-# Copyright (C) 2016 all contributors <meta@public-inbox.org>
+# Copyright (C) 2016-2019 all contributors <meta@public-inbox.org>
# License: AGPL-3.0+ <https://www.gnu.org/licenses/agpl-3.0.txt>
#
# Generic PSGI server for convenience. It aims to provide
# a consistent experience for public-inbox admins so they don't have
# to learn different ways to admin both NNTP and HTTP components.
-# There's nothing public-inbox-specific, here.
+# There's nothing which depends on public-inbox, here.
# Each instance of this class represents a HTTP client socket
package PublicInbox::HTTP;
use strict;
use warnings;
-use base qw(Danga::Socket);
-use fields qw(httpd env rbuf input_left remote_addr remote_port);
+use base qw(PublicInbox::DS);
+use fields qw(httpd env input_left remote_addr remote_port forward alive);
+use bytes (); # only for bytes::length
use Fcntl qw(:seek);
use Plack::HTTPParser qw(parse_http_request); # XS or pure Perl
+use Plack::Util;
use HTTP::Status qw(status_message);
use HTTP::Date qw(time2str);
-use IO::File;
+use IO::Handle; # ->write
+use PublicInbox::DS qw(msg_more);
+use PublicInbox::Syscall qw(EPOLLIN EPOLLONESHOT);
+use PublicInbox::Tmpfile;
use constant {
CHUNK_START => -1, # [a-f0-9]+\r\n
CHUNK_END => -2, # \r\n
CHUNK_ZEND => -3, # \r\n
CHUNK_MAX_HDR => 256,
};
+use Errno qw(EAGAIN);
-# FIXME: duplicated code with NNTP.pm
-my $WEAKEN = {}; # string(inbox) -> inbox
-my $WEAKTIMER;
-sub weaken_task () {
- $WEAKTIMER = undef;
- $_->weaken_all for values %$WEAKEN;
- $WEAKEN = {};
+my $pipelineq = [];
+sub process_pipelineq () {
+ my $q = $pipelineq;
+ $pipelineq = [];
+ foreach (@$q) {
+ next unless $_->{sock};
+ rbuf_process($_);
+ }
}
# Use the same configuration parameter as git since this is primarily
our $MAX_REQUEST_BUFFER = $ENV{GIT_HTTP_MAX_REQUEST_BUFFER} ||
(10 * 1024 * 1024);
-my $null_io = IO::File->new('/dev/null', '<');
+open(my $null_io, '<', '/dev/null') or die "failed to open /dev/null: $!";
my $http_date;
my $prev = 0;
sub http_date () {
sub new ($$$) {
my ($class, $sock, $addr, $httpd) = @_;
my $self = fields::new($class);
- $self->SUPER::new($sock);
+ my $ev = EPOLLIN;
+ my $wbuf;
+ if (ref($sock) eq 'IO::Socket::SSL' && !$sock->accept_SSL) {
+ return CORE::close($sock) if $! != EAGAIN;
+ $ev = PublicInbox::TLS::epollbit();
+ $wbuf = [ \&PublicInbox::DS::accept_tls_step ];
+ }
+ $self->SUPER::new($sock, $ev | EPOLLONESHOT);
$self->{httpd} = $httpd;
- $self->{rbuf} = '';
+ $self->{wbuf} = $wbuf if $wbuf;
($self->{remote_addr}, $self->{remote_port}) =
PublicInbox::Daemon::host_with_port($addr);
- $self->watch_read(1);
$self;
}
-sub event_read { # called by Danga::Socket
+sub event_step { # called by PublicInbox::DS
my ($self) = @_;
- return event_read_input($self) if defined $self->{env};
+ return unless $self->flush_write && $self->{sock};
- my $off = length($self->{rbuf});
- my $r = sysread($self->{sock}, $self->{rbuf}, 8192, $off);
- if (defined $r) {
- return $self->close if $r == 0;
- return rbuf_process($self);
- }
- return if $!{EAGAIN}; # no need to call watch_read(1) again
+ # only read more requests if we've drained the write buffer,
+ # otherwise we can be buffering infinitely w/o backpressure
- # common for clients to break connections without warning,
- # would be too noisy to log here:
- return $self->close;
+ return read_input($self) if defined $self->{env};
+ my $rbuf = $self->{rbuf} // (\(my $x = ''));
+ $self->do_read($rbuf, 8192, bytes::length($$rbuf)) or return;
+ rbuf_process($self, $rbuf);
}
sub rbuf_process {
- my ($self) = @_;
+ my ($self, $rbuf) = @_;
+ $rbuf //= $self->{rbuf} // (\(my $x = ''));
my %env = %{$self->{httpd}->{env}}; # full hash copy
