$self->{remote_id}}} > $::MAX_TRANSFERS_IP) {
            my $timeout = $::NOW + 10;
            while (keys %{$conn{$self->{remote_id}}} > $::MAX_TRANSFERS_IP) {
               if ($timeout < $::NOW) {
                  $self->block($::BLOCKTIME, "too many connections");
               } else {
                  $httpevent->wait;
               }
            }
         }

         $self->handle_file ($queue_file, $tbf_top);
      } else {
         $self->err (404, "not found");
      }
   }
}

sub handle_dir {
   my $self = shift;
   my $idx = $self->diridx;

   $self->response (200, "ok",
         {
            "Content-Type"   => "text/html; charset=utf-8",
            "Content-Length" => length $idx,
            "Last-Modified"  => time2str ($self->{stat}[9]),
         },
         $idx);
}

sub handle_file {
   my ($self, $queue, $tbf) = @_;
   my $length = $self->{stat}[7];
   my $hdr = {
      "Last-Modified" => time2str ((stat _)[9]),
      "Accept-Ranges" => "bytes",
   };

   my @code = (200, "ok");
   my ($l, $h);

   if ($self->{h}{range} =~ /^bytes=(.*)$/i) {
      for (split /,/, $1) {
         if (/^-(\d+)$/) {
            ($l, $h) = ($length - $1, $length - 1);
         } elsif (/^(\d+)-(\d*)$/) {
            ($l, $h) = ($1, ($2 ne "" || $2 >= $length) ? $2 : $length - 1);
         } else {
            ($l, $h) = (0, $length - 1);
            goto ignore;
         }
         goto satisfiable if $l >= 0 && $l < $length && $h >= 0 && $h >= $l;
      }
      $hdr->{"Content-Range"} = "bytes */$length";
      $hdr->{"Content-Length"} = $length;
      $self->err (416, "not satisfiable", $hdr, "");

satisfiable:
      # check for segmented downloads
      if ($l && $::NO_SEGMENTED) {
         my $timeout = $::NOW + 15;
         while (keys %{$uri{$self->{remote_id}}{$self->{uri}}} > 1) {
            if ($timeout <= $::NOW) {
               $self->block ($::BLOCKTIME, "segmented downloads are forbidden");
               #$self->err_segmented_download;
            } else {
               $httpevent->wait;
            }
         }
      }

      $hdr->{"Content-Range"} = "bytes $l-$h/$length";
      @code = (206, "partial content");
      $length = $h - $l + 1;

ignore:
   } else {
      ($l, $h) = (0, $length - 1);
   }

   $self->{path} =~ /\.([^.]+)$/;
   $hdr->{"Content-Type"} = $mimetype{lc $1} || "application/octet-stream";
   $hdr->{"Content-Length"} = $length;

   $self->response (@code, $hdr, "");

   if ($self->{method} eq "GET") {
      $self->{time} = $::NOW;
      $self->{written} = 0;

      my $fh = Coro::AIO::aio_open $self->{path}, Fcntl::O_RDONLY, 0
         or die "$self->{path}: late open failure ($!)";

      $h -= $l - 1;

      my $transfer = $queue->start_transfer ($h);
      my $locked;
      my $bufsize = $::WAIT_BUFSIZE; # initial buffer size

      while ($h > 0) {
         Coro::cede;
         my $transfer_lock = $TRANSFER_LOCK->guard ($self->{remote_id});

         unless ($locked) {
            if ($locked ||= $transfer->try ($::WAIT_INTERVAL)) {
               $bufsize = $::BUFSIZE;
               $self->{time} = $::NOW;
               $self->{written} = 0;
            }
         }

         if ($blocked{$self->{remote_id}}) {
            $self->{h}{connection} = "close";
            die bless {}, err::;
         }

         Coro::AIO::aio_read $fh, $l, ($h > $bufsize ? $bufsize : $h), my $buf, 0
            or last;

