-# Copyright (C) 2016-2018 all contributors <meta@public-inbox.org>
+# Copyright (C) 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
# to learn different ways to admin both NNTP and HTTP components.
# There's nothing which depends on public-inbox, here.
# Each instance of this class represents a HTTP client socket
-
+#
+# fields:
+# httpd: PublicInbox::HTTPD ref
+# env: PSGI env hashref
+# input_left: bytes left to read in request body (e.g. POST/PUT)
+# remote_addr: remote IP address as a string (e.g. "127.0.0.1")
+# remote_port: peer port
+# forward: response body object, response to ->getline + ->close
+# alive: HTTP keepalive state:
+# 0: drop connection when done
+# 1: keep connection when done
+# 2: keep connection, chunk responses
package PublicInbox::HTTP;
use strict;
-use warnings;
-use base qw(PublicInbox::DS);
-use fields qw(httpd env rbuf input_left remote_addr remote_port forward pull);
-use bytes (); # only for bytes::length
+use parent qw(PublicInbox::DS);
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::Handle;
-require PublicInbox::EvCleanup;
-PublicInbox::DS->import(qw(msg_more write_in_full));
+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
};
use Errno qw(EAGAIN);
-my $pipelineq = [];
-my $pipet;
-sub process_pipelineq () {
- my $q = $pipelineq;
- $pipet = undef;
- $pipelineq = [];
- foreach (@$q) {
- next unless $_->{sock};
- rbuf_process($_);
- }
-}
-
# Use the same configuration parameter as git since this is primarily
# a slow-client sponge for git-http-backend
# TODO: support per-respository http.maxRequestBuffer somehow...
}
sub new ($$$) {
- my ($class, $sock, $addr, $httpd) = @_;
- my $self = fields::new($class);
- $self->SUPER::new($sock, EPOLLIN | EPOLLONESHOT);
- $self->{httpd} = $httpd;
- $self->{rbuf} = '';
+ my ($class, $sock, $addr, $srv_env) = @_;
+ my $self = bless { srv_env => $srv_env }, $class;
+ my $ev = EPOLLIN;
+ my $wbuf;
+ if ($sock->can('accept_SSL') && !$sock->accept_SSL) {
+ return CORE::close($sock) if $! != EAGAIN;
+ $ev = PublicInbox::TLS::epollbit() or return CORE::close($sock);
+ $wbuf = [ \&PublicInbox::DS::accept_tls_step ];
+ }
+ $self->{wbuf} = $wbuf if $wbuf;
($self->{remote_addr}, $self->{remote_port}) =
PublicInbox::Daemon::host_with_port($addr);
- $self;
+ $self->SUPER::new($sock, $ev | EPOLLONESHOT);
}
sub event_step { # called by PublicInbox::DS
my ($self) = @_;
-
+ local $SIG{__WARN__} = $self->{srv_env}->{'pi-httpd.warn_cb'};
return unless $self->flush_write && $self->{sock};
# only read more requests if we've drained the write buffer,
# otherwise we can be buffering infinitely w/o backpressure
- return read_input($self) if defined $self->{env};
-
- 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);
- }
-
- # common for clients to break connections without warning,
- # would be too noisy to log here:
- $! == EAGAIN ? $self->watch_in1 : $self->close;
-}
-
-sub rbuf_process {
- my ($self) = @_;
-
- my %env = %{$self->{httpd}->{env}}; # full hash copy
- my $r = parse_http_request($self->{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($self->{rbuf}) > 0x4000)) {
- return quit($self, 400);
+ return read_input($self) if ref($self->{env});
+
+ my $rbuf = $self->{rbuf} // (\(my $x = ''));
+ my %env = %{$self->{srv_env}}; # full hash copy
+ my $r;
+ while (($r = parse_http_request($$rbuf, \%env)) < 0) {
+ # We do not support Trailers in chunked requests, for
+ # now (they are rarely-used and git (as of 2.7.2) does
+ # not use them).