- my $r = parse_http_request($self->{rbuf}, \%env);
+ my $r = parse_http_request($$rbuf, \%env);
# We do not support Trailers in chunked requests, for now
# (they are rarely-used and git (as of 2.7.2) does not use them)
if ($r == -1 || $env{HTTP_TRAILER} ||
# this length-check is necessary for PURE_PERL=1:
- ($r == -2 && length($self->{rbuf}) > 0x4000)) {
+ ($r == -2 && bytes::length($$rbuf) > 0x4000)) {
return quit($self, 400);
}
- return $self->watch_read(1) if $r < 0; # incomplete
- $self->{rbuf} = substr($self->{rbuf}, $r);
-
+ if ($r < 0) { # incomplete
+ $self->rbuf_idle($rbuf);
+ return $self->requeue;
+ }
+ $$rbuf = substr($$rbuf, $r);
my $len = input_prepare($self, \%env);
- defined $len or return write_err($self); # EMFILE/ENFILE
+ defined $len or return write_err($self, undef); # EMFILE/ENFILE
- $len ? event_read_input($self) : app_dispatch($self);
+ $len ? read_input($self, $rbuf) : app_dispatch($self, undef, $rbuf);
}
-sub event_read_input ($) {
- my ($self) = @_;
+# IO::Handle::write returns boolean, this returns bytes written:
+sub xwrite ($$$) {
+ my ($fh, $rbuf, $max) = @_;
+ my $w = bytes::length($$rbuf);
+ $w = $max if $w > $max;
+ $fh->write($$rbuf, $w) or return;
+ $w;
+}
+
+sub read_input ($;$) {
+ my ($self, $rbuf) = @_;
+ $rbuf //= $self->{rbuf} // (\(my $x = ''));
my $env = $self->{env};
- return event_read_input_chunked($self) if env_chunked($env);
+ return if $env->{REMOTE_ADDR}; # in app dispatch
+ return read_input_chunked($self, $rbuf) if env_chunked($env);
# env->{CONTENT_LENGTH} (identity)
- my $sock = $self->{sock};
- my $len = $self->{input_left};
- $self->{input_left} = undef;
- my $rbuf = \($self->{rbuf});
+ my $len = delete $self->{input_left};
my $input = $env->{'psgi.input'};
while ($len > 0) {
if ($$rbuf ne '') {
- my $w = write_in_full($input, $rbuf, $len);
- return write_err($self) unless $w;
+ my $w = xwrite($input, $rbuf, $len);
+ return write_err($self, $len) unless $w;
$len -= $w;
die "BUG: $len < 0 (w=$w)" if $len < 0;
if ($len == 0) { # next request may be pipelined
}
$$rbuf = '';
}
- my $r = sysread($sock, $$rbuf, 8192);
- return recv_err($self, $r, $len) unless $r;
+ $self->do_read($rbuf, 8192) or return recv_err($self, $len);
# continue looping if $r > 0;
}
- app_dispatch($self, $input);
+ app_dispatch($self, $input, $rbuf);
}
sub app_dispatch {
- my ($self, $input) = @_;
- $self->watch_read(0);
+ my ($self, $input, $rbuf) = @_;
+ $self->rbuf_idle($rbuf);
my $env = $self->{env};
$env->{REMOTE_ADDR} = $self->{remote_addr};
$env->{REMOTE_PORT} = $self->{remote_port};
if (my $host = $env->{HTTP_HOST}) {
- $host =~ s/:(\d+)\z// and $env->{SERVER_PORT} = $1;
+ $host =~ s/:([0-9]+)\z// and $env->{SERVER_PORT} = $1;
$env->{SERVER_NAME} = $host;
}
if (defined $input) {
sysseek($input, 0, SEEK_SET) or
die "BUG: psgi.input seek failed: $!";
}
- # note: NOT $self->{sock}, we want our close (+ Danga::Socket::close),
+ # note: NOT $self->{sock}, we want our close (+ PublicInbox::DS::close),
# to do proper cleanup:
- $env->{'psgix.io'} = $self; # only for ->close
+ $env->{'psgix.io'} = $self; # for ->close or async_pass
my $res = Plack::Util::run_app($self->{httpd}->{app}, $env);
eval {
if (ref($res) eq 'CODE') {
my $conn = $env->{HTTP_CONNECTION} || '';
my $term = defined($len) || $chunked;
- my $alive = $term &&
- (($proto eq 'HTTP/1.1' && $conn !~ /\bclose\b/i) ||
- ($conn =~ /\bkeep-alive\b/i));
-
- $h .= 'Connection: ' . ($alive ? 'keep-alive' : 'close');
- $h .= "\r\nDate: " . http_date() . "\r\n\r\n";
+ my $prot_persist = ($proto eq 'HTTP/1.1') && ($conn !~ /\bclose\b/i);
+ my $alive;
+ if (!$term && $prot_persist) { # auto-chunk