         $tbf->request (length $buf);
         my $w = $self->{fh}->syswrite ($buf)
            or last;
         $::written += $w;
         $self->{written} += $w;
         $l += $w;
      }

      close $fh;
   }
}

1

Revision:	1.90
Committed:	Wed Jan 27 20:06:57 2010 UTC (14 years, 5 months ago) by root
Content type:	text/plain
Branch:	MAIN
Changes since 1.89:	+18 -13 lines
Log Message:	increase fairness between downloads from different ips
#	User	Rev	Content
1	root	1.1	use Coro;
2			use Coro::Semaphore;
3	root	1.90	use Coro::SemaphoreSet;
4	root	1.84	use Coro::EV;
5	root	1.1	use Coro::Socket;
6	root	1.44	use Coro::Signal;
7	root	1.79	use Coro::AIO ();
8	root	1.1
9	root	1.90	use Fcntl;
10	root	1.32	use HTTP::Date;
11	root	1.47	use POSIX ();
12	root	1.32
13	root	1.63	use Compress::Zlib ();
14
15	root	1.90	use common::sense;
16	root	1.1
17			# at least on my machine, this thingy serves files
18			# quite a bit faster than apache, ;)
19			# and quite a bit slower than thttpd :(
20
21			$SIG{PIPE} = 'IGNORE';
22	root	1.27
23			our $accesslog;
24	root	1.56	our $errorlog;
25	root	1.90	our @listen_sockets;
26	root	1.56
27			our $NOW;
28			our $HTTP_NOW;
29
30	root	1.80	our $ERROR_LOG;
31			our $ACCESS_LOG;
32	root	1.90	our $TRANSFER_LOCK = new Coro::SemaphoreSet; # lock to be acquired per ip
33	root	1.80
34	root	1.84	our $update_time = EV::periodic 0, 1, undef, sub {
35	root	1.56	$NOW = time;
36			$HTTP_NOW = time2str $NOW;
37	root	1.84	};
38			$update_time->invoke;
39	root	1.56
40			if ($ERROR_LOG) {
41			use IO::Handle;
42			open $errorlog, ">>$ERROR_LOG"
43			or die "$ERROR_LOG: $!";
44			$errorlog->autoflush(1);
45			}
46	root	1.27
47			if ($ACCESS_LOG) {
48			use IO::Handle;
49			open $accesslog, ">>$ACCESS_LOG"
50			or die "$ACCESS_LOG: $!";
51			$accesslog->autoflush(1);
52			}
53
54	root	1.1	sub slog {
55			my $level = shift;
56			my $format = shift;
57	root	1.56	my $NOW = (POSIX::strftime "%Y-%m-%d %H:%M:%S", gmtime $::NOW);
58			printf "$NOW: $format\n", @_;
59			printf $errorlog "$NOW: $format\n", @_ if $errorlog;
60	root	1.1	}
61
62	root	1.90	our $connections = new Coro::Semaphore $::MAX_CONNECTS \|\| 250;
63	root	1.44	our $httpevent = new Coro::Signal;
64	root	1.34
65	root	1.90	our $queue_file = new transferqueue $::MAX_TRANSFERS;
66	root	1.63	our $queue_index = new transferqueue 10;
67	root	1.62
68	root	1.90	our $tbf_top = new tbf rate => $::TBF_RATE \|\| 100000;
69	root	1.67
70	root	1.68	my $unused_bytes = 0;
71			my $unused_last = time;
72
73			sub unused_bandwidth {
74			$unused_bytes += $_[0];
75			if ($unused_last < $NOW - 30 && $unused_bytes / ($NOW - $unused_last) > 50000) {
76			$unused_last = $NOW;
77			$unused_bytes = 0;
78			$queue_file->force_wake_next;
79			slog 1, "forced filetransfer due to unused bandwidth";
80			}
81			}
82
83	root	1.40	sub listen_on {
84			my $listen = $_[0];
85
86			push @listen_sockets, $listen;
87
88			# the "main thread"
89			async {
90			slog 1, "accepting connections";
91			while () {
92			$connections->down;
93	root	1.88	my @conn = $listen->accept;
94	root	1.40	#slog 3, "accepted @$connections ".scalar(@pool);
