1 # Copyright (C) 2016-2019 all contributors <meta@public-inbox.org>
2 # License: AGPL-3.0+ <https://www.gnu.org/licenses/agpl-3.0.txt>
4 # Generic PSGI server for convenience. It aims to provide
5 # a consistent experience for public-inbox admins so they don't have
6 # to learn different ways to admin both NNTP and HTTP components.
7 # There's nothing which depends on public-inbox, here.
8 # Each instance of this class represents a HTTP client socket
10 package PublicInbox::HTTP;
13 use base qw(PublicInbox::DS);
14 use fields qw(httpd env input_left remote_addr remote_port forward);
15 use bytes (); # only for bytes::length
17 use Plack::HTTPParser qw(parse_http_request); # XS or pure Perl
18 use HTTP::Status qw(status_message);
19 use HTTP::Date qw(time2str);
21 require PublicInbox::EvCleanup;
22 use PublicInbox::DS qw(msg_more);
23 use PublicInbox::Syscall qw(EPOLLIN EPOLLONESHOT);
24 use PublicInbox::Tmpfile;
26 CHUNK_START => -1, # [a-f0-9]+\r\n
27 CHUNK_END => -2, # \r\n
28 CHUNK_ZEND => -3, # \r\n
34 sub process_pipelineq () {
38 next unless $_->{sock};
43 # Use the same configuration parameter as git since this is primarily
44 # a slow-client sponge for git-http-backend
45 # TODO: support per-respository http.maxRequestBuffer somehow...
46 our $MAX_REQUEST_BUFFER = $ENV{GIT_HTTP_MAX_REQUEST_BUFFER} ||
49 open(my $null_io, '<', '/dev/null') or die "failed to open /dev/null: $!";
54 $now == $prev ? $http_date : ($http_date = time2str($prev = $now));
58 my ($class, $sock, $addr, $httpd) = @_;
59 my $self = fields::new($class);
62 if (ref($sock) eq 'IO::Socket::SSL' && !$sock->accept_SSL) {
63 return CORE::close($sock) if $! != EAGAIN;
64 $ev = PublicInbox::TLS::epollbit();
65 $wbuf = [ \&PublicInbox::DS::accept_tls_step ];
67 $self->SUPER::new($sock, $ev | EPOLLONESHOT);
68 $self->{httpd} = $httpd;
69 $self->{wbuf} = $wbuf if $wbuf;
70 ($self->{remote_addr}, $self->{remote_port}) =
71 PublicInbox::Daemon::host_with_port($addr);
75 sub event_step { # called by PublicInbox::DS
78 return unless $self->flush_write && $self->{sock};
80 # only read more requests if we've drained the write buffer,
81 # otherwise we can be buffering infinitely w/o backpressure
83 return read_input($self) if defined $self->{env};
84 my $rbuf = $self->{rbuf} // (\(my $x = ''));
85 $self->do_read($rbuf, 8192, bytes::length($$rbuf)) or return;
86 rbuf_process($self, $rbuf);
90 my ($self, $rbuf) = @_;
91 $rbuf //= $self->{rbuf} // (\(my $x = ''));
93 my %env = %{$self->{httpd}->{env}}; # full hash copy
94 my $r = parse_http_request($$rbuf, \%env);
96 # We do not support Trailers in chunked requests, for now
97 # (they are rarely-used and git (as of 2.7.2) does not use them)
98 if ($r == -1 || $env{HTTP_TRAILER} ||
99 # this length-check is necessary for PURE_PERL=1:
100 ($r == -2 && bytes::length($$rbuf) > 0x4000)) {
101 return quit($self, 400);
103 if ($r < 0) { # incomplete
104 $self->rbuf_idle($rbuf);
105 return $self->requeue;
107 $$rbuf = substr($$rbuf, $r);
108 my $len = input_prepare($self, \%env);
109 defined $len or return write_err($self, undef); # EMFILE/ENFILE
111 $len ? read_input($self, $rbuf) : app_dispatch($self, undef, $rbuf);
114 # IO::Handle::write returns boolean, this returns bytes written:
116 my ($fh, $rbuf, $max) = @_;
117 my $w = bytes::length($$rbuf);
118 $w = $max if $w > $max;
119 $fh->write($$rbuf, $w) or return;
123 sub read_input ($;$) {
124 my ($self, $rbuf) = @_;
125 $rbuf //= $self->{rbuf} // (\(my $x = ''));
126 my $env = $self->{env};
127 return if $env->{REMOTE_ADDR}; # in app dispatch
128 return read_input_chunked($self, $rbuf) if env_chunked($env);
130 # env->{CONTENT_LENGTH} (identity)
131 my $len = delete $self->{input_left};
132 my $input = $env->{'psgi.input'};
136 my $w = xwrite($input, $rbuf, $len);
137 return write_err($self, $len) unless $w;
139 die "BUG: $len < 0 (w=$w)" if $len < 0;
140 if ($len == 0) { # next request may be pipelined
141 $$rbuf = substr($$rbuf, $w);
146 $self->do_read($rbuf, 8192) or return recv_err($self, $len);
147 # continue looping if $r > 0;
149 app_dispatch($self, $input, $rbuf);
153 my ($self, $input, $rbuf) = @_;
154 $self->rbuf_idle($rbuf);
155 my $env = $self->{env};
156 $env->{REMOTE_ADDR} = $self->{remote_addr};
157 $env->{REMOTE_PORT} = $self->{remote_port};
158 if (my $host = $env->{HTTP_HOST}) {
159 $host =~ s/:([0-9]+)\z// and $env->{SERVER_PORT} = $1;
160 $env->{SERVER_NAME} = $host;
162 if (defined $input) {
163 sysseek($input, 0, SEEK_SET) or
164 die "BUG: psgi.input seek failed: $!";
166 # note: NOT $self->{sock}, we want our close (+ PublicInbox::DS::close),
167 # to do proper cleanup:
168 $env->{'psgix.io'} = $self; # only for ->close
169 my $res = Plack::Util::run_app($self->{httpd}->{app}, $env);
171 if (ref($res) eq 'CODE') {
172 $res->(sub { response_write($self, $env, $_[0]) });
174 response_write($self, $env, $res);
180 sub response_header_write {
181 my ($self, $env, $res) = @_;
182 my $proto = $env->{SERVER_PROTOCOL} or return; # HTTP/0.9 :P
183 my $status = $res->[0];
184 my $h = "$proto $status " . status_message($status) . "\r\n";
186 my $headers = $res->[1];
188 for (my $i = 0; $i < @$headers; $i += 2) {
189 my $k = $headers->[$i];
190 my $v = $headers->[$i + 1];
191 next if $k =~ /\A(?:Connection|Date)\z/i;
193 $len = $v if $k =~ /\AContent-Length\z/i;
194 if ($k =~ /\ATransfer-Encoding\z/i && $v =~ /\bchunked\b/i) {
200 my $conn = $env->{HTTP_CONNECTION} || '';
201 my $term = defined($len) || $chunked;
202 my $prot_persist = ($proto eq 'HTTP/1.1') && ($conn !~ /\bclose\b/i);
204 if (!$term && $prot_persist) { # auto-chunk
205 $chunked = $alive = 2;
206 $h .= "Transfer-Encoding: chunked\r\n";
207 # no need for "Connection: keep-alive" with HTTP/1.1
208 } elsif ($term && ($prot_persist || ($conn =~ /\bkeep-alive\b/i))) {
210 $h .= "Connection: keep-alive\r\n";
213 $h .= "Connection: close\r\n";
215 $h .= 'Date: ' . http_date() . "\r\n\r\n";
217 if (($len || $chunked) && $env->{REQUEST_METHOD} ne 'HEAD') {
225 # middlewares such as Deflater may write empty strings
226 sub chunked_wcb ($) {
229 return if $_[0] eq '';
230 msg_more($self, sprintf("%x\r\n", bytes::length($_[0])));
231 msg_more($self, $_[0]);
233 # use $self->write(\"\n\n") if you care about real-time
234 # streaming responses, public-inbox WWW does not.
235 msg_more($self, "\r\n");
239 sub identity_wcb ($) {
241 sub { $self->write(\($_[0])) if $_[0] ne '' }
244 sub next_request ($) {
247 # avoid recursion for pipelined requests
248 PublicInbox::DS::requeue(\&process_pipelineq) if !@$pipelineq;
249 push @$pipelineq, $self;
250 } else { # wait for next request
255 sub response_done_cb ($$) {
256 my ($self, $alive) = @_;
258 delete $self->{env}; # we're no longer busy
259 $self->write(\"0\r\n\r\n") if $alive == 2;
260 $self->write($alive ? \&next_request : \&close);
264 sub getline_response ($$$) {
265 my ($self, $write, $close) = @_;
266 my $pull; # DANGER: self-referential
268 my $forward = $self->{forward};
269 # limit our own running time for fairness with other
270 # clients and to avoid buffering too much:
277 $write->($buf); # may close in PublicInbox::DS::write
280 my $wbuf = $self->{wbuf} ||= [];
283 # wbuf may be populated by $write->($buf),
284 # no need to rearm if so:
285 $self->requeue if scalar(@$wbuf) == 1;
289 err($self, "response ->getline error: $@");
293 $pull = undef; # all done!