+ $chunked = $alive = 2;
+ $h .= "Transfer-Encoding: chunked\r\n";
+ # no need for "Connection: keep-alive" with HTTP/1.1
+ } elsif ($term && ($prot_persist || ($conn =~ /\bkeep-alive\b/i))) {
+ $alive = 1;
+ $h .= "Connection: keep-alive\r\n";
+ } else {
+ $alive = 0;
+ $h .= "Connection: close\r\n";
+ }
+ $h .= 'Date: ' . http_date() . "\r\n\r\n";
if (($len || $chunked) && $env->{REQUEST_METHOD} ne 'HEAD') {
- more($self, $h);
+ msg_more($self, $h);
} else {
- $self->write($h);
+ $self->write(\$h);
}
$alive;
}
-sub response_write {
- my ($self, $env, $res) = @_;
- my $alive = response_header_write($self, $env, $res);
+# middlewares such as Deflater may write empty strings
+sub chunked_write ($$) {
+ my $self = $_[0];
+ return if $_[1] eq '';
+ msg_more($self, sprintf("%x\r\n", bytes::length($_[1])));
+ msg_more($self, $_[1]);
+
+ # use $self->write(\"\n\n") if you care about real-time
+ # streaming responses, public-inbox WWW does not.
+ msg_more($self, "\r\n");
+}
- # middlewares such as Deflater may write empty strings
- my $write = sub { $self->write($_[0]) if $_[0] ne '' };
- my $close = sub {
- if ($alive) {
- $self->event_write; # watch for readability if done
+sub identity_write ($$) {
+ my $self = $_[0];
+ $self->write(\($_[1])) if $_[1] ne '';
+}
+
+sub next_request ($) {
+ my ($self) = @_;
+ if ($self->{rbuf}) {
+ # avoid recursion for pipelined requests
+ PublicInbox::DS::requeue(\&process_pipelineq) if !@$pipelineq;
+ push @$pipelineq, $self;
+ } else { # wait for next request
+ $self->requeue;
+ }
+}
+
+sub response_done {
+ my ($self, $alive) = @_;
+ delete $self->{env}; # we're no longer busy
+ $self->write(\"0\r\n\r\n") if $alive == 2;
+ $self->write($alive ? \&next_request : \&close);
+}
+
+sub getline_pull {
+ my ($self) = @_;
+ my $forward = $self->{forward};
+
+ # limit our own running time for fairness with other
+ # clients and to avoid buffering too much:
+ my $buf = eval {
+ local $/ = \8192;
+ $forward->getline;
+ } if $forward;
+
+ if (defined $buf) {
+ # may close in PublicInbox::DS::write
+ if ($self->{alive} == 2) {
+ chunked_write($self, $buf);
} else {
- Danga::Socket::write($self, sub { $self->close });
+ identity_write($self, $buf);
}
- if (my $obj = $env->{'pi-httpd.inbox'}) {
- # grace period for reaping resources
- $WEAKEN->{"$obj"} = $obj;
- $WEAKTIMER ||= Danga::Socket->AddTimer(60, *weaken_task);
+
+ if ($self->{sock}) {
+ my $wbuf = $self->{wbuf} //= [];
+ push @$wbuf, \&getline_pull;
+
+ # wbuf may be populated by {chunked,identity}_write()
+ # above, no need to rearm if so:
+ $self->requeue if scalar(@$wbuf) == 1;
+ return; # likely
}
- $self->{env} = undef;
- };
+ } elsif ($@) {
+ err($self, "response ->getline error: $@");
+ $self->close;
+ }
+ # avoid recursion
+ if (delete $self->{forward}) {
+ eval { $forward->close };
+ if ($@) {
+ err($self, "response ->close error: $@");
+ $self->close; # idempotent
+ }
+ }
+ response_done($self, delete $self->{alive});
+}
+sub response_write {
+ my ($self, $env, $res) = @_;
+ my $alive = response_header_write($self, $env, $res);
if (defined(my $body = $res->[2])) {
if (ref $body eq 'ARRAY') {
- $write->($_) foreach @$body;
- $close->();
+ if ($alive == 2) {
+ chunked_write($self, $_) for @$body;
+ } else {
+ identity_write($self, $_) for @$body;
+ }
+ response_done($self, $alive);
} else {
- my $pull;
- $pull = sub {
- local $/ = \8192;
- while (defined(my $buf = $body->getline)) {
- $write->($buf);
- if ($self->{write_buf}) {
- $self->write($pull);
- return;
- }
- }
- $pull = undef;
- $body->close();
- $close->();
- };
- $pull->();
+ $self->{forward} = $body;
+ $self->{alive} = $alive;
+ getline_pull($self); # kick-off!