95	root	1.88
96			async_pool {
97			eval {
98			conn->new (@conn)->handle;
99			};
100			slog 1, "$@" if $@ && !ref $@;
101
102			$httpevent->broadcast; # only for testing, but doesn't matter much
103
104			$connections->up;
105	root	1.40	}
106			}
107			};
108			}
109
110	root	1.4	my $http_port = new Coro::Socket
111	root	1.90	LocalAddr => $::SERVER_HOST,
112			LocalPort => $::SERVER_PORT,
113	root	1.4	ReuseAddr => 1,
114	root	1.81	Listen => 50,
115	root	1.4	or die "unable to start server";
116
117	root	1.40	listen_on $http_port;
118
119	root	1.90	if ($::SERVER_PORT2) {
120	root	1.41	my $http_port = new Coro::Socket
121	root	1.90	LocalAddr => $::SERVER_HOST,
122			LocalPort => $::SERVER_PORT2,
123	root	1.41	ReuseAddr => 1,
124	root	1.81	Listen => 50,
125	root	1.41	or die "unable to start server";
126	root	1.40
127	root	1.41	listen_on $http_port;
128			}
129	root	1.4
130	root	1.1	package conn;
131
132	root	1.90	use common::sense;
133	root	1.80
134	root	1.1	use Socket;
135			use HTTP::Date;
136	root	1.2	use Convert::Scalar 'weaken';
137	root	1.77	use IO::AIO;
138	root	1.16
139	root	1.77	IO::AIO::min_parallel $::AIO_PARALLEL;
140	root	1.16
141	root	1.84	our $AIO_WATCHER = EV::io IO::AIO::poll_fileno, EV::READ, \&IO::AIO::poll_cb;
142	root	1.16
143	root	1.26	our %conn; # $conn{ip}{self} => connobj
144			our %uri; # $uri{ip}{uri}{self}
145	root	1.3	our %blocked;
146	root	1.9	our %mimetype;
147
148			sub read_mimetypes {
149	root	1.81	if (open my $fh, "<mime_types") {
150			while (<$fh>) {
151	root	1.9	if (/^([^#]\S+)\t+(\S+)$/) {
152			$mimetype{lc $1} = $2;
153			}
154			}
155			} else {
156	root	1.10	print "cannot open mime_types\n";
157	root	1.9	}
158			}
159	root	1.1
160	root	1.10	read_mimetypes;
161
162	root	1.1	sub new {
163			my $class = shift;
164	root	1.42	my $fh = shift;
165	root	1.6	my $peername = shift;
166	root	1.2	my $self = bless { fh => $fh }, $class;
167	root	1.6	my (undef, $iaddr) = unpack_sockaddr_in $peername
168	root	1.81	or $self->err (500, "unable to decode peername");
169	root	1.7
170	root	1.50	$self->{remote_addr} =
171			$self->{remote_id} = inet_ntoa $iaddr;
172	root	1.60
173	root	1.11	$self->{time} = $::NOW;
174	root	1.2
175	root	1.49	weaken ($Coro::current->{conn} = $self);
176
177	root	1.82	++$::conns;
178	root	1.53	$::maxconns = $::conns if $::conns > $::maxconns;
179	root	1.13
180	root	1.81	$self
181	root	1.2	}
182
183			sub DESTROY {
184	root	1.82	my $self = shift;
185
186	root	1.81	--$::conns;
187	root	1.1	}
188
189	root	1.73	sub prune_cache {
190			my $hash = $_[0];
191
192			for (keys %$hash) {
193			if (ref $hash->{$_} eq HASH::) {
194			prune_cache($hash->{$_});
195			unless (scalar keys %{$hash->{$_}}) {
196			delete $hash->{$_};
197			}
198			}
199			}
200			}
201
202			sub prune_caches {
203			prune_cache \%conn;
204			prune_cache \%uri;
205
206			for (keys %blocked) {
207	root	1.74	delete $blocked{$_} unless $blocked{$_}[0] > $::NOW;
208	root	1.73	}
209			}
210
211	root	1.84	our $PRUNE_WATCHER = EV::timer 60, 60, \&prune_caches;
212	root	1.73
213	root	1.1	sub slog {
214	root	1.4	my $self = shift;
