1 # Copyright (C) 2021 all contributors <meta@public-inbox.org>
2 # License: AGPL-3.0+ <https://www.gnu.org/licenses/agpl-3.0.txt>
4 # "lei add-external --mirror" support (also "public-inbox-clone");
5 package PublicInbox::LeiMirror;
8 use parent qw(PublicInbox::IPC);
9 use PublicInbox::Config;
10 use PublicInbox::AutoReap;
11 use IO::Uncompress::Gunzip qw(gunzip $GunzipError);
12 use IO::Compress::Gzip qw(gzip $GzipError);
13 use PublicInbox::Spawn qw(popen_rd spawn);
15 use Fcntl qw(SEEK_SET O_CREAT O_EXCL O_WRONLY);
18 sub _wq_done_wait { # dwaitpid callback (via wq_eof)
20 my ($mrr, $lei) = @$arg;
21 my $f = "$mrr->{dst}/mirror.done";
23 $lei->child_error($?);
24 } elsif (!unlink($f)) {
25 warn("unlink($f): $!\n") unless $!{ENOENT};
27 if ($lei->{cmd} ne 'public-inbox-clone') {
28 # calls _finish_add_external
29 $lei->lazy_cb('add-external', '_finish_'
30 )->($lei, $mrr->{dst});
32 $lei->qerr("# mirrored $mrr->{src} => $mrr->{dst}");
37 # for old installations without manifest.js.gz
40 my $uri = URI->new($self->{src});
41 my $lei = $self->{lei};
42 my $curl = $self->{curl} //= PublicInbox::LeiCurl->new($lei) or return;
43 my $cmd = $curl->for_uri($lei, $uri, '--compressed');
44 my $opt = { 0 => $lei->{0}, 2 => $lei->{2} };
45 my $fh = popen_rd($cmd, undef, $opt);
46 my $html = do { local $/; <$fh> } // die "read(curl $uri): $!";
47 close($fh) or return $lei->child_error($?, "@$cmd failed");
49 # we grep with URL below, we don't want Subject/From headers
50 # making us clone random URLs
51 my @html = split(/<hr>/, $html);
52 my @urls = ($html[-1] =~ m!\bgit clone --mirror ([a-z\+]+://\S+)!g);
53 my $url = $uri->as_string;
54 chop($url) eq '/' or die "BUG: $uri not canonicalized";
56 # since this is for old instances w/o manifest.js.gz, try v1 first
57 return clone_v1($self) if grep(m!\A\Q$url\E/*\z!, @urls);
58 if (my @v2_urls = grep(m!\A\Q$url\E/[0-9]+\z!, @urls)) {
60 my ($n) = (m!/([0-9]+)\z!);
63 return clone_v2($self, \%v2_epochs);
66 # filter out common URLs served by WWW (e.g /$MSGID/T/)
67 if (@urls && $url =~ s!/+[^/]+\@[^/]+/.*\z!! &&
68 grep(m!\A\Q$url\E/*\z!, @urls)) {
70 E: confused by scraping <$uri>, did you mean <$url>?
74 E: confused by scraping <$uri>, got ambiguous results:
77 die "E: scraping <$uri> revealed nothing\n";
83 $opt->{$_} = $lei->{$_} for (0..2);
84 # we support "-c $key=$val" for arbitrary git config options
85 # e.g.: git -c http.proxy=socks5h://127.0.0.1:9050
86 push(@cmd, '-c', $_) for @{$lei->{opt}->{c} // []};
87 push @cmd, qw(clone --mirror);
88 push @cmd, '-q' if $lei->{opt}->{quiet};
89 push @cmd, '-v' if $lei->{opt}->{verbose};
90 # XXX any other options to support?
91 # --reference is tricky with multiple epochs...