}
+ # these are returned to the calling application:
+ } elsif ($alive == 2) {
+ bless [ $self, $alive ], 'PublicInbox::HTTP::Chunked';
} else {
- # this is returned to the calling application:
- Plack::Util::inline_object(write => $write, close => $close);
- }
-}
-
-use constant MSG_MORE => ($^O eq 'linux') ? 0x8000 : 0;
-sub more ($$) {
- my $self = $_[0];
- if (MSG_MORE && !$self->{write_buf_size}) {
- my $n = send($self->{sock}, $_[1], MSG_MORE);
- if (defined $n) {
- my $dlen = length($_[1]);
- return 1 if $n == $dlen; # all done!
- $_[1] = substr($_[1], $n, $dlen - $n);
- # fall through to normal write:
- }
+ bless [ $self, $alive ], 'PublicInbox::HTTP::Identity';
}
- $self->write($_[1]);
-}
-
-my $pipelineq = [];
-my $next_tick;
-sub process_pipelineq () {
- $next_tick = undef;
- my $q = $pipelineq;
- $pipelineq = [];
- rbuf_process($_) foreach @$q;
}
-# overrides existing Danga::Socket method
-sub event_write {
- my ($self) = @_;
- # only continue watching for readability when we are done writing:
- return if $self->write(undef) != 1;
-
- if ($self->{rbuf} eq '') { # wait for next request
- $self->watch_read(1);
- } else { # avoid recursion for pipelined requests
- push @$pipelineq, $self;
- $next_tick ||= Danga::Socket->AddTimer(0, *process_pipelineq);
- }
+sub input_tmpfile ($) {
+ my $input = tmpfile('http.input', $_[0]->{sock}) or return;
+ $input->autoflush(1);
+ $input;
}
sub input_prepare {
my ($self, $env) = @_;
- my $input = $null_io;
+ my $input;
my $len = $env->{CONTENT_LENGTH};
if ($len) {
if ($len > $MAX_REQUEST_BUFFER) {
quit($self, 413);
return;
}
- $input = IO::File->new_tmpfile;
+ $input = input_tmpfile($self);
} elsif (env_chunked($env)) {
$len = CHUNK_START;
- $input = IO::File->new_tmpfile;
+ $input = input_tmpfile($self);
+ } else {
+ $input = $null_io;
}
# TODO: expire idle clients on ENFILE / EMFILE
sub env_chunked { ($_[0]->{HTTP_TRANSFER_ENCODING} || '') =~ /\bchunked\b/i }
+sub err ($$) {
+ eval { $_[0]->{httpd}->{env}->{'psgi.errors'}->print($_[1]."\n") };
+}
+
sub write_err {
- my ($self) = @_;
- my $err = $self->{httpd}->{env}->{'psgi.errors'};
+ my ($self, $len) = @_;
my $msg = $! || '(zero write)';
- $err->print("error buffering to input: $msg\n");
+ $msg .= " ($len bytes remaining)" if defined $len;
+ err($self, "error buffering to input: $msg");
quit($self, 500);
}
sub recv_err {
- my ($self, $r, $len) = @_;
- return $self->close if (defined $r && $r == 0);
- if ($!{EAGAIN}) {
+ my ($self, $len) = @_;
+ if ($! == EAGAIN) { # epoll/kevent watch already set by do_read
$self->{input_left} = $len;
- return;
- }
- my $err = $self->{httpd}->{env}->{'psgi.errors'};
- $err->print("error reading for input: $! ($len bytes remaining)\n");
- quit($self, 500);
-}
-
-sub write_in_full {
- my ($fh, $rbuf, $len) = @_;
- my $rv = 0;
- my $off = 0;
- while ($len > 0) {
- my $w = syswrite($fh, $$rbuf, $len, $off);
- return ($rv ? $rv : $w) unless $w; # undef or 0
- $rv += $w;
- $off += $w;
- $len -= $w;
+ } else {
+ err($self, "error reading input: $! ($len bytes remaining)");
}
- $rv
}
-sub event_read_input_chunked { # unlikely...
