X-Git-Url: http://www.git.stargrave.org/?a=blobdiff_plain;f=lib%2FPublicInbox%2FGitHTTPBackend.pm;h=b4851920fe04f55b51ede42fa6770b70982cf17e;hb=52052329acedddc0082487283a5a0390a8c86602;hp=50bb4331f5aaaa017b21d846fdbd040d32d0ca79;hpb=310819ea86ac7f85d39a2a8720c4d179c8c94277;p=public-inbox.git diff --git a/lib/PublicInbox/GitHTTPBackend.pm b/lib/PublicInbox/GitHTTPBackend.pm index 50bb4331..b4851920 100644 --- a/lib/PublicInbox/GitHTTPBackend.pm +++ b/lib/PublicInbox/GitHTTPBackend.pm @@ -7,7 +7,10 @@ package PublicInbox::GitHTTPBackend; use strict; use warnings; use Fcntl qw(:seek); -use PublicInbox::Spawn qw(spawn); +use IO::File; +use HTTP::Date qw(time2str); +use HTTP::Status qw(status_message); +use PublicInbox::Qspawn; # n.b. serving "description" and "cloneurl" should be innocuous enough to # not cause problems. serving "config" might... @@ -20,70 +23,103 @@ my @binary = qw! 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, $path) = @_; + + # Documentation/technical/http-protocol.txt in git.git + # requires one and exactly one query parameter: + if ($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'}) { + $io->close; # this is Danga::Socket::close + } +} + +sub serve_dumb { + my ($env, $git, $path) = @_; + + my @h; my $type; if ($path =~ /\A(?:$BIN)\z/o) { $type = 'application/octet-stream'; + push @h, 'Expires', time2str(time + 31536000); + push @h, 'Cache-Control', 'public, max-age=31536000'; } elsif ($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 $f = (ref $git ? $git->{git_dir} : $git) . '/' . $path; + return r(404) unless -f $f && -r _; # just in case it's a FIFO :P my @st = stat(_); my $size = $st[7]; - # TODO: If-Modified-Since and Last-Modified + # 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 $r = sysread($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 '') { @@ -112,12 +148,12 @@ sub prepare_range { 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, $len); @@ -125,26 +161,14 @@ sub prepare_range { # returns undef if 403 so it falls back to dumb HTTP sub serve_smart { - my ($cgi, $git, $path) = @_; - my $env = $cgi->{env}; - - my $input = $env->{'psgi.input'}; - my $buf; - my $in; - my $err = $env->{'psgi.errors'}; - my $fd = eval { fileno($input) }; - if (defined $fd && $fd >= 0) { - $in = $input; - } else { + my ($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_HTTP_EXPORT_ALL, GIT_COMMITTER_NAME, GIT_COMMITTER_EMAIL + # 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 @@ -155,83 +179,71 @@ sub serve_smart { my $val = $env->{$name}; $env{$name} = $val if defined $val; } - my $git_dir = $git->{git_dir}; + my $git_dir = ref $git ? $git->{git_dir} : $git; $env{GIT_HTTP_EXPORT_ALL} = '1'; $env{PATH_TRANSLATED} = "$git_dir/$path"; - my %rdr = ( 0 => fileno($in), 1 => fileno($wpipe) ); - my $pid = spawn([qw(git http-backend)], \%env, \%rdr); - unless (defined $pid) { - $err->print("error spawning: $!\n"); - return r(500); - } - $wpipe = $in = undef; - $buf = ''; - my ($vin, $fh, $res); + my %rdr = ( 0 => fileno($in) ); + my $x = PublicInbox::Qspawn->new([qw(git http-backend)], \%env, \%rdr); + my ($fh, $rpipe); my $end = sub { - if ($fh) { - $fh->close; - $fh = undef; - } else { - $res->(r(500)) if $res; - } - if ($rpipe) { - $rpipe->close; # _may_ be Danga::Socket::close - $rpipe = undef; - } - if (defined $pid) { - my $wpid = $pid; - $pid = undef; - return if $wpid == waitpid($wpid, 0); - $err->print("git http-backend ($git_dir): $?\n"); + if (my $err = $x->finish) { + err($env, "git http-backend ($git_dir): $err"); } + $fh->close if $fh; # async-only }; - my $fail = sub { - my ($e) = @_; - if ($e eq 'EAGAIN') { - select($vin, undef, undef, undef) if defined $vin; - # $vin is undef on async, so this is a noop on EAGAIN - return; + + # 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'}; + my $io = $env->{'psgix.io'}; + my $cb = sub { + 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->(); + } + return $res->($r); + } + if ($async) { + $fh = $res->($r); + return $async->async_pass($io, $fh, \$buf); } - $end->(); - $err->print("git http-backend ($git_dir): $e\n"); + + # for synchronous PSGI servers + require PublicInbox::GetlineBody; + $r->[2] = PublicInbox::GetlineBody->new($rpipe, $end, $buf); + $res->($r); }; - my $cb = sub { # read git-http-backend output and stream to client - my $r = $rpipe ? $rpipe->sysread($buf, 8192, length($buf)) : 0; - return $fail->($!{EAGAIN} ? 'EAGAIN' : $!) unless defined $r; - return $end->() if $r == 0; # EOF - if ($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) { - ($code) = ($v =~ /\b(\d+)\b/); - } else { - push @h, $k, $v; - } + sub { + ($res) = @_; + + # hopefully this doesn't break any middlewares, + # holding the input here is a waste of FDs and memory + $env->{'psgi.input'} = undef; + + $x->start(sub { # may run later, much later... + ($rpipe) = @_; + $in = undef; + if ($async) { + $async = $async->($rpipe, $cb, $end); + } else { # generic PSGI + $cb->() while $rd_hdr; } - # write response header: - $fh = $res->([ $code, \@h ]); - $res = undef; - $fh->write($buf); - $buf = ''; - } # else { keep reading ... } + }); }; - if (my $async = $env->{'pi-httpd.async'}) { - $rpipe = $async->($rpipe, $cb); - sub { ($res) = @_ } # let Danga::Socket handle the rest. - } else { # synchronous loop for other PSGI servers - $vin = ''; - vec($vin, fileno($rpipe), 1) = 1; - sub { - ($res) = @_; - while ($rpipe) { $cb->() } - } - } } sub input_to_file { @@ -242,8 +254,7 @@ sub input_to_file { while (1) { my $r = $input->read($buf, 8192); unless (defined $r) { - my $err = $env->{'psgi.errors'}; - $err->print("error reading input: $!\n"); + err($env, "error reading input: $!"); return; } last if ($r == 0); @@ -254,4 +265,21 @@ sub input_to_file { 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;