+ # this length-check is necessary for PURE_PERL=1:
+ if ($r == -1 || $env{HTTP_TRAILER} ||
+ ($r == -2 && length($$rbuf) > 0x4000)) {
+ return quit($self, 400);
+ }
+ $self->do_read($rbuf, 8192, length($$rbuf)) or return;
}
- return $self->watch_in1 if $r < 0; # incomplete
- $self->{rbuf} = substr($self->{rbuf}, $r);
-
- my $len = input_prepare($self, \%env);
- defined $len or return write_err($self, undef); # EMFILE/ENFILE
+ return quit($self, 400) if grep(/\s/, keys %env); # stop smugglers
+ $$rbuf = substr($$rbuf, $r);
+ my $len = input_prepare($self, \%env) //
+ return write_err($self, undef); # EMFILE/ENFILE
- $len ? read_input($self) : app_dispatch($self);
+ $len ? read_input($self, $rbuf) : app_dispatch($self, undef, $rbuf);
}
-sub read_input ($) {
- my ($self) = @_;
+sub read_input ($;$) {
+ my ($self, $rbuf) = @_;
+ $rbuf //= $self->{rbuf} // (\(my $x = ''));
my $env = $self->{env};
- return if $env->{REMOTE_ADDR}; # in app dispatch
- return read_input_chunked($self) if env_chunked($env);
+ return read_input_chunked($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 ($$rbuf ne '') {
- my $w = write_in_full($input, $rbuf, $len, 0);
+ my $w = syswrite($input, $$rbuf, $len);
return write_err($self, $len) unless $w;
$len -= $w;
die "BUG: $len < 0 (w=$w)" if $len < 0;
}
$$rbuf = '';
}
- my $r = sysread($sock, $$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, $input) = @_;
+ my ($self, $input, $rbuf) = @_;
+ $self->rbuf_idle($rbuf);
my $env = $self->{env};
+ $self->{env} = undef; # for exists() check in ->busy
$env->{REMOTE_ADDR} = $self->{remote_addr};
$env->{REMOTE_PORT} = $self->{remote_port};
- if (my $host = $env->{HTTP_HOST}) {
- $host =~ s/:([0-9]+)\z// and $env->{SERVER_PORT} = $1;
+ if (defined(my $host = $env->{HTTP_HOST})) {
+ $host =~ s/:([0-9]+)\z// and $env->{SERVER_PORT} = $1 + 0;
$env->{SERVER_NAME} = $host;
}
if (defined $input) {
}
# note: NOT $self->{sock}, we want our close (+ PublicInbox::DS::close),
# to do proper cleanup:
- $env->{'psgix.io'} = $self; # only for ->close
- my $res = Plack::Util::run_app($self->{httpd}->{app}, $env);
+ $env->{'psgix.io'} = $self; # for ->close or async_pass
+ my $res = Plack::Util::run_app($env->{'pi-httpd.app'}, $env);
eval {
if (ref($res) eq 'CODE') {
$res->(sub { response_write($self, $env, $_[0]) });
response_write($self, $env, $res);
}
};
- $self->close if $@;
+ if ($@) {
+ warn "response_write error: $@";
+ $self->close;
+ }
}
sub response_header_write {
my $alive;
if (!$term && $prot_persist) { # auto-chunk
$chunked = $alive = 2;
+ $alive = 3 if $env->{REQUEST_METHOD} eq 'HEAD';
$h .= "Transfer-Encoding: chunked\r\n";
# no need for "Connection: keep-alive" with HTTP/1.1
} elsif ($term && ($prot_persist || ($conn =~ /\bkeep-alive\b/i))) {
}
# middlewares such as Deflater may write empty strings
-sub chunked_wcb ($) {
- my ($self) = @_;
- sub {
- return if $_[0] eq '';
- msg_more($self, sprintf("%x\r\n", bytes::length($_[0])));
- msg_more($self, $_[0]);
-
- # use $self->write(\"\n\n") if you care about real-time
- # streaming responses, public-inbox WWW does not.