96 my ($ft, $dst, $open_mode) = @_;
97 my $fn = $ft->filename;
99 my $mode = @st ? ($st[2] & 07777) : ($open_mode & ~umask);
100 chmod($mode, $ft) or croak "E: chmod $fn: $!";
101 rename($fn, $dst) or croak "E: rename($fn => $ft): $!";
102 $ft->unlink_on_destroy(0);
105 sub _get_txt { # non-fatal
106 my ($self, $endpoint, $file, $mode) = @_;
107 my $uri = URI->new($self->{src});
108 my $lei = $self->{lei};
109 my $path = $uri->path;
110 chop($path) eq '/' or die "BUG: $uri not canonicalized";
111 $uri->path("$path/$endpoint");
112 my $ft = File::Temp->new(TEMPLATE => "$file-XXXX", DIR => $self->{dst});
113 my $opt = { 0 => $lei->{0}, 1 => $lei->{1}, 2 => $lei->{2} };
114 my $cmd = $self->{curl}->for_uri($lei, $uri,
115 qw(--compressed -R -o), $ft->filename);
116 my $cerr = run_reap($lei, $cmd, $opt);
117 return "$uri missing" if ($cerr >> 8) == 22;
118 return "# @$cmd failed (non-fatal)" if $cerr;
119 ft_rename($ft, "$self->{dst}/$file", $mode);
123 # tries the relatively new /$INBOX/_/text/config/raw endpoint
126 my $dst = $self->{dst};
127 if (!-d $dst || !mkdir($dst)) {
129 File::Path::mkpath($dst);
130 -d $dst or die "mkpath($dst): $!\n";
132 my $err = _get_txt($self,
133 qw(_/text/config/raw inbox.config.example), 0444);
134 return warn($err, "\n") if $err;
135 my $f = "$self->{dst}/inbox.config.example";
136 my $cfg = PublicInbox::Config->git_config_dump($f, $self->{lei}->{2});
137 my $ibx = $self->{ibx} = {};
138 for my $sec (grep(/\Apublicinbox\./, @{$cfg->{-section_order}})) {
139 for (qw(address newsgroup nntpmirror)) {
140 $ibx->{$_} = $cfg->{"$sec.$_"};
145 sub set_description ($) {
147 my $f = "$self->{dst}/description";
148 open my $fh, '+>>', $f or die "open($f): $!";
149 seek($fh, 0, SEEK_SET) or die "seek($f): $!";
150 chomp(my $d = do { local $/; <$fh> } // die "read($f): $!");
151 if ($d eq '($INBOX_DIR/description missing)' ||
152 $d =~ /^Unnamed repository/ || $d !~ /\S/) {
153 seek($fh, 0, SEEK_SET) or die "seek($f): $!";
154 truncate($fh, 0) or die "truncate($f): $!";
155 print $fh "mirror of $self->{src}\n" or die "print($f): $!";
156 close $fh or die "close($f): $!";
160 sub index_cloned_inbox {
161 my ($self, $iv) = @_;
162 my $lei = $self->{lei};
163 my $err = _get_txt($self, qw(description description), 0666);
164 warn($err, "\n") if $err; # non fatal
165 eval { set_description($self) };
168 # n.b. public-inbox-clone works w/o (SQLite || Xapian)
169 # lei is useless without Xapian + SQLite
170 if ($lei->{cmd} ne 'public-inbox-clone') {
171 my $ibx = delete($self->{ibx}) // {
172 address => [ 'lei@example.com' ],
175 $ibx->{inboxdir} = $self->{dst};
176 PublicInbox::Inbox->new($ibx);
177 PublicInbox::InboxWritable->new($ibx);
179 for my $sw ($lei->index_opt) {
180 my ($k) = ($sw =~ /\A([\w-]+)/);
181 $opt->{$k} = $lei->{opt}->{$k};
183 # force synchronous dwaitpid for v2:
184 local $PublicInbox::DS::in_loop = 0;
185 my $cfg = PublicInbox::Config->new(undef, $lei->{2});
186 my $env = PublicInbox::Admin::index_prepare($opt, $cfg);
187 local %ENV = (%ENV, %$env) if $env;
188 PublicInbox::Admin::progress_prepare($opt, $lei->{2});
189 PublicInbox::Admin::index_inbox($ibx, undef, $opt);
191 open my $x, '>', "$self->{dst}/mirror.done"; # for _wq_done_wait
195 my ($lei, $cmd, $opt) = @_;
196 $lei->qerr("# @$cmd");
197 my $ar = PublicInbox::AutoReap->new(spawn($cmd, undef, $opt));
200 $? = 0; # don't let it influence normal exit
206 my $lei = $self->{lei};
207 my $curl = $self->{curl} //= PublicInbox::LeiCurl->new($lei) or return;
208 my $uri = URI->new($self->{src});
209 defined($lei->{opt}->{epoch}) and
210 die "$uri is a v1 inbox, --epoch is not supported\n";
211 my $pfx = $curl->torsocks($lei, $uri) or return;
212 my $cmd = [ @$pfx, clone_cmd($lei, my $opt = {}),
213 $uri->as_string, $self->{dst} ];
214 my $cerr = run_reap($lei, $cmd, $opt);
215 return $lei->child_error($cerr, "@$cmd failed") if $cerr;
217 write_makefile($self->{dst}, 1);
218 index_cloned_inbox($self, 1);
221 sub parse_epochs ($$) {
222 my ($opt_epochs, $v2_epochs) = @_; # $epcohs "LOW..HIGH"
223 $opt_epochs // return; # undef => all epochs
224 my ($lo, $dotdot, $hi, @extra) = split(/(\.\.)/, $opt_epochs);
225 undef($lo) if ($lo // '') eq '';
226 my $re = qr/\A~?[0-9]+\z/;
227 if (@extra || (($lo // '0') !~ $re) ||
228 (($hi // '0') !~ $re) ||
229 !(grep(defined, $lo, $hi))) {
231 --epoch=$opt_epochs not in the form of `LOW..HIGH', `LOW..', nor `..HIGH'
234 my @n = sort { $a <=> $b } keys %$v2_epochs;
235 for (grep(defined, $lo, $hi)) {
238 "`$_' exceeds maximum available epoch ($n[-1])\n";
240 "`$_' is lower than minimum available epoch ($n[0])\n";
241 } elsif (/\A~([0-9]+)/) {
243 $n[$off] // die "`$_' is out of range\n";
245 } else { die "`$_' not understood\n" }
247 defined($lo) && defined($hi) && $lo > $hi and die
248 "low value (`$lo') exceeds high (`$hi')\n";
249 $lo //= $n[0] if $dotdot;
250 $hi //= $n[-1] if $dotdot;
254 if (defined $v2_epochs->{$_}) {
258 "# epoch $_ is not available (non-fatal, $lo..$hi)\n";
264 sub init_placeholder ($$) {
265 my ($src, $edst) = @_;
266 PublicInbox::Import::init_bare($edst);
267 my $f = "$edst/config";
268 open my $fh, '>>', $f or die "open($f): $!";
269 print $fh <<EOM or die "print($f): $!";
272 fetch = +refs/*:refs/*
275 ; This git epoch was created read-only and "public-inbox-fetch"
276 ; will not fetch updates for it unless write permission is added.
278 close $fh or die "close:($f): $!";
281 sub clone_v2 ($$;$) {
282 my ($self, $v2_epochs, $m) = @_; # $m => manifest.js.gz hashref
283 my $lei = $self->{lei};
284 my $curl = $self->{curl} //= PublicInbox::LeiCurl->new($lei) or return;
285 my $pfx = $curl->torsocks($lei, (values %$v2_epochs)[0]) or return;
286 my $dst = $self->{dst};
287 my $want = parse_epochs($lei->{opt}->{epoch}, $v2_epochs);
288 my (@src_edst, @read_only, @skip_nr);
289 for my $nr (sort { $a <=> $b } keys %$v2_epochs) {
290 my $uri = $v2_epochs->{$nr};
291 my $src = $uri->as_string;
293 $src =~ m!/([0-9]+)(?:\.git)?\z! or die <<"";
294 failed to extract epoch number from $src
296 $1 + 0 == $nr or die "BUG: <$uri> miskeyed $1 != $nr";
297 $edst .= "/git/$nr.git";
298 if (!$want || $want->{$nr}) {
299 push @src_edst, $src, $edst;
300 } else { # create a placeholder so users only need to chmod +w
301 init_placeholder($src, $edst);
302 push @read_only, $edst;
306 if (@skip_nr) { # filter out the epochs we skipped