- my ($self) = @_;
+sub read_input_chunked { # unlikely...
+ my ($self, $rbuf) = @_;
+ $rbuf //= $self->{rbuf} // (\(my $x = ''));
my $input = $self->{env}->{'psgi.input'};
- my $sock = $self->{sock};
- my $len = $self->{input_left};
- $self->{input_left} = undef;
- my $rbuf = \($self->{rbuf});
+ my $len = delete $self->{input_left};
while (1) { # chunk start
if ($len == CHUNK_ZEND) {
$$rbuf =~ s/\A\r\n//s and
- return app_dispatch($self, $input);
- return quit($self, 400) if length($$rbuf) > 2;
+ return app_dispatch($self, $input, $rbuf);
+
+ return quit($self, 400) if bytes::length($$rbuf) > 2;
}
if ($len == CHUNK_END) {
if ($$rbuf =~ s/\A\r\n//s) {
$len = CHUNK_START;
- } elsif (length($$rbuf) > 2) {
+ } elsif (bytes::length($$rbuf) > 2) {
return quit($self, 400);
}
}
if (($len + -s $input) > $MAX_REQUEST_BUFFER) {
return quit($self, 413);
}
- } elsif (length($$rbuf) > CHUNK_MAX_HDR) {
+ } elsif (bytes::length($$rbuf) > CHUNK_MAX_HDR) {
return quit($self, 400);
}
# will break from loop since $len >= 0
}
if ($len < 0) { # chunk header is trickled, read more
- my $off = length($$rbuf);
- my $r = sysread($sock, $$rbuf, 8192, $off);
- return recv_err($self, $r, $len) unless $r;
+ $self->do_read($rbuf, 8192, bytes::length($$rbuf)) or
+ return recv_err($self, $len);
# (implicit) goto chunk_start if $r > 0;
}
$len = CHUNK_ZEND if $len == 0;
# drain the current chunk
until ($len <= 0) {
if ($$rbuf ne '') {
- my $w = write_in_full($input, $rbuf, $len);
- return write_err($self) unless $w;
+ my $w = xwrite($input, $rbuf, $len);
+ return write_err($self, "$len chunk") if !$w;
$len -= $w;
if ($len == 0) {
# we may have leftover data to parse
}
if ($$rbuf eq '') {
# read more of current chunk
- my $r = sysread($sock, $$rbuf, 8192);
- return recv_err($self, $r, $len) unless $r;
+ $self->do_read($rbuf, 8192) or
+ return recv_err($self, $len);
}
}
}
sub quit {
my ($self, $status) = @_;
my $h = "HTTP/1.1 $status " . status_message($status) . "\r\n\r\n";
- $self->write($h);
+ $self->write(\$h);
$self->close;
}
-# callbacks for Danga::Socket
-
-sub event_hup { $_[0]->close }
-sub event_err { $_[0]->close }
-
sub close {
- my $self = shift;
- $self->{env} = undef;
- $self->SUPER::close(@_);
+ my $self = $_[0];
+ delete $self->{env}; # prevent circular references
+ if (my $forward = delete $self->{forward}) {
+ eval { $forward->close };
+ err($self, "forward ->close error: $@") if $@;
+ }
+ $self->SUPER::close; # PublicInbox::DS::close
}
# for graceful shutdown in PublicInbox::Daemon:
sub busy () {
my ($self) = @_;
- ($self->{rbuf} ne '' || $self->{env} || $self->{write_buf_size});
+ ($self->{rbuf} || $self->{env} || $self->{wbuf});
+}
+
+# Chunked and Identity packages are used for writing responses.
+# They may be exposed to the PSGI application when the PSGI app
+# returns a CODE ref for "push"-based responses
+package PublicInbox::HTTP::Chunked;
+use strict;
+
+sub write {
+ # ([$http], $buf) = @_;
+ PublicInbox::HTTP::chunked_write($_[0]->[0], $_[1])
+}
+
+sub close {
+ # $_[0] = [$http, $alive]
+ PublicInbox::HTTP::response_done(@{$_[0]});
+}
+
+package PublicInbox::HTTP::Identity;
+use strict;
+our @ISA = qw(PublicInbox::HTTP::Chunked);
+
+sub write {
+ # ([$http], $buf) = @_;
+ PublicInbox::HTTP::identity_write($_[0]->[0], $_[1]);
}
1;