- msg_more($self, "\r\n");
- }
+sub chunked_write ($$) {
+ my $self = $_[0];
+ return if $_[1] eq '';
+ msg_more($self, sprintf("%x\r\n", 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");
}
-sub identity_wcb ($) {
- my ($self) = @_;
- sub { $self->write(\($_[0])) if $_[0] ne '' }
-}
-
-sub next_request ($) {
- my ($self) = @_;
- if ($self->{rbuf} eq '') { # wait for next request
- $self->watch_in1;
- } else { # avoid recursion for pipelined requests
- push @$pipelineq, $self;
- $pipet ||= PublicInbox::EvCleanup::asap(*process_pipelineq);
- }
+sub identity_write ($$) {
+ my $self = $_[0];
+ $self->write(\($_[1])) if $_[1] ne '';
}
-sub response_done_cb ($$) {
+sub response_done {
my ($self, $alive) = @_;
- sub {
- my $env = $self->{env};
- $self->{env} = undef;
- $self->write(\"0\r\n\r\n") if $alive == 2;
- $self->write(sub{$alive ? next_request($self) : $self->close});
- }
+ delete $self->{env}; # we're no longer busy
+ # HEAD requests set $alive = 3 so we don't send "0\r\n\r\n";
+ $self->write(\"0\r\n\r\n") if $alive == 2;
+ $self->write($alive ? $self->can('requeue') : \&close);
}
-sub getline_cb ($$$) {
- my ($self, $write, $close) = @_;
- local $/ = \8192;
+sub getline_pull {
+ my ($self) = @_;
my $forward = $self->{forward};
+
# limit our own running time for fairness with other
# clients and to avoid buffering too much:
- if ($forward) {
- my $buf = eval { $forward->getline };
- if (defined $buf) {
- $write->($buf); # may close in PublicInbox::DS::write
- if ($self->{sock}) {
- my $next = $self->{pull};
- if ($self->{wbuf}) {
- $self->write($next);
- } else {
- PublicInbox::EvCleanup::asap($next);
- }
- return;
- }
- } elsif ($@) {
- err($self, "response ->getline error: $@");
- $forward = undef;
- $self->close;
+ my $buf = eval {
+ local $/ = \65536;
+ $forward->getline;
+ } if $forward;
+
+ if (defined $buf) {
+ # may close in PublicInbox::DS::write
+ if ($self->{alive} == 2) {
+ chunked_write($self, $buf);
+ } else {
+ identity_write($self, $buf);
}
- }
- $self->{forward} = $self->{pull} = undef;
+ if ($self->{sock}) {
+ # autovivify wbuf
+ my $new_size = push(@{$self->{wbuf}}, \&getline_pull);
+
+ # wbuf may be populated by {chunked,identity}_write()
+ # above, no need to rearm if so:
+ $self->requeue if $new_size == 1;
+ return; # likely
+ }
+ } elsif ($@) {
+ warn "response ->getline error: $@";
+ $self->close;
+ }
# avoid recursion
- if ($forward) {
+ if (delete $self->{forward}) {
eval { $forward->close };
if ($@) {
- err($self, "response ->close error: $@");
+ warn "response ->close error: $@";
$self->close; # idempotent
}
}
- $close->();
-}
-
-sub getline_response ($$$) {
- my ($self, $write, $close) = @_;
- my $pull = $self->{pull} = sub { getline_cb($self, $write, $close) };
- $pull->();
+ response_done($self, delete $self->{alive});
}
sub response_write {
my ($self, $env, $res) = @_;
my $alive = response_header_write($self, $env, $res);
- my $close = response_done_cb($self, $alive);
- my $write = $alive == 2 ? chunked_wcb($self) : identity_wcb($self);
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 {
$self->{forward} = $body;
- getline_response($self, $write, $close);
+ $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);
+ bless [ $self, $alive ], 'PublicInbox::HTTP::Identity';
}
}
sub input_prepare {
my ($self, $env) = @_;
- my $input;
- my $len = $env->{CONTENT_LENGTH};
- if ($len) {
- if ($len > $MAX_REQUEST_BUFFER) {
- quit($self, 413);
- return;
- }
- open($input, '+>', undef);
- } elsif (env_chunked($env)) {
+ my ($input, $len);
+
+ # rfc 7230 3.3.2, 3.3.3,: favor Transfer-Encoding over Content-Length
+ my $hte = $env->{HTTP_TRANSFER_ENCODING};
+ if (defined $hte) {
+ # rfc7230 3.3.3, point 3 says only chunked is accepted
+ # as the final encoding. Since neither public-inbox-httpd,
+ # git-http-backend, or our WWW-related code uses "gzip",
+ # "deflate" or "compress" as the Transfer-Encoding, we'll
+ # reject them:
+ return quit($self, 400) if $hte !~ /\Achunked\z/i;
+
$len = CHUNK_START;
- open($input, '+>', undef);
+ $input = tmpfile('http.input', $self->{sock});
} else {
- $input = $null_io;
+ $len = $env->{CONTENT_LENGTH};
+ if (defined $len) {
+ # rfc7230 3.3.3.4
+ return quit($self, 400) if $len !~ /\A[0-9]+\z/;
+ return quit($self, 413) if $len > $MAX_REQUEST_BUFFER;
+ $input = $len ? tmpfile('http.input', $self->{sock})
+ : $null_io;
+ } else {
+ $input = $null_io;
+ }
}
# TODO: expire idle clients on ENFILE / EMFILE
- return unless $input;
-
- $env->{'psgi.input'} = $input;
+ $env->{'psgi.input'} = $input // return;
$self->{env} = $env;
$self->{input_left} = $len || 0;
}
-sub env_chunked { ($_[0]->{HTTP_TRANSFER_ENCODING} || '') =~ /\bchunked\b/i }
-
-sub err ($$) {
- eval { $_[0]->{httpd}->{env}->{'psgi.errors'}->print($_[1]."\n") };
-}
+sub env_chunked { ($_[0]->{HTTP_TRANSFER_ENCODING} // '') =~ /\Achunked\z/i }
sub write_err {
my ($self, $len) = @_;
my $msg = $! || '(zero write)';
$msg .= " ($len bytes remaining)" if defined $len;
- err($self, "error buffering to input: $msg");
+ warn "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 $self->watch_in1;
+ } else {
+ warn "error reading input: $! ($len bytes remaining)";
}
- err($self, "error reading for input: $! ($len bytes remaining)");
- quit($self, 500);
}
sub read_input_chunked { # unlikely...
- my ($self) = @_;
+ 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) {
$$rbuf =~ s/\A\r\n//s and
- return app_dispatch($self, $input);
+ return app_dispatch($self, $input, $rbuf);
+
return quit($self, 400) if length($$rbuf) > 2;
}
if ($len == CHUNK_END) {
}
if ($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, 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, 0);
+ my $w = syswrite($input, $$rbuf, $len);
return write_err($self, "$len chunk") if !$w;
$len -= $w;
if ($len == 0) {
}
if ($$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);
}
}
}
my $h = "HTTP/1.1 $status " . status_message($status) . "\r\n\r\n";
$self->write(\$h);
$self->close;
+ undef; # input_prepare expects this
}
sub close {
- my $self = shift;
- my $forward = $self->{forward};
- my $env = $self->{env};
- delete $env->{'psgix.io'} if $env; # prevent circular references
- $self->{pull} = $self->{forward} = $self->{env} = undef;
- if ($forward) {
+ my $self = $_[0];
+ if (my $forward = delete $self->{forward}) {
eval { $forward->close };
- err($self, "forward ->close error: $@") if $@;
+ warn "forward ->close error: $@" if $@;
}
- $self->SUPER::close(@_);
+ $self->SUPER::close; # PublicInbox::DS::close
}
-# for graceful shutdown in PublicInbox::Daemon:
-sub busy () {
+sub busy { # for graceful shutdown in PublicInbox::Daemon:
my ($self) = @_;
- ($self->{rbuf} ne '' || $self->{env} || $self->{wbuf});
+ defined($self->{rbuf}) || exists($self->{env}) || defined($self->{wbuf})
+}
+
+# runs $cb on the next iteration of the event loop at earliest
+sub next_step {
+ my ($self, $cb) = @_;
+ return unless exists $self->{sock};
+ $self->requeue if 1 == push(@{$self->{wbuf}}, $cb);
+}
+
+# Chunked and Identity packages are used for writing responses.
+# They may be exposed to the PSGI application when the PSGI app
+# returns a CODE ref for "push"-based responses
+package PublicInbox::HTTP::Chunked;
+use strict;
+
+sub write {
+ # ([$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;