307 my $re = join('|', @skip_nr);
308 my @del = grep(m!/git/$re\.git\z!, keys %$m);
310 $self->{-culled_manifest} = 1;
312 my $lk = bless { lock_path => "$dst/inbox.lock" }, 'PublicInbox::Lock';
314 my $on_destroy = $lk->lock_for_scope($$);
315 my @cmd = clone_cmd($lei, my $opt = {});
316 while (my ($src, $edst) = splice(@src_edst, 0, 2)) {
317 my $cmd = [ @$pfx, @cmd, $src, $edst ];
318 my $cerr = run_reap($lei, $cmd, $opt);
319 return $lei->child_error($cerr, "@$cmd failed") if $cerr;
321 require PublicInbox::MultiGit;
322 my $mg = PublicInbox::MultiGit->new($dst, 'all.git', 'git');
323 $mg->fill_alternates;
324 for my $i ($mg->git_epochs) { $mg->epoch_cfg_set($i) }
325 for my $edst (@read_only) {
326 my @st = stat($edst) or die "stat($edst): $!";
327 chmod($st[2] & 0555, $edst) or die "chmod(a-w, $edst): $!";
329 write_makefile($self->{dst}, 2);
330 undef $on_destroy; # unlock
331 index_cloned_inbox($self, 2);
334 # PSGI mount prefixes and manifest.js.gz prefixes don't always align...
335 sub deduce_epochs ($$) {
337 my ($v1_ent, @v2_epochs);
341 $v1_ent = $m->{$path};
342 @v2_epochs = grep(m!\A\Q$path\E/git/[0-9]+\.git\z!, keys %$m);
343 } while (!defined($v1_ent) && !@v2_epochs &&
344 $path =~ s!\A(/[^/]+)/!/! and $path_pfx .= $1);
345 ($path_pfx, $v1_ent ? $path : undef, @v2_epochs);
348 sub decode_manifest ($$$) {
349 my ($fh, $fn, $uri) = @_;
351 my $gz = do { local $/; <$fh> } // die "slurp($fn): $!";
352 gunzip(\$gz => \$js, MultiStream => 1) or
353 die "gunzip($uri): $GunzipError\n";
354 my $m = eval { PublicInbox::Config->json->decode($js) };
355 die "$uri: error decoding `$js': $@\n" if $@;
356 ref($m) eq 'HASH' or die "$uri unknown type: ".ref($m);
362 my $uri = URI->new($self->{src});
363 my $lei = $self->{lei};
364 my $curl = $self->{curl} //= PublicInbox::LeiCurl->new($lei) or return;
365 my $path = $uri->path;
366 chop($path) eq '/' or die "BUG: $uri not canonicalized";
367 $uri->path($path . '/manifest.js.gz');
368 my $pdir = $lei->rel2abs($self->{dst});
369 $pdir =~ s!/[^/]+/?\z!!;
370 my $ft = File::Temp->new(TEMPLATE => 'm-XXXX',
371 UNLINK => 1, DIR => $pdir, SUFFIX => '.tmp');
372 my $fn = $ft->filename;
373 my ($bn) = ($fn =~ m!/([^/]+)\z!);
374 my $cmd = $curl->for_uri($lei, $uri, '-R', '-o', $bn);
375 my $opt = { -C => $pdir };
376 $opt->{$_} = $lei->{$_} for (0..2);
377 my $cerr = run_reap($lei, $cmd, $opt);
379 return try_scrape($self) if ($cerr >> 8) == 22; # 404 missing
380 return $lei->child_error($cerr, "@$cmd failed");
382 my $m = eval { decode_manifest($ft, $fn, $uri) };
385 return try_scrape($self);
387 my ($path_pfx, $v1_path, @v2_epochs) = deduce_epochs($m, $path);
389 # It may be possible to have v1 + v2 in parallel someday:
390 warn(<<EOM) if defined $v1_path;
391 # `$v1_path' appears to be a v1 inbox while v2 epochs exist:
393 # ignoring $v1_path (use --inbox-version=1 to force v1 instead)
395 my %v2_epochs = map {
396 $uri->path($path_pfx.$_);
397 my ($n) = ("$uri" =~ m!/([0-9]+)\.git\z!);
400 clone_v2($self, \%v2_epochs, $m);
401 } elsif (defined $v1_path) {
404 die "E: confused by <$uri>, possible matches:\n\t",
405 join(', ', sort keys %$m), "\n";
