X-Git-Url: http://www.git.stargrave.org/?a=blobdiff_plain;f=lib%2FPublicInbox%2FGit.pm;h=49c0d5d60760c7c3d0a7c2149df22fc4c53811bc;hb=af0b0fb7a454470a32c452119d0392e0dedb3fe1;hp=8d3f87d577565ea2f07123d890cde23a043185a5;hpb=fe20d568e82cdb3645b42f18f1691d64271aaf7b;p=public-inbox.git diff --git a/lib/PublicInbox/Git.pm b/lib/PublicInbox/Git.pm index 8d3f87d5..49c0d5d6 100644 --- a/lib/PublicInbox/Git.pm +++ b/lib/PublicInbox/Git.pm @@ -1,4 +1,4 @@ -# Copyright (C) 2014-2018 all contributors +# Copyright (C) 2014-2021 all contributors # License: GPLv2 or later # # Used to read files from a git repository without excessive forking. @@ -8,12 +8,29 @@ # There are also API changes to simplify our usage and data set. package PublicInbox::Git; use strict; -use warnings; -use POSIX qw(dup2); -require IO::Handle; -use PublicInbox::Spawn qw(spawn popen_rd); -use base qw(Exporter); -our @EXPORT_OK = qw(git_unquote); +use v5.10.1; +use parent qw(Exporter); +use POSIX (); +use IO::Handle; # ->autoflush +use Errno qw(EINTR EAGAIN); +use File::Glob qw(bsd_glob GLOB_NOSORT); +use File::Spec (); +use Time::HiRes qw(stat); +use PublicInbox::Spawn qw(popen_rd); +use PublicInbox::Tmpfile; +use IO::Poll qw(POLLIN); +use Carp qw(croak); +use Digest::SHA (); +use PublicInbox::DS qw(dwaitpid); +our @EXPORT_OK = qw(git_unquote git_quote); +our $PIPE_BUFSIZ = 65536; # Linux default +our $in_cleanup; +our $RDTIMEO = 60_000; # milliseconds + +use constant MAX_INFLIGHT => + (($^O eq 'linux' ? 4096 : POSIX::_POSIX_PIPE_BUF()) * 3) + / + 65; # SHA-256 hex size + "\n" in preparation for git using non-SHA1 my %GIT_ESC = ( a => "\a", @@ -23,145 +40,316 @@ my %GIT_ESC = ( r => "\r", t => "\t", v => "\013", + '"' => '"', + '\\' => '\\', ); +my %ESC_GIT = map { $GIT_ESC{$_} => $_ } keys %GIT_ESC; + # unquote pathnames used by git, see quote.c::unquote_c_style.c in git.git sub git_unquote ($) { - my ($s) = @_; - return $s unless ($s =~ /\A"(.*)"\z/); - $s = $1; - $s =~ s/\\([abfnrtv])/$GIT_ESC{$1}/g; - $s =~ s/\\([0-7]{1,3})/chr(oct($1))/ge; - $s; + return $_[0] unless ($_[0] =~ /\A"(.*)"\z/); + $_[0] = $1; + $_[0] =~ s/\\([\\"abfnrtv])/$GIT_ESC{$1}/g; + $_[0] =~ s/\\([0-7]{1,3})/chr(oct($1))/ge; + $_[0]; +} + +sub git_quote ($) { + if ($_[0] =~ s/([\\"\a\b\f\n\r\t\013]|[^[:print:]])/ + '\\'.($ESC_GIT{$1}||sprintf("%0o",ord($1)))/egs) { + return qq{"$_[0]"}; + } + $_[0]; } sub new { my ($class, $git_dir) = @_; - my @st; - $st[7] = $st[10] = 0; - bless { git_dir => $git_dir, st => \@st }, $class + # may contain {-tmp} field for File::Temp::Dir + bless { git_dir => $git_dir, alt_st => '', -git_path => {} }, $class +} + +sub git_path ($$) { + my ($self, $path) = @_; + $self->{-git_path}->{$path} ||= do { + local $/ = "\n"; + chomp(my $str = $self->qx(qw(rev-parse --git-path), $path)); + + # git prior to 2.5.0 did not understand --git-path + if ($str eq "--git-path\n$path") { + $str = "$self->{git_dir}/$path"; + } + $str; + }; } sub alternates_changed { my ($self) = @_; - my $alt = "$self->{git_dir}/objects/info/alternates"; + my $alt = git_path($self, 'objects/info/alternates'); my @st = stat($alt) or return 0; - my $old_st = $self->{st}; - # 10 - ctime, 7 - size - return 0 if ($st[10] == $old_st->[10] && $st[7] == $old_st->[7]); - $self->{st} = \@st; + + # can't rely on 'q' on some 32-bit builds, but `d' works + my $st = pack('dd', $st[10], $st[7]); # 10: ctime, 7: size + return 0 if $self->{alt_st} eq $st; + $self->{alt_st} = $st; # always a true value } -sub _bidi_pipe { - my ($self, $batch, $in, $out, $pid) = @_; - return