1 # Copyright (C) 2016-2021 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
11 # httpd: PublicInbox::HTTPD ref
12 # env: PSGI env hashref
13 # input_left: bytes left to read in request body (e.g. POST/PUT)
14 # remote_addr: remote IP address as a string (e.g. "127.0.0.1")
15 # remote_port: peer port
16 # forward: response body object, response to ->getline + ->close
17 # alive: HTTP keepalive state:
18 # 0: drop connection when done
19 # 1: keep connection when done
20 # 2: keep connection, chunk responses
21 package PublicInbox::HTTP;
23 use parent qw(PublicInbox::DS);
25 use Plack::HTTPParser qw(parse_http_request); # XS or pure Perl
27 use HTTP::Status qw(status_message);
28 use HTTP::Date qw(time2str);
29 use IO::Handle; # ->write
30 use PublicInbox::DS qw(msg_more);
31 use PublicInbox::Syscall qw(EPOLLIN EPOLLONESHOT);
32 use PublicInbox::Tmpfile;
34 CHUNK_START => -1, # [a-f0-9]+\r\n
35 CHUNK_END => -2, # \r\n
36 CHUNK_ZEND => -3, # \r\n
42 sub process_pipelineq () {
46 next unless $_->{sock};
51 # Use the same configuration parameter as git since this is primarily
52 # a slow-client sponge for git-http-backend
53 # TODO: support per-respository http.maxRequestBuffer somehow...
54 our $MAX_REQUEST_BUFFER = $ENV{GIT_HTTP_MAX_REQUEST_BUFFER} ||
57 open(my $null_io, '<', '/dev/null') or die "failed to open /dev/null: $!";
62 $now == $prev ? $http_date : ($http_date = time2str($prev = $now));
66 my ($class, $sock, $addr, $httpd) = @_;
67 my $self = bless { httpd => $httpd }, $class;
70 if ($sock->can('accept_SSL') && !$sock->accept_SSL) {
71 return CORE::close($sock) if $! != EAGAIN;
72 $ev = PublicInbox::TLS::epollbit() or return CORE::close($sock);
73 $wbuf = [ \&PublicInbox::DS::accept_tls_step ];
75 $self->{wbuf} = $wbuf if $wbuf;
76 ($self->{remote_addr}, $self->{remote_port}) =
77 PublicInbox::Daemon::host_with_port($addr);
78 $self->SUPER::new($sock, $ev | EPOLLONESHOT);
81 sub event_step { # called by PublicInbox::DS
84 return unless $self->flush_write && $self->{sock};
86 # only read more requests if we've drained the write buffer,
87 # otherwise we can be buffering infinitely w/o backpressure
89 return read_input($self) if ref($self->{env});
90 my $rbuf = $self->{rbuf} // (\(my $x = ''));
91 $self->do_read($rbuf, 8192, length($$rbuf)) or return;
92 rbuf_process($self, $rbuf);
96 my ($self, $rbuf) = @_;
97 $rbuf //= $self->{rbuf} // (\(my $x = ''));
99 my %env = %{$self->{httpd}->{env}}; # full hash copy
100 my $r = parse_http_request($$rbuf, \%env);
102 # We do not support Trailers in chunked requests, for now
103 # (they are rarely-used and git (as of 2.7.2) does not use them)
104 if ($r == -1 || $env{HTTP_TRAILER} ||
105 # this length-check is necessary for PURE_PERL=1:
106 ($r == -2 && length($$rbuf) > 0x4000)) {
107 return quit($self, 400);
109 if ($r < 0) { # incomplete
110 $self->rbuf_idle($rbuf);
111 return $self->requeue;
113 $$rbuf = substr($$rbuf, $r);
114 my $len = input_prepare($self, \%env);
115 defined $len or return write_err($self, undef); # EMFILE/ENFILE
117 $len ? read_input($self, $rbuf) : app_dispatch($self, undef, $rbuf);
120 # IO::Handle::write returns boolean, this returns bytes written:
122 my ($fh, $rbuf, $max) = @_;
123 my $w = length($$rbuf);
124 $w = $max if $w > $max;
125 $fh->write($$rbuf, $w) or return;
129 sub read_input ($;$) {
130 my ($self, $rbuf) = @_;
131 $rbuf //= $self->{rbuf} // (\(my $x = ''));
132 my $env = $self->{env};
133 return read_input_chunked($self, $rbuf) if env_chunked($env);
135 # env->{CONTENT_LENGTH} (identity)
136 my $len = delete $self->{input_left};
137 my $input = $env->{'psgi.input'};
141 my $w = xwrite($input, $rbuf, $len);
142 return write_err($self, $len) unless $w;
144 die "BUG: $len < 0 (w=$w)" if $len < 0;
145 if ($len == 0) { # next request may be pipelined
146 $$rbuf = substr($$rbuf, $w);
151 $self->do_read($rbuf, 8192) or return recv_err($self, $len);
152 # continue looping if $r > 0;
