use strict;
use warnings;
use Fcntl qw(:seek);
-use POSIX qw(dup2);
+use IO::Handle;
+use HTTP::Date qw(time2str);
+use HTTP::Status qw(status_message);
+use Plack::Util;
+use PublicInbox::Qspawn;
+
+# 32 is same as the git-daemon connection limit
+my $default_limiter = PublicInbox::Qspawn::Limiter->new(32);
# n.b. serving "description" and "cloneurl" should be innocuous enough to
# not cause problems. serving "config" might...
objects/pack/pack-[a-f0-9]{40}\.(?:pack|idx)
!;
-our $ANY = join('|', @binary, @text);
+our $ANY = join('|', @binary, @text, 'git-upload-pack');
my $BIN = join('|', @binary);
my $TEXT = join('|', @text);
-sub r {
- [ $_[0] , [qw(Content-Type text/plain Content-Length 0) ], [] ]
+my @no_cache = ('Expires', 'Fri, 01 Jan 1980 00:00:00 GMT',
+ 'Pragma', 'no-cache',
+ 'Cache-Control', 'no-cache, max-age=0, must-revalidate');
+
+sub r ($;$) {
+ my ($code, $msg) = @_;
+ $msg ||= status_message($code);
+ my $len = length($msg);
+ [ $code, [qw(Content-Type text/plain Content-Length), $len, @no_cache],
+ [$msg] ]
}
sub serve {
- my ($cgi, $git, $path) = @_;
- my $service = $cgi->param('service') || '';
- if ($service =~ /\Agit-\w+-pack\z/ || $path =~ /\Agit-\w+-pack\z/) {
- my $ok = serve_smart($cgi, $git, $path);
+ my ($env, $git, $path) = @_;
+
+ # XXX compatibility... ugh, can we stop supporting this?
+ $git = PublicInbox::Git->new($git) unless ref($git);
+
+ # Documentation/technical/http-protocol.txt in git.git
+ # requires one and exactly one query parameter:
+ if ($env->{QUERY_STRING} =~ /\Aservice=git-\w+-pack\z/ ||
+ $path =~ /\Agit-\w+-pack\z/) {
+ my $ok = serve_smart($env, $git, $path);
return $ok if $ok;
}
+ serve_dumb($env, $git, $path);
+}
+
+sub err ($@) {
+ my ($env, @msg) = @_;
+ $env->{'psgi.errors'}->print(@msg, "\n");
+}
+
+sub drop_client ($) {
+ if (my $io = $_[0]->{'psgix.io'}) {
+ $io->close; # this is Danga::Socket::close
+ }
+}
+
+my $prev = 0;
+my $exp;
+sub cache_one_year {
+ my ($h) = @_;
+ my $t = time + 31536000;
+ push @$h, 'Expires', $t == $prev ? $exp : ($exp = time2str($prev = $t)),
+ 'Cache-Control', 'public, max-age=31536000';
+}
+
+sub serve_dumb {
+ my ($env, $git, $path) = @_;
+
+ my @h;
my $type;
- if ($path =~ /\A(?:$BIN)\z/o) {
- $type = 'application/octet-stream';
+ if ($path =~ m!\Aobjects/[a-f0-9]{2}/[a-f0-9]{38}\z!) {
+ $type = 'application/x-git-loose-object';
+ cache_one_year(\@h);
+ } elsif ($path =~ m!\Aobjects/pack/pack-[a-f0-9]{40}\.pack\z!) {
+ $type = 'application/x-git-packed-objects';
+ cache_one_year(\@h);
+ } elsif ($path =~ m!\Aobjects/pack/pack-[a-f0-9]{40}\.idx\z!) {
+ $type = 'application/x-git-packed-objects-toc';
+ cache_one_year(\@h);
} elsif ($path =~ /\A(?:$TEXT)\z/o) {
$type = 'text/plain';
+ push @h, @no_cache;
} else {
return r(404);
}
- my $f = "$git->{git_dir}/$path";
- return r(404) unless -f $f && -r _;
- my @st = stat(_);
- my $size = $st[7];
- # TODO: If-Modified-Since and Last-Modified
+ my $f = $git->{git_dir} . '/' . $path;
+ return r(404) unless -f $f && -r _; # just in case it's a FIFO :P
+ my $size = -s _;
+
+ # TODO: If-Modified-Since and Last-Modified?
open my $in, '<', $f or return r(404);
- my $code = 200;
my $len = $size;
- my @h;
-
- my $env = $cgi->{env};
- my $range = $env->{HTTP_RANGE};
- if (defined $range && $range =~ /\bbytes=(\d*)-(\d*)\z/) {
- ($code, $len) = prepare_range($cgi, $in, \@h, $1, $2, $size);
+ my $code = 200;
+ push @h, 'Content-Type', $type;
+ if (($env->{HTTP_RANGE} || '') =~ /\bbytes=(\d*)-(\d*)\z/) {
+ ($code, $len) = prepare_range($env, $in, \@h, $1, $2, $size);
if ($code == 416) {
push @h, 'Content-Range', "bytes */$size";
return [ 416, \@h, [] ];
}
}
-
- push @h, 'Content-Type', $type, 'Content-Length', $len;
- sub {
- my ($res) = @_; # Plack callback
- my $fh = $res->([ $code, \@h ]);
- my $buf;
- my $n = 8192;
- while ($len > 0) {
+ push @h, 'Content-Length', $len;
+ my $n = 65536;
+ [ $code, \@h, Plack::Util::inline_object(close => sub { close $in },
+ getline => sub {
+ return if $len == 0;
$n = $len if $len < $n;
- my $r = read($in, $buf, $n);
- last if (!defined($r) || $r <= 0);
- $len -= $r;
- $fh->write($buf);
- }
- $fh->close;
- }
+ my $r = sysread($in, my $buf, $n);
+ if (!defined $r) {
+ err($env, "$f read error: $!");
+ } elsif ($r <= 0) {
+ err($env, "$f EOF with $len bytes left");
+ } else {
+ $len -= $r;
+ $n = 8192;
+ return $buf;
+ }
+ drop_client($env);
+ return;
+ })]
}
sub prepare_range {
- my ($cgi, $in, $h, $beg, $end, $size) = @_;
+ my ($env, $in, $h, $beg, $end, $size) = @_;
my $code = 200;
my $len = $size;
if ($beg eq '') {
if ($len <= 0) {
$code = 416;
} else {
- seek($in, $beg, SEEK_SET) or return [ 500, [], [] ];
+ sysseek($in, $beg, SEEK_SET) or return [ 500, [], [] ];
push @$h, qw(Accept-Ranges bytes Content-Range);
push @$h, "bytes $beg-$end/$size";
# FIXME: Plack::Middleware::Deflater bug?
- $cgi->{env}->{'psgix.no-compress'} = 1;
+ $env->{'psgix.no-compress'} = 1;
}
}
($code, $len);
# returns undef if 403 so it falls back to dumb HTTP
sub serve_smart {
- my ($cgi, $git, $path) = @_;
- my $env = $cgi->{env};
-
- my $input = $env->{'psgi.input'};
- my $buf;
- my $in;
- my $err = $env->{'psgi.errors'};
- if (fileno($input) >= 0) {
- $in = $input;
- } else { # FIXME untested
- $in = IO::File->new_tmpfile;
- while (1) {
- my $r = $input->read($buf, 8192);
- unless (defined $r) {
- $err->print('error reading input: ', $!, "\n");
- return r(500);
- }
- last if ($r == 0);
- $in->write($buf);
- }
- $in->flush;
- $in->sysseek(0, SEEK_SET);
+ my ($env, $git, $path) = @_;
+ my $in = $env->{'psgi.input'};
+ my $fd = eval { fileno($in) };
+ unless (defined $fd && $fd >= 0) {
+ $in = input_to_file($env) or return r(500);
}
- my ($rpipe, $wpipe);
- unless (pipe($rpipe, $wpipe)) {
- $err->print('error creating pipe', $!, "\n");
- return r(500);
+ my %env = %ENV;
+ # GIT_COMMITTER_NAME, GIT_COMMITTER_EMAIL
+ # may be set in the server-process and are passed as-is
+ foreach my $name (qw(QUERY_STRING
+ REMOTE_USER REMOTE_ADDR
+ HTTP_CONTENT_ENCODING
+ CONTENT_TYPE
+ SERVER_PROTOCOL
+ REQUEST_METHOD)) {
+ my $val = $env->{$name};
+ $env{$name} = $val if defined $val;
}
- my $pid = fork; # TODO: vfork under Linux...
- unless (defined $pid) {
- $err->print('error forking: ', $!, "\n");
- return r(500);
- }
- if ($pid == 0) {
- # GIT_HTTP_EXPORT_ALL, GIT_COMMITTER_NAME, GIT_COMMITTER_EMAIL
- # may be set in the server-process and are passed as-is
- foreach my $name (qw(QUERY_STRING
- REMOTE_USER REMOTE_ADDR
- HTTP_CONTENT_ENCODING
- CONTENT_TYPE
- SERVER_PROTOCOL
- REQUEST_METHOD)) {
- my $val = $env->{$name};
- $ENV{$name} = $val if defined $val;
- }
- # $ENV{GIT_PROJECT_ROOT} = $git->{git_dir};
- $ENV{GIT_HTTP_EXPORT_ALL} = '1';
- $ENV{PATH_TRANSLATED} = "$git->{git_dir}/$path";
- dup2(fileno($in), 0) or die "redirect stdin failed: $!\n";
- dup2(fileno($wpipe), 1) or die "redirect stdout failed: $!\n";
- my @cmd = qw(git http-backend);
- exec(@cmd) or die 'exec `' . join(' ', @cmd). "' failed: $!\n";
- }
- $wpipe = $in = undef;
- $rpipe->blocking(0);
- $buf = '';
- my $vin;
- vec($vin, fileno($rpipe), 1) = 1;
- my ($fh, $res);
- my $fail = sub {
- my ($e) = @_;
- if ($e eq 'EAGAIN') {
- select($vin, undef, undef, undef);
- } else {
- $rpipe = undef;
- $fh->close if $fh;
- $err->print('git http-backend error: ', $e, "\n");
+ my $limiter = $git->{-httpbackend_limiter} || $default_limiter;
+ my $git_dir = $git->{git_dir};
+ $env{GIT_HTTP_EXPORT_ALL} = '1';
+ $env{PATH_TRANSLATED} = "$git_dir/$path";
+ my $rdr = { 0 => fileno($in) };
+ my $qsp = PublicInbox::Qspawn->new([qw(git http-backend)], \%env, $rdr);
+ my ($fh, $rpipe);
+ my $end = sub {
+ if (my $err = $qsp->finish) {
+ err($env, "git http-backend ($git_dir): $err");
}
+ $fh->close if $fh; # async-only
};
+
+ # Danga::Socket users, we queue up the read_enable callback to
+ # fire after pending writes are complete:
+ my $buf = '';
+ my $rd_hdr = sub {
+ my $r = sysread($rpipe, $buf, 1024, length($buf));
+ return if !defined($r) && ($!{EINTR} || $!{EAGAIN});
+ return r(500, 'http-backend error') unless $r;
+ $r = parse_cgi_headers(\$buf) or return; # incomplete headers
+ $r->[0] == 403 ? serve_dumb($env, $git, $path) : $r;
+ };
+ my $res;
+ my $async = $env->{'pi-httpd.async'}; # XXX unstable API
my $cb = sub {
- my $r = sysread($rpipe, $buf, 8192, length($buf));
- return $fail->($!{EAGAIN} ? 'EAGAIN' : $!) unless defined $r;
- if ($r == 0) { # EOF
- $rpipe = undef;
- $fh->close if $fh;
- return;
- }
- if ($fh) { # stream body from git-http-backend to HTTP client
- $fh->write($buf);
- $buf = '';
- } elsif ($buf =~ s/\A(.*?)\r?\n\r?\n//s) { # parse headers
- my $h = $1;
- my $code = 200;
- my @h;
- foreach my $l (split(/\r?\n/, $h)) {
- my ($k, $v) = split(/:\s*/, $l, 2);
- if ($k =~ /\AStatus\z/i) {
- $code = int($v);
- } else {
- push @h, $k, $v;
- }
+ my $r = $rd_hdr->() or return;
+ $rd_hdr = undef;
+ if (scalar(@$r) == 3) { # error:
+ if ($async) {
+ $async->close; # calls rpipe->close
+ } else {
+ $rpipe->close;
+ $end->();
}
- # write response header:
- $fh = $res->([ $code, \@h ]);
- $fh->write($buf);
- $buf = '';
- } # else { keep reading ... }
+ $res->($r);
+ } elsif ($async) {
+ $fh = $res->($r);
+ $async->async_pass($env->{'psgix.io'}, $fh, \$buf);
+ } else { # for synchronous PSGI servers
+ require PublicInbox::GetlineBody;
+ $r->[2] = PublicInbox::GetlineBody->new($rpipe, $end,
+ $buf);
+ $res->($r);
+ }
};
sub {
($res) = @_;
- while ($rpipe) { $cb->() }
+
+ # hopefully this doesn't break any middlewares,
+ # holding the input here is a waste of FDs and memory
+ $env->{'psgi.input'} = undef;
+
+ $qsp->start($limiter, sub { # may run later, much later...
+ ($rpipe) = @_;
+ $in = undef;
+ if ($async) {
+ $async = $async->($rpipe, $cb, $end);
+ } else { # generic PSGI
+ $cb->() while $rd_hdr;
+ }
+ });
};
}
+sub input_to_file {
+ my ($env) = @_;
+ open(my $in, '+>', undef);
+ unless (defined $in) {
+ err($env, "could not open temporary file: $!");
+ return;
+ }
+ my $input = $env->{'psgi.input'};
+ my $buf;
+ while (1) {
+ my $r = $input->read($buf, 8192);
+ unless (defined $r) {
+ err($env, "error reading input: $!");
+ return;
+ }
+ my $off = 0;
+ while ($r > 0) {
+ my $w = syswrite($in, $buf, $r, $off);
+ if (defined $w) {
+ $r -= $w;
+ $off += $w;
+ } else {
+ err($env, "error writing temporary file: $!");
+ return;
+ }
+ }
+ }
+ unless (defined(sysseek($in, 0, SEEK_SET))) {
+ err($env, "error seeking temporary file: $!");
+ return;
+ }
+ return $in;
+}
+
+sub parse_cgi_headers {
+ my ($bref) = @_;
+ $$bref =~ s/\A(.*?)\r\n\r\n//s or return;
+ my $h = $1;
+ my $code = 200;
+ my @h;
+ foreach my $l (split(/\r\n/, $h)) {
+ my ($k, $v) = split(/:\s*/, $l, 2);
+ if ($k =~ /\AStatus\z/i) {
+ ($code) = ($v =~ /\b(\d+)\b/);
+ } else {
+ push @h, $k, $v;
+ }
+ }
+ [ $code, \@h ]
+}
+
1;