if $self->{$pid}; - my ($in_r, $in_w, $out_r, $out_w); +sub last_check_err { + my ($self) = @_; + my $fh = $self->{err_c} or return; + sysseek($fh, 0, 0) or $self->fail("sysseek failed: $!"); + defined(sysread($fh, my $buf, -s $fh)) or + $self->fail("sysread failed: $!"); + $buf; +} - pipe($in_r, $in_w) or fail($self, "pipe failed: $!"); - pipe($out_r, $out_w) or fail($self, "pipe failed: $!"); - if ($^O eq 'linux') { # 1031: F_SETPIPE_SZ - fcntl($out_w, 1031, 4096); - fcntl($in_w, 1031, 4096) if $batch eq '--batch-check'; +sub _bidi_pipe { + my ($self, $batch, $in, $out, $pid, $err) = @_; + if ($self->{$pid}) { + if (defined $err) { # "err_c" + my $fh = $self->{$err}; + sysseek($fh, 0, 0) or $self->fail("sysseek failed: $!"); + truncate($fh, 0) or $self->fail("truncate failed: $!"); + } + return; } - - my @cmd = ('git', "--git-dir=$self->{git_dir}", qw(cat-file), $batch); - my $redir = { 0 => fileno($out_r), 1 => fileno($in_w) }; - my $p = spawn(\@cmd, undef, $redir); - defined $p or fail($self, "spawn failed: $!"); + my ($out_r, $out_w); + pipe($out_r, $out_w) or $self->fail("pipe failed: $!"); + my @cmd = (qw(git), "--git-dir=$self->{git_dir}", + qw(-c core.abbrev=40 cat-file), $batch); + my $redir = { 0 => $out_r }; + if ($err) { + my $id = "git.$self->{git_dir}$batch.err"; + my $fh = tmpfile($id) or $self->fail("tmpfile($id): $!"); + $self->{$err} = $fh; + $redir->{2} = $fh; + } + my ($in_r, $p) = popen_rd(\@cmd, undef, $redir); $self->{$pid} = $p; + $self->{"$pid.owner"} = $$; $out_w->autoflush(1); + if ($^O eq 'linux') { # 1031: F_SETPIPE_SZ + fcntl($out_w, 1031, 4096); + fcntl($in_r, 1031, 4096) if $batch eq '--batch-check'; + } $self->{$out} = $out_w; $self->{$in} = $in_r; } -sub cat_file { - my ($self, $obj, $ref) = @_; - my ($retried, $in, $head); +sub poll_in ($) { IO::Poll::_poll($RDTIMEO, fileno($_[0]), my $ev = POLLIN) } -again: - batch_prepare($self); - $self->{out}->print($obj, "\n") or fail($self, "write error: $!"); +sub my_read ($$$) { + my ($fh, $rbuf, $len) = @_; + my $left = $len - length($$rbuf); + my $r; + while ($left > 0) { + $r = sysread($fh, $$rbuf, $PIPE_BUFSIZ, length($$rbuf)); + if ($r) { + $left -= $r; + } elsif (defined($r)) { # EOF + return 0; + } else { + next if ($! == EAGAIN and poll_in($fh)); + next if $! == EINTR; # may be set by sysread or poll_in + return; # unrecoverable error + } + } + \substr($$rbuf, 0, $len, ''); +} - $in = $self->{in}; - local $/ = "\n"; - $head = $in->getline; - if ($head =~ / missing$/) { - if (!$retried && alternates_changed($self)) { - $retried = 1; - cleanup($self); - goto again; +sub my_readline ($$) { + my ($fh, $rbuf) = @_; + while (1) { + if ((my $n = index($$rbuf, "\n")) >= 0) { + return substr($$rbuf, 0, $n + 1, ''); } - return; + my $r = sysread($fh, $$rbuf, $PIPE_BUFSIZ, length($$rbuf)) + and next; + + # return whatever's left on EOF + return substr($$rbuf, 0, length($$rbuf)+1, '') if defined($r); + + next if ($! == EAGAIN and poll_in($fh)); + next if $! == EINTR; # may be set by sysread or poll_in + return; # unrecoverable error } - $head =~ /^[0-9a-f]{40} \S+ (\d+)$/ or - fail($self, "Unexpected result from git cat-file: $head"); - - my $size = $1; - my $ref_type = $ref ? ref($ref) : ''; - - my $rv; - my $left = $size; - $$ref = $size if ($ref_type eq 'SCALAR'); - my $cb_err; - - if ($ref_type eq 'CODE') { - $rv = eval { $ref->($in, \$left) }; - $cb_err = $@; - # drain the rest - my $max = 8192; - while ($left > 0) { - my $r = read($in, my $x, $left > $max ? $max : $left); - defined($r) or fail($self, "read failed: $!"); - $r == 0 and fail($self, 'exited unexpectedly'); - $left -= $r; +} + +sub cat_async_retry ($$$$$) { + my ($self, $inflight, $req, $cb, $arg) = @_; + + # {inflight} may be non-existent, but if it isn't we delete it + # here to prevent cleanup() from waiting: + delete $self->{inflight}; + cleanup($self); + + $self->{inflight} = $inflight; + batch_prepare($self); + my $buf = "$req\n"; + for (my $i = 0; $i < @$inflight; $i += 3) { + $buf .