154 app_dispatch($self, $input, $rbuf);
158 my ($self, $input, $rbuf) = @_;
159 $self->rbuf_idle($rbuf);
160 my $env = $self->{env};
161 $self->{env} = undef; # for exists() check in ->busy
162 $env->{REMOTE_ADDR} = $self->{remote_addr};
163 $env->{REMOTE_PORT} = $self->{remote_port};
164 if (defined(my $host = $env->{HTTP_HOST})) {
165 $host =~ s/:([0-9]+)\z// and $env->{SERVER_PORT} = $1;
166 $env->{SERVER_NAME} = $host;
168 if (defined $input) {
169 sysseek($input, 0, SEEK_SET) or
170 die "BUG: psgi.input seek failed: $!";
172 # note: NOT $self->{sock}, we want our close (+ PublicInbox::DS::close),
173 # to do proper cleanup:
174 $env->{'psgix.io'} = $self; # for ->close or async_pass
175 my $res = Plack::Util::run_app($self->{httpd}->{app}, $env);
177 if (ref($res) eq 'CODE') {
178 $res->(sub { response_write($self, $env, $_[0]) });
180 response_write($self, $env, $res);
184 err($self, "response_write error: $@");
189 sub response_header_write {
190 my ($self, $env, $res) = @_;
191 my $proto = $env->{SERVER_PROTOCOL} or return; # HTTP/0.9 :P
192 my $status = $res->[0];
193 my $h = "$proto $status " . status_message($status) . "\r\n";
195 my $headers = $res->[1];
197 for (my $i = 0; $i < @$headers; $i += 2) {
198 my $k = $headers->[$i];
199 my $v = $headers->[$i + 1];
200 next if $k =~ /\A(?:Connection|Date)\z/i;
202 $len = $v if $k =~ /\AContent-Length\z/i;
203 if ($k =~ /\ATransfer-Encoding\z/i && $v =~ /\bchunked\b/i) {
209 my $conn = $env->{HTTP_CONNECTION} || '';
210 my $term = defined($len) || $chunked;
211 my $prot_persist = ($proto eq 'HTTP/1.1') && ($conn !~ /\bclose\b/i);
213 if (!$term && $prot_persist) { # auto-chunk
214 $chunked = $alive = 2;
215 $h .= "Transfer-Encoding: chunked\r\n";
216 # no need for "Connection: keep-alive" with HTTP/1.1
217 } elsif ($term && ($prot_persist || ($conn =~ /\bkeep-alive\b/i))) {
219 $h .= "Connection: keep-alive\r\n";
222 $h .= "Connection: close\r\n";
224 $h .= 'Date: ' . http_date() . "\r\n\r\n";
226 if (($len || $chunked) && $env->{REQUEST_METHOD} ne 'HEAD') {
234 # middlewares such as Deflater may write empty strings
235 sub chunked_write ($$) {
237 return if $_[1] eq '';
238 msg_more($self, sprintf("%x\r\n", length($_[1])));
239 msg_more($self, $_[1]);
241 # use $self->write(\"\n\n") if you care about real-time
242 # streaming responses, public-inbox WWW does not.
243 msg_more($self, "\r\n");
246 sub identity_write ($$) {
248 $self->write(\($_[1])) if $_[1] ne '';
251 sub next_request ($) {
254 # avoid recursion for pipelined requests
255 PublicInbox::DS::requeue(\&process_pipelineq) if !@$pipelineq;
256 push @$pipelineq, $self;
257 } else { # wait for next request
263 my ($self, $alive) = @_;
264 delete $self->{env}; # we're no longer busy
265 $self->write(\"0\r\n\r\n") if $alive == 2;
266 $self->write($alive ? \&next_request : \&close);
271 my $forward = $self->{forward};
273 # limit our own running time for fairness with other
274 # clients and to avoid buffering too much:
281 # may close in PublicInbox::DS::write
282 if ($self->{alive} == 2) {
283 chunked_write($self, $buf);
285 identity_write($self, $buf);
290 my $new_size = push(@{$self->{wbuf}}, \&getline_pull);
292 # wbuf may be populated by {chunked,identity}_write()
293 # above, no need to rearm if so:
294 $self->requeue if $new_size == 1;
298 err($self, "response ->getline error: $@");
302 if (delete $self->{forward}) {
303 eval { $forward->close };
305 err($self, "response ->close error: $@");
306 $self->close; # idempotent
309 response_done($self, delete $self->{alive});
313 my ($self, $env, $res) = @_;
314 my $alive = response_header_write($self, $env, $res);
315 if (defined(my $body = $res->[2])) {
316 if (ref $body eq 'ARRAY') {
318 chunked_write($self, $_) for @$body;
320 identity_write($self, $_) for @$body;
322 response_done($self, $alive);
324 $self->{forward} = $body;
325 $self->{alive} = $alive;
326 getline_pull($self); # kick-off!