295 if (delete $self->{forward}) {
296 eval { $forward->close };
298 err($self, "response ->close error: $@");
299 $self->close; # idempotent
303 $close->(); # call response_done_cb
306 $pull->(); # kick-off!
310 my ($self, $env, $res) = @_;
311 my $alive = response_header_write($self, $env, $res);
312 my $close = response_done_cb($self, $alive);
313 my $write = $alive == 2 ? chunked_wcb($self) : identity_wcb($self);
314 if (defined(my $body = $res->[2])) {
315 if (ref $body eq 'ARRAY') {
316 $write->($_) foreach @$body;
319 $self->{forward} = $body;
320 getline_response($self, $write, $close);
323 # this is returned to the calling application:
324 Plack::Util::inline_object(write => $write, close => $close);
328 sub input_tmpfile ($) {
329 my $input = tmpfile('http.input', $_[0]->{sock}) or return;
330 $input->autoflush(1);
335 my ($self, $env) = @_;
337 my $len = $env->{CONTENT_LENGTH};
339 if ($len > $MAX_REQUEST_BUFFER) {
343 $input = input_tmpfile($self);
344 } elsif (env_chunked($env)) {
346 $input = input_tmpfile($self);
351 # TODO: expire idle clients on ENFILE / EMFILE
352 return unless $input;
354 $env->{'psgi.input'} = $input;
356 $self->{input_left} = $len || 0;
359 sub env_chunked { ($_[0]->{HTTP_TRANSFER_ENCODING} || '') =~ /\bchunked\b/i }
362 eval { $_[0]->{httpd}->{env}->{'psgi.errors'}->print($_[1]."\n") };
366 my ($self, $len) = @_;
367 my $msg = $! || '(zero write)';
368 $msg .= " ($len bytes remaining)" if defined $len;
369 err($self, "error buffering to input: $msg");
374 my ($self, $len) = @_;
375 if ($! == EAGAIN) { # epoll/kevent watch already set by do_read
376 $self->{input_left} = $len;
378 err($self, "error reading input: $! ($len bytes remaining)");
382 sub read_input_chunked { # unlikely...
383 my ($self, $rbuf) = @_;
384 $rbuf //= $self->{rbuf} // (\(my $x = ''));
385 my $input = $self->{env}->{'psgi.input'};
386 my $len = delete $self->{input_left};
388 while (1) { # chunk start
389 if ($len == CHUNK_ZEND) {
390 $$rbuf =~ s/\A\r\n//s and
391 return app_dispatch($self, $input, $rbuf);
393 return quit($self, 400) if bytes::length($$rbuf) > 2;
395 if ($len == CHUNK_END) {
396 if ($$rbuf =~ s/\A\r\n//s) {
398 } elsif (bytes::length($$rbuf) > 2) {
399 return quit($self, 400);
402 if ($len == CHUNK_START) {
403 if ($$rbuf =~ s/\A([a-f0-9]+).*?\r\n//i) {
405 if (($len + -s $input) > $MAX_REQUEST_BUFFER) {
406 return quit($self, 413);
408 } elsif (bytes::length($$rbuf) > CHUNK_MAX_HDR) {
409 return quit($self, 400);
411 # will break from loop since $len >= 0
414 if ($len < 0) { # chunk header is trickled, read more
415 $self->do_read($rbuf, 8192, bytes::length($$rbuf)) or
416 return recv_err($self, $len);
417 # (implicit) goto chunk_start if $r > 0;
419 $len = CHUNK_ZEND if $len == 0;
421 # drain the current chunk
424 my $w = xwrite($input, $rbuf, $len);
425 return write_err($self, "$len chunk") if !$w;
428 # we may have leftover data to parse
430 $$rbuf = substr($$rbuf, $w);
433 die "BUG: len < 0: $len";
439 # read more of current chunk
440 $self->do_read($rbuf, 8192) or
441 return recv_err($self, $len);
448 my ($self, $status) = @_;
449 my $h = "HTTP/1.1 $status " . status_message($status) . "\r\n\r\n";
456 delete $self->{env}; # prevent circular references
457 if (my $forward = delete $self->{forward}) {
458 eval { $forward->close };
459 err($self, "forward ->close error: $@") if $@;
461 $self->SUPER::close; # PublicInbox::DS::close
464 # for graceful shutdown in PublicInbox::Daemon:
467 ($self->{rbuf} || $self->{env} || $self->{wbuf});