215	root	1.50	main::slog($_[0], "$self->{remote_id}> $_[1]");
216	root	1.1	}
217
218	root	1.4	sub response {
219	root	1.1	my ($self, $code, $msg, $hdr, $content) = @_;
220	root	1.17	my $res = "HTTP/1.1 $code $msg\015\012";
221	root	1.63	my $GZ = "";
222	root	1.1
223	root	1.46	if (exists $hdr->{Connection}) {
224			if ($hdr->{Connection} =~ /close/) {
225			$self->{h}{connection} = "close"
226			}
227			} else {
228			if ($self->{version} < 1.1) {
229			if ($self->{h}{connection} =~ /keep-alive/i) {
230			$hdr->{Connection} = "Keep-Alive";
231			} else {
232			$self->{h}{connection} = "close"
233			}
234			}
235			}
236	root	1.28
237	root	1.63	if ($self->{method} ne "HEAD"
238			&& $self->{h}{"accept-encoding"} =~ /\bgzip\b/
239	root	1.64	&& 400 < length $content
240	root	1.63	&& $hdr->{"Content-Length"} == length $content
241			&& !exists $hdr->{"Content-Encoding"}
242			) {
243			my $orig = length $content;
244			$hdr->{"Content-Encoding"} = "gzip";
245			$content = Compress::Zlib::memGzip(\$content);
246			$hdr->{"Content-Length"} = length $content;
247			$GZ = sprintf "GZ%02d", 100 - 100*((length $content) / $orig);
248			}
249
250	root	1.32	$res .= "Date: $HTTP_NOW\015\012";
251	root	1.71	$res .= "Server: $::NAME\015\012";
252	root	1.1
253			while (my ($h, $v) = each %$hdr) {
254			$res .= "$h: $v\015\012"
255			}
256	root	1.10	$res .= "\015\012";
257	root	1.4
258	root	1.13	$res .= $content if defined $content and $self->{method} ne "HEAD";
259	root	1.1
260	root	1.65	my $log = (POSIX::strftime "%Y-%m-%d %H:%M:%S", gmtime $::NOW).
261	root	1.63	" $self->{remote_id} \"$self->{uri}\" $code ".$hdr->{"Content-Length"}.$GZ.
262	root	1.50	" \"$self->{h}{referer}\"\n";
263	root	1.27
264	root	1.65	print $::accesslog $log if $::accesslog;
265	root	1.27	print STDERR $log;
266	root	1.2
267	root	1.67	$tbf_top->request(length $res, 1e6);
268			$self->{written} += print {$self->{fh}} $res;
269	root	1.1	}
270
271			sub err {
272			my $self = shift;
273			my ($code, $msg, $hdr, $content) = @_;
274
275			unless (defined $content) {
276	root	1.35	$content = "$code $msg\n";
277	root	1.1	$hdr->{"Content-Type"} = "text/plain";
278			$hdr->{"Content-Length"} = length $content;
279			}
280	root	1.17	$hdr->{"Connection"} = "close";
281	root	1.1
282	root	1.81	$self->response ($code, $msg, $hdr, $content);
283	root	1.1
284	root	1.81	die bless {}, err::
285	root	1.1	}
286
287			sub handle {
288			my $self = shift;
289			my $fh = $self->{fh};
290
291	root	1.29	my $host;
292
293	root	1.17	$fh->timeout($::REQ_TIMEOUT);
294	root	1.81	while () {
295	root	1.17	$self->{reqs}++;
296	root	1.1
297			# read request and parse first line
298			my $req = $fh->readline("\015\012\015\012");
299
300	root	1.17	unless (defined $req) {
301			if (exists $self->{version}) {
302			last;
303			} else {
304			$self->err(408, "request timeout");
305			}
306			}
307
308			$self->{h} = {};
309	root	1.1
310	root	1.17	$fh->timeout($::RES_TIMEOUT);
311	root	1.3
312	root	1.1	$req =~ /^(?:\015\012)?
313			(GET\|HEAD) \040+
314			([^\040]+) \040+
315			HTTP\/([0-9]+\.[0-9]+)
316			\015\012/gx
317	root	1.14	or $self->err(405, "method not allowed", { Allow => "GET,HEAD" });
