1 # Copyright (C) all contributors <meta@public-inbox.org>
2 # License: AGPL-3.0+ <https://www.gnu.org/licenses/agpl-3.0.txt>
4 # when no endpoints match, fallback to this and serve a static file
5 # or smart HTTP. This is our wrapper for git-http-backend(1)
6 package PublicInbox::GitHTTPBackend;
10 use IO::Handle; # ->flush
11 use HTTP::Date qw(time2str);
12 use PublicInbox::Qspawn;
13 use PublicInbox::Tmpfile;
14 use PublicInbox::WwwStatic qw(r @NO_CACHE);
17 # 32 is same as the git-daemon connection limit
18 my $default_limiter = PublicInbox::Qspawn::Limiter->new(32);
20 # n.b. serving "description" and "cloneurl" should be innocuous enough to
21 # not cause problems. serving "config" might...
22 my @text = qw[HEAD info/refs info/attributes
23 objects/info/(?:http-alternates|alternates|packs)
24 cloneurl description];
26 my @binary = ('objects/[a-f0-9]{2}/[a-f0-9]{38,62}',
27 'objects/pack/pack-[a-f0-9]{40,64}\.(?:pack|idx)');
29 our $ANY = join('|', @binary, @text, 'git-upload-pack');
30 my $TEXT = join('|', @text);
33 my ($env, $git, $path) = @_;
35 # Documentation/technical/http-protocol.txt in git.git
36 # requires one and exactly one query parameter:
37 if ($env->{QUERY_STRING} =~ /\Aservice=git-[A-Za-z0-9_]+-pack\z/ ||
38 $path =~ /\Agit-[A-Za-z0-9_]+-pack\z/) {
39 my $ok = serve_smart($env, $git, $path);
43 serve_dumb($env, $git, $path);
46 sub ucarp { Carp::carp(@_); undef }
52 my $t = time + 31536000;
53 push @$h, 'Expires', $t == $prev ? $exp : ($exp = time2str($prev = $t)),
54 'Cache-Control', 'public, max-age=31536000';
58 my ($env, $git, $path) = @_;
62 if ($path =~ m!\Aobjects/[a-f0-9]{2}/[a-f0-9]{38,62}\z!) {
63 $type = 'application/x-git-loose-object';
65 } elsif ($path =~ m!\Aobjects/pack/pack-[a-f0-9]{40,64}\.pack\z!) {
66 $type = 'application/x-git-packed-objects';
68 } elsif ($path =~ m!\Aobjects/pack/pack-[a-f0-9]{40,64}\.idx\z!) {
69 $type = 'application/x-git-packed-objects-toc';
71 } elsif ($path =~ /\A(?:$TEXT)\z/o) {
77 $path = "$git->{git_dir}/$path";
78 PublicInbox::WwwStatic::response($env, $h, $path, $type);
81 sub git_parse_hdr { # {parse_hdr} for Qspawn
82 my ($r, $bref, $dumb_args) = @_;
83 my $res = parse_cgi_headers($r, $bref) or return; # incomplete
84 $res->[0] == 403 ? serve_dumb(@$dumb_args) : $res;
87 # returns undef if 403 so it falls back to dumb HTTP
89 my ($env, $git, $path) = @_;
91 # GIT_COMMITTER_NAME, GIT_COMMITTER_EMAIL
92 # may be set in the server-process and are passed as-is
93 foreach my $name (qw(QUERY_STRING
94 REMOTE_USER REMOTE_ADDR
100 my $val = $env->{$name};
101 $env{$name} = $val if defined $val;
103 my $limiter = $git->{-httpbackend_limiter} || $default_limiter;
104 $env{GIT_HTTP_EXPORT_ALL} = '1';
105 $env{PATH_TRANSLATED} = "$git->{git_dir}/$path";
106 my $rdr = input_prepare($env) or return r(500);
107 my $qsp = PublicInbox::Qspawn->new([qw(git http-backend)], \%env, $rdr);
108 $qsp->psgi_return($env, $limiter, \&git_parse_hdr, [$env, $git, $path]);
114 my $input = $env->{'psgi.input'};
115 my $fd = eval { fileno($input) };
116 return { 0 => $fd } if (defined $fd && $fd >= 0);
117 my $id = "git-http.input.$env->{REMOTE_ADDR}:$env->{REMOTE_PORT}";
118 my $in = tmpfile($id) // return ucarp("tmpfile: $!");
121 my $r = $input->read($buf, 8192) // return ucarp("read $!");
123 print $in $buf // return ucarp("print: $!");
125 # ensure it's visible to git-http-backend(1):
126 $in->flush // return ucarp("flush: $!");
127 sysseek($in, 0, SEEK_SET) // return ucarp($env, "seek: $!");
131 sub parse_cgi_headers { # {parse_hdr} for Qspawn
132 my ($r, $bref, $ctx) = @_;
133 return r(500) unless defined $r && $r >= 0;
134 $$bref =~ s/\A(.*?)\r?\n\r?\n//s or return $r == 0 ? r(500) : undef;
138 foreach my $l (split(/\r?\n/, $h)) {
139 my ($k, $v) = split(/:\s*/, $l, 2);
140 if ($k =~ /\AStatus\z/i) {
141 ($code) = ($v =~ /\b([0-9]+)\b/);
147 # fallback to WwwCoderepo if cgit 404s. Duplicating $ctx prevents
148 # ->finalize from the current Qspawn from using qspawn.wcb.
149 # This makes qspawn skip ->async_pass and causes
150 # PublicInbox::HTTPD::Async::event_step to close shortly after
151 if ($code == 404 && $ctx->{www} && !$ctx->{_coderepo_tried}++) {
153 $ctx{env} = +{ %{$ctx->{env}} };
154 delete $ctx->{env}->{'qspawn.wcb'};
155 $ctx->{env}->{'plack.skip-deflater'} = 1; # prevent 2x gzip
156 my $res = $ctx->{www}->coderepo->srv(\%ctx);
157 # for ->psgi_return_init_cb:
158 $ctx->{env}->{'qspawn.wcb'} = $ctx{env}->{'qspawn.wcb'};
159 $res; # CODE or ARRAY ref