328 # these are returned to the calling application:
329 } elsif ($alive == 2) {
330 bless [ $self, $alive ], 'PublicInbox::HTTP::Chunked';
332 bless [ $self, $alive ], 'PublicInbox::HTTP::Identity';
336 sub input_tmpfile ($) {
337 my $input = tmpfile('http.input', $_[0]->{sock}) or return;
338 $input->autoflush(1);
343 my ($self, $env) = @_;
346 # rfc 7230 3.3.2, 3.3.3,: favor Transfer-Encoding over Content-Length
347 my $hte = $env->{HTTP_TRANSFER_ENCODING};
349 # rfc7230 3.3.3, point 3 says only chunked is accepted
350 # as the final encoding. Since neither public-inbox-httpd,
351 # git-http-backend, or our WWW-related code uses "gzip",
352 # "deflate" or "compress" as the Transfer-Encoding, we'll
354 return quit($self, 400) if $hte !~ /\Achunked\z/i;
357 $input = input_tmpfile($self);
359 $len = $env->{CONTENT_LENGTH};
362 return quit($self, 400) if $len !~ /\A[0-9]+\z/;
364 return quit($self, 413) if $len > $MAX_REQUEST_BUFFER;
365 $input = $len ? input_tmpfile($self) : $null_io;
371 # TODO: expire idle clients on ENFILE / EMFILE
372 return unless $input;
374 $env->{'psgi.input'} = $input;
376 $self->{input_left} = $len || 0;
379 sub env_chunked { ($_[0]->{HTTP_TRANSFER_ENCODING} // '') =~ /\Achunked\z/i }
382 eval { $_[0]->{httpd}->{env}->{'psgi.errors'}->print($_[1]."\n") };
386 my ($self, $len) = @_;
387 my $msg = $! || '(zero write)';
388 $msg .= " ($len bytes remaining)" if defined $len;
389 err($self, "error buffering to input: $msg");
394 my ($self, $len) = @_;
395 if ($! == EAGAIN) { # epoll/kevent watch already set by do_read
396 $self->{input_left} = $len;
398 err($self, "error reading input: $! ($len bytes remaining)");
402 sub read_input_chunked { # unlikely...
403 my ($self, $rbuf) = @_;
404 $rbuf //= $self->{rbuf} // (\(my $x = ''));
405 my $input = $self->{env}->{'psgi.input'};
406 my $len = delete $self->{input_left};
408 while (1) { # chunk start
409 if ($len == CHUNK_ZEND) {
410 $$rbuf =~ s/\A\r\n//s and
411 return app_dispatch($self, $input, $rbuf);
413 return quit($self, 400) if length($$rbuf) > 2;
415 if ($len == CHUNK_END) {
416 if ($$rbuf =~ s/\A\r\n//s) {
418 } elsif (length($$rbuf) > 2) {
419 return quit($self, 400);
422 if ($len == CHUNK_START) {
423 if ($$rbuf =~ s/\A([a-f0-9]+).*?\r\n//i) {
425 if (($len + -s $input) > $MAX_REQUEST_BUFFER) {
426 return quit($self, 413);
428 } elsif (length($$rbuf) > CHUNK_MAX_HDR) {
429 return quit($self, 400);
431 # will break from loop since $len >= 0
434 if ($len < 0) { # chunk header is trickled, read more
435 $self->do_read($rbuf, 8192, length($$rbuf)) or
436 return recv_err($self, $len);
437 # (implicit) goto chunk_start if $r > 0;
439 $len = CHUNK_ZEND if $len == 0;
441 # drain the current chunk
444 my $w = xwrite($input, $rbuf, $len);
445 return write_err($self, "$len chunk") if !$w;
448 # we may have leftover data to parse
450 $$rbuf = substr($$rbuf, $w);
453 die "BUG: len < 0: $len";
459 # read more of current chunk
460 $self->do_read($rbuf, 8192) or
461 return recv_err($self, $len);
468 my ($self, $status) = @_;
469 my $h = "HTTP/1.1 $status " . status_message($status) . "\r\n\r\n";
472 undef; # input_prepare expects this
477 if (my $forward = delete $self->{forward}) {
478 eval { $forward->close };
479 err($self, "forward ->close error: $@") if $@;
481 $self->SUPER::close; # PublicInbox::DS::close
484 # for graceful shutdown in PublicInbox::Daemon:
487 ($self->{rbuf} || exists($self->{env}) || $self->{wbuf});
490 # runs $cb on the next iteration of the event loop at earliest
492 my ($self, $cb) = @_;
493 return unless exists $self->{sock};
494 $self->requeue if 1 == push(@{$self->{wbuf}}, $cb);
497 # Chunked and Identity packages are used for writing responses.
498 # They may be exposed to the PSGI application when the PSGI app
499 # returns a CODE ref for "push"-based responses
500 package PublicInbox::HTTP::Chunked;
504 # ([$http], $buf) = @_;
505 PublicInbox::HTTP::chunked_write($_[0]->[0], $_[1])
509 # $_[0] = [$http, $alive]
510 PublicInbox::HTTP::response_done(@{$_[0]});
513 package PublicInbox::HTTP::Identity;
515 our @ISA = qw(PublicInbox::HTTP::Chunked);
518 # ([$http], $buf) = @_;
519 PublicInbox::HTTP::identity_write($_[0]->[0], $_[1]);