318	root	1.1
319			$self->{method} = $1;
320			$self->{uri} = $2;
321	root	1.17	$self->{version} = $3;
322
323	root	1.20	$3 =~ /^1\./
324	root	1.17	or $self->err(506, "http protocol version $3 not supported");
325	root	1.1
326			# parse headers
327			{
328			my (%hdr, $h, $v);
329
330			$hdr{lc $1} .= ",$2"
331			while $req =~ /\G
332			([^:\000-\040]+):
333	root	1.66	[\011\040]*
334			((?: [^\015\012]+ \| \015\012[\011\040] )*)
335	root	1.1	\015\012
336			/gxc;
337
338			$req =~ /\G\015\012$/
339			or $self->err(400, "bad request");
340
341			$self->{h}{$h} = substr $v, 1
342			while ($h, $v) = each %hdr;
343			}
344
345	root	1.36	# remote id should be unique per user
346			my $id = $self->{remote_addr};
347
348			if (exists $self->{h}{"client-ip"}) {
349			$id .= "[".$self->{h}{"client-ip"}."]";
350			} elsif (exists $self->{h}{"x-forwarded-for"}) {
351			$id .= "[".$self->{h}{"x-forwarded-for"}."]";
352			}
353
354			$self->{remote_id} = $id;
355
356	root	1.56	weaken (local $conn{$id}{$self*1} = $self);
357
358	root	1.36	if ($blocked{$id}) {
359	root	1.56	$self->err_blocked
360			if $blocked{$id}[0] > $::NOW;
361	root	1.36
362			delete $blocked{$id};
363			}
364
365	root	1.29	# find out server name and port
366			if ($self->{uri} =~ s/^http:\/\/([^\/?#]*)//i) {
367			$host = $1;
368			} else {
369			$host = $self->{h}{host};
370			}
371
372			if (defined $host) {
373			$self->{server_port} = $host =~ s/:([0-9]+)$// ? $1 : 80;
374			} else {
375			($self->{server_port}, $host)
376	root	1.43	= unpack_sockaddr_in $self->{fh}->sockname
377	root	1.29	or $self->err(500, "unable to get socket name");
378			$host = inet_ntoa $host;
379			}
380
381			$self->{server_name} = $host;
382
383	root	1.56	weaken (local $uri{$id}{$self->{uri}}{$self*1} = $self);
384	root	1.1
385	root	1.24	eval {
386			$self->map_uri;
387			$self->respond;
388			};
389
390			die if $@ && !ref $@;
391	root	1.17
392	root	1.56	last if $self->{h}{connection} =~ /close/i;
393	root	1.17
394	root	1.44	$httpevent->broadcast;
395
396	root	1.17	$fh->timeout($::PER_TIMEOUT);
397			}
398	root	1.1	}
399
400	root	1.56	sub block {
401			my $self = shift;
402
403			$blocked{$self->{remote_id}} = [$::NOW + $_[0], $_[1]];
404			$self->slog(2, "blocked ip $self->{remote_id}");
405			$self->err_blocked;
406			}
407
408	root	1.1	# uri => path mapping
409			sub map_uri {
410			my $self = shift;
411	root	1.29	my $host = $self->{server_name};
412	root	1.1	my $uri = $self->{uri};
413
414	root	1.85	$host =~ /[\/\\]/
415			and $self->err(400, "bad request");
416
417	root	1.1	# some massaging, also makes it more secure
418			$uri =~ s/%([0-9a-fA-F][0-9a-fA-F])/chr hex $1/ge;
419			$uri =~ s%//+%/%g;
420			$uri =~ s%/\.(?=/\|$)%%g;
421			1 while $uri =~ s%/[^/]+/\.\.(?=/\|$)%%;
422
423			$uri =~ m%^/?\.\.(?=/\|$)%
424			and $self->err(400, "bad request");
425
426			$self->{name} = $uri;
427
428			# now do the path mapping
429			$self->{path} = "$::DOCROOT/$host$uri";
430	root	1.7
431			$self->access_check;
432	root	1.1	}
433
434			sub _cgi {
435			my $self = shift;
436			my $path = shift;