407 if (delete $self->{-culled_manifest}) { # set by clone_v2
408 # write the smaller manifest if epochs were skipped so
409 # users won't have to delete manifest if they +w an
410 # epoch they no longer want to skip
411 my $json = PublicInbox::Config->json->encode($m);
412 my $mtime = (stat($fn))[9];
413 gzip(\$json => $fn) or die "gzip: $GzipError";
414 utime($mtime, $mtime, $fn) or die "utime(..., $fn): $!";
416 ft_rename($ft, "$self->{dst}/manifest.js.gz", 0666);
419 sub start_clone_url {
421 return try_manifest($self) if $self->{src} =~ m!\Ahttps?://!;
422 die "TODO: non-HTTP/HTTPS clone of $self->{src} not supported, yet";
425 sub do_mirror { # via wq_io_do
427 my $lei = $self->{lei};
428 umask($lei->{client_umask}) if defined $lei->{client_umask};
430 my $iv = $lei->{opt}->{'inbox-version'};
432 return clone_v1($self) if $iv == 1;
433 return try_scrape($self) if $iv == 2;
434 die "bad --inbox-version=$iv\n";
436 return start_clone_url($self) if $self->{src} =~ m!://!;
437 die "TODO: cloning local directories not supported, yet";
439 $lei->fail($@) if $@;
443 my ($cls, $lei, $src, $dst) = @_;
444 my $self = bless { src => $src, dst => $dst }, $cls;
445 if ($src =~ m!https?://!) {
447 require PublicInbox::LeiCurl;
449 require PublicInbox::Lock;
450 require PublicInbox::Inbox;
451 require PublicInbox::Admin;
452 require PublicInbox::InboxWritable;
454 my ($op_c, $ops) = $lei->workers_start($self, 1);
456 $self->wq_io_do('do_mirror', []);
458 $lei->wait_wq_events($op_c, $ops);
461 sub ipc_atfork_child {
463 $self->{lei}->_lei_atfork_child;
464 $self->SUPER::ipc_atfork_child;
468 my ($dir, $ibx_ver) = @_;
469 my $f = "$dir/Makefile";
470 if (sysopen my $fh, $f, O_CREAT|O_EXCL|O_WRONLY) {
471 print $fh <<EOM or die "print($f) $!";
472 # This is a v$ibx_ver public-inbox, see the public-inbox-v$ibx_ver-format(5)
473 # manpage for more information on the format. This Makefile is
474 # intended as a familiar wrapper for users unfamiliar with
475 # public-inbox-* commands.
477 # See the respective manpages for public-inbox-fetch(1),
478 # public-inbox-index(1), etc for more information on
479 # some of the commands used by this Makefile.
481 # This Makefile will not be modified nor read by public-inbox,
482 # so you may edit it freely with your own convenience targets
483 # and notes. public-inbox-fetch will recreate it if removed.
485 print $fh <<'EOM' or die "print($f): $!";
486 # the default target:
488 @echo Common targets:
489 @echo ' make fetch - fetch from remote git repostorie(s)'
490 @echo ' make update - fetch and update index '
492 @echo Rarely needed targets:
493 @echo ' make reindex - may be needed for new features/bugfixes'
494 @echo ' make compact - rewrite Xapian storage to save space'
499 @if ! public-inbox-fetch --exit-code; \
502 test $$c -eq 127 && exit 0; \
504 elif test -f msgmap.sqlite3 || test -f public-inbox/msgmap.sqlite3; \
506 public-inbox-index; \
508 echo 'public-inbox index not initialized'; \
509 echo 'see public-inbox-index(1) man page'; \
512 public-inbox-index --reindex
516 .PHONY : help fetch update reindex compact
518 close $fh or die "close($f): $!";
520 die "open($f): $!" unless $!{EEXIST};