= "$inflight->[$i]\n"; + } + print { $self->{out} } $buf or $self->fail("write error: $!"); + unshift(@$inflight, \$req, $cb, $arg); # \$ref to indicate retried + + cat_async_step($self, $inflight); # take one step +} + +sub cat_async_step ($$) { + my ($self, $inflight) = @_; + die 'BUG: inflight empty or odd' if scalar(@$inflight) < 3; + my ($req, $cb, $arg) = splice(@$inflight, 0, 3); + my $rbuf = delete($self->{cat_rbuf}) // \(my $new = ''); + my ($bref, $oid, $type, $size); + my $head = my_readline($self->{in}, $rbuf); + # ->fail may be called via Gcf2Client.pm + if ($head =~ /^([0-9a-f]{40,}) (\S+) ([0-9]+)$/) { + ($oid, $type, $size) = ($1, $2, $3 + 0); + $bref = my_read($self->{in}, $rbuf, $size + 1) or + $self->fail(defined($bref) ? 'read EOF' : "read: $!"); + chop($$bref) eq "\n" or $self->fail('LF missing after blob'); + } elsif ($head =~ s/ missing\n//s) { + $oid = $head; + # ref($req) indicates it's already been retried + # -gcf2 retries internally, so it never hits this path: + if (!ref($req) && !$in_cleanup && $self->alternates_changed) { + return cat_async_retry($self, $inflight, + $req, $cb, $arg); } + $type = 'missing'; + $oid = ref($req) ? $$req : $req if $oid eq ''; } else { - my $offset = 0; - my $buf = ''; - while ($left > 0) { - my $r = read($in, $buf, $left, $offset); - defined($r) or fail($self, "read failed: $!"); - $r == 0 and fail($self, 'exited unexpectedly'); - $left -= $r; - $offset += $r; - } - $rv = \$buf; + my $err = $! ? " ($!)" : ''; + $self->fail("bad result from async cat-file: $head$err"); } + $self->{cat_rbuf} = $rbuf if $$rbuf ne ''; + eval { $cb->($bref, $oid, $type, $size, $arg) }; + warn "E: $oid: $@\n" if $@; +} + +sub cat_async_wait ($) { + my ($self) = @_; + my $inflight = $self->{inflight} or return; + while (scalar(@$inflight)) { + cat_async_step($self, $inflight); + } +} + +sub batch_prepare ($) { + _bidi_pipe($_[0], qw(--batch in out pid)); +} - my $r = read($in, my $buf, 1); - defined($r) or fail($self, "read failed: $!"); - fail($self, 'newline missing after blob') if ($r != 1 || $buf ne "\n"); - die $cb_err if $cb_err; +sub _cat_file_cb { + my ($bref, undef, undef, $size, $result) = @_; + @$result = ($bref, $size); +} - $rv; +sub cat_file { + my ($self, $oid, $sizeref) = @_; + my $result = []; + cat_async($self, $oid, \&_cat_file_cb, $result); + cat_async_wait($self); + $$sizeref = $result->[1] if $sizeref; + $result->[0]; } -sub batch_prepare ($) { _bidi_pipe($_[0], qw(--batch in out pid)) } +sub check_async_step ($$) { + my ($self, $inflight_c) = @_; + die 'BUG: inflight empty or odd' if scalar(@$inflight_c) < 3; + my ($req, $cb, $arg) = splice(@$inflight_c, 0, 3); + my $rbuf = delete($self->{chk_rbuf}) // \(my $new = ''); + chomp(my $line = my_readline($self->{in_c}, $rbuf)); + my ($hex, $type, $size) = split(/ /, $line); + + # Future versions of git.git may have type=ambiguous, but for now, + # we must handle 'dangling' below (and maybe some other oddball + # stuff): + # https://public-inbox.org/git/20190118033845.s2vlrb3wd3m2jfzu@dcvr/T/ + if ($hex eq 'dangling' || $hex eq 'notdir' || $hex eq 'loop') { + my $ret = my_read($self->{in_c}, $rbuf, $type + 1); + $self->fail(defined($ret) ? 