437			my $fh;
438
439			# no two-way xxx supported
440			if (0 == fork) {
441			open STDOUT, ">&".fileno($self->{fh});
442			if (chdir $::DOCROOT) {
443			$ENV{SERVER_SOFTWARE} = "thttpd-myhttpd"; # we are thttpd-alike
444	root	1.29	$ENV{HTTP_HOST} = $self->{server_name};
445			$ENV{HTTP_PORT} = $self->{server_port};
446	root	1.1	$ENV{SCRIPT_NAME} = $self->{name};
447	root	1.10	exec $path;
448	root	1.1	}
449			Coro::State::_exit(0);
450			} else {
451	root	1.29	die;
452	root	1.1	}
453			}
454
455	root	1.29	sub server_hostport {
456			$_[0]{server_port} == 80
457			? $_[0]{server_name}
458			: "$_[0]{server_name}:$_[0]{server_port}";
459			}
460
461	root	1.1	sub respond {
462			my $self = shift;
463			my $path = $self->{path};
464
465	root	1.49	if ($self->{name} =~ s%^/internal/([^/]+)%%) {
466			if ($::internal{$1}) {
467			$::internal{$1}->($self);
468			} else {
469	root	1.81	$self->err (404, "not found");
470	root	1.49	}
471			} else {
472	root	1.1
473	root	1.89	Coro::AIO::aio_stat $path
474			and $self->err (404, "not found");
475	root	1.10
476	root	1.49	$self->{stat} = [stat _];
477	root	1.1
478	root	1.49	# idiotic netscape sends idiotic headers AGAIN
479			my $ims = $self->{h}{"if-modified-since"} =~ /^([^;]+)/
480			? str2time $1 : 0;
481
482			if (-d _ && -r _) {
483			# directory
484			if ($path !~ /\/$/) {
485			# create a redirect to get the trailing "/"
486			# we don't try to avoid the :80
487	root	1.81	$self->err (301, "moved permanently", { Location => "http://".$self->server_hostport."$self->{uri}/" });
488	root	1.25	} else {
489	root	1.49	$ims < $self->{stat}[9]
490	root	1.81	or $self->err (304, "not modified");
491	root	1.49
492			if (-r "$path/index.html") {
493			# replace directory "size" by index.html filesize
494	root	1.55	$self->{stat} = [stat ($self->{path} .= "/index.html")];
495	root	1.81	$self->handle_file ($queue_index, $tbf_top);
496	root	1.49	} else {
497			$self->handle_dir;
498			}
499	root	1.1	}
500	root	1.49	} elsif (-f _ && -r _) {
501	root	1.81	-x _ and $self->err (403, "forbidden");
502	root	1.56
503	root	1.58	if (keys %{$conn{$self->{remote_id}}} > $::MAX_TRANSFERS_IP) {
504	root	1.56	my $timeout = $::NOW + 10;
505	root	1.58	while (keys %{$conn{$self->{remote_id}}} > $::MAX_TRANSFERS_IP) {
506	root	1.56	if ($timeout < $::NOW) {
507			$self->block($::BLOCKTIME, "too many connections");
508			} else {
509			$httpevent->wait;
510			}
511			}
512			}
513
514	root	1.81	$self->handle_file ($queue_file, $tbf_top);
515	root	1.49	} else {
516	root	1.81	$self->err (404, "not found");
517	root	1.1	}
518			}
519			}
520
521			sub handle_dir {
522			my $self = shift;
523	root	1.10	my $idx = $self->diridx;
524
525	root	1.81	$self->response (200, "ok",
526	root	1.10	{
527	root	1.76	"Content-Type" => "text/html; charset=utf-8",
528	root	1.10	"Content-Length" => length $idx,
529	root	1.55	"Last-Modified" => time2str ($self->{stat}[9]),
530	root	1.10	},
531			$idx);
532	root	1.1	}
533
534			sub handle_file {
535	root	1.67	my ($self, $queue, $tbf) = @_;
536	root	1.34	my $length = $self->{stat}[7];
537	root	1.1	my $hdr = {
538	root	1.75	"Last-Modified" => time2str ((stat _)[9]),
539			"Accept-Ranges" => "bytes",
540	root	1.1	};
541
542			my @code = (200, "ok");
543			my ($l, $h);
544
545	root	1.86	if ($self->{h}{range} =~ /^bytes=(.*)$/i) {
546	root	1.1	for (split /,/, $1) {
547			if (/^-(\d+)$/) {
548			($l, $h) = ($length - $1, $length - 1);
549			} elsif (/^(\d+)-(\d*)$/) {
550			($l, $h) = ($1, ($2 ne "" \|\| $2 >= $length) ? $2 : $length - 1);
551			} else {
552			($l, $h) = (0, $length - 1);
553			goto ignore;
554			}
555	root	1.26	goto satisfiable if $l >= 0 && $l < $length && $h >= 0 && $h >= $l;
556	root	1.1	}
557			$hdr->{"Content-Range"} = "bytes */$length";
558	root	1.24	$hdr->{"Content-Length"} = $length;
559	root	1.81	$self->err (416, "not satisfiable", $hdr, "");
560	root	1.1
561			satisfiable:
562	root	1.4	# check for segmented downloads
563	root	1.10	if ($l && $::NO_SEGMENTED) {
564	root	1.56	my $timeout = $::NOW + 15;
565	root	1.58	while (keys %{$uri{$self->{remote_id}}{$self->{uri}}} > 1) {
566	root	1.56	if ($timeout <= $::NOW) {
567	root	1.81	$self->block ($::BLOCKTIME, "segmented downloads are forbidden");
568	root	1.56	#$self->err_segmented_download;
569	root	1.29	} else {
570	root	1.46	$httpevent->wait;
571	root	1.29	}
572	root	1.4	}
573			}
574
575	root	1.1	$hdr->{"Content-Range"} = "bytes $l-$h/$length";
576			@code = (206, "partial content");
577			$length = $h - $l + 1;
578
579			ignore:
580			} else {
581			($l, $h) = (0, $length - 1);
582			}
583
584	root	1.9	$self->{path} =~ /\.([^.]+)$/;
585			$hdr->{"Content-Type"} = $mimetype{lc $1} \|\| "application/octet-stream";
586	root	1.1	$hdr->{"Content-Length"} = $length;
587
588	root	1.81	$self->response (@code, $hdr, "");
589	root	1.1
590			if ($self->{method} eq "GET") {
591	root	1.32	$self->{time} = $::NOW;
592	root	1.71	$self->{written} = 0;
593	root	1.32
594	root	1.90	my $fh = Coro::AIO::aio_open $self->{path}, Fcntl::O_RDONLY, 0
595	root	1.1	or die "$self->{path}: late open failure ($!)";
596
597			$h -= $l - 1;
598
599	root	1.81	my $transfer = $queue->start_transfer ($h);
600	root	1.51	my $locked;
601	root	1.48	my $bufsize = $::WAIT_BUFSIZE; # initial buffer size
602	root	1.19
603	root	1.1	while ($h > 0) {
604	root	1.90	Coro::cede;
605			my $transfer_lock = $TRANSFER_LOCK->guard ($self->{remote_id});
606
607	root	1.51	unless ($locked) {
608	root	1.81	if ($locked \|\|= $transfer->try ($::WAIT_INTERVAL)) {
609	root	1.48	$bufsize = $::BUFSIZE;
610			$self->{time} = $::NOW;
611	root	1.71	$self->{written} = 0;
612	root	1.48	}
613	root	1.56	}
614
615			if ($blocked{$self->{remote_id}}) {
616			$self->{h}{connection} = "close";
617	root	1.59	die bless {}, err::;
618	root	1.48	}
619
620	root	1.80	Coro::AIO::aio_read $fh, $l, ($h > $bufsize ? $bufsize : $h), my $buf, 0
621	root	1.79	or last;
622	root	1.67
623	root	1.79	$tbf->request (length $buf);
624	root	1.87	my $w = $self->{fh}->syswrite ($buf)
625	root	1.1	or last;
626	root	1.11	$::written += $w;
627			$self->{written} += $w;
628	root	1.80	$l += $w;
629	root	1.1	}
630	root	1.32
631			close $fh;
632	root	1.1	}
633	root	1.7	}
634
635	root	1.78	1
636	root	1.79