'read EOF' : "read: $!") if !$ret; + } + $self->{chk_rbuf} = $rbuf if $$rbuf ne ''; + eval { $cb->($hex, $type, $size, $arg, $self) }; + warn "E: check($req) $@\n" if $@; +} + +sub check_async_wait ($) { + my ($self) = @_; + my $inflight_c = $self->{inflight_c} or return; + while (scalar(@$inflight_c)) { + check_async_step($self, $inflight_c); + } +} + +sub check_async_begin ($) { + my ($self) = @_; + cleanup($self) if alternates_changed($self); + _bidi_pipe($self, qw(--batch-check in_c out_c pid_c err_c)); + die 'BUG: already in async check' if $self->{inflight_c}; + $self->{inflight_c} = []; +} + +sub check_async ($$$$) { + my ($self, $oid, $cb, $arg) = @_; + my $inflight_c = $self->{inflight_c} // check_async_begin($self); + while (scalar(@$inflight_c) >= MAX_INFLIGHT) { + check_async_step($self, $inflight_c); + } + print { $self->{out_c} } $oid, "\n" or $self->fail("write error: $!"); + push(@$inflight_c, $oid, $cb, $arg); +} + +sub _check_cb { # check_async callback + my ($hex, $type, $size, $result) = @_; + @$result = ($hex, $type, $size); +} sub check { - my ($self, $obj) = @_; - $self->_bidi_pipe(qw(--batch-check in_c out_c pid_c)); - $self->{out_c}->print($obj, "\n") or fail($self, "write error: $!"); - local $/ = "\n"; - chomp(my $line = $self->{in_c}->getline); - my ($hex, $type, $size) = split(' ', $line); - return if $type eq 'missing'; + my ($self, $oid) = @_; + my $result = []; + check_async($self, $oid, \&_check_cb, $result); + check_async_wait($self); + my ($hex, $type, $size) = @$result; + + # Future versions of git.git may show 'ambiguous', but for now, + # we must handle 'dangling' below (and maybe some other oddball + # stuff): + # https://public-inbox.org/git/20190118033845.s2vlrb3wd3m2jfzu@dcvr/T/ + return if $type eq 'missing' || $type eq 'ambiguous'; + return if $hex eq 'dangling' || $hex eq 'notdir' || $hex eq 'loop'; ($hex, $type, $size); } sub _destroy { - my ($self, $in, $out, $pid) = @_; + my ($self, $rbuf, $in, $out, $pid, $err) = @_; + delete @$self{($rbuf, $in, $out)}; + delete $self->{$err} if $err; # `err_c' + + # GitAsyncCat::event_step may delete {pid} my $p = delete $self->{$pid} or return; - foreach my $f ($in, $out) { - delete $self->{$f}; + dwaitpid($p) if $$ == $self->{"$pid.owner"}; +} + +sub cat_async_abort ($) { + my ($self) = @_; + if (my $inflight = $self->{inflight}) { + while (@$inflight) { + my ($req, $cb, $arg) = splice(@$inflight, 0, 3); + $req =~ s/ .*//; # drop git_dir for Gcf2Client + eval { $cb->(undef, $req, undef, undef, $arg) }; + warn "E: $req: $@ (in abort)\n" if $@; + } + delete $self->{cat_rbuf}; + delete $self->{inflight}; } - waitpid $p, 0; + cleanup($self); } -sub fail { +sub fail { # may be augmented in subclasses my ($self, $msg) = @_; - cleanup($self); - die $msg; + cat_async_abort($self); + croak(ref($self) . ' ' . ($self->{git_dir} // '') . ": $msg"); } sub popen { @@ -173,26 +361,43 @@ sub popen { sub qx { my ($self, @cmd) = @_; my $fh = $self->popen(@cmd); - defined $fh or return; - local $/ = "\n"; - return <$fh> if wantarray; - local $/; - <$fh> + local $/ = wantarray ? "\n" : undef; + <$fh>; +} + +# check_async and cat_async may trigger the other, so ensure they're +# both completely done by using this: +sub async_wait_all ($) { + my ($self) = @_; + while (scalar(@{$self->{inflight_c} // []}) || + scalar(@{$self->{inflight} // []})) { + $self->check_async_wait; + $self->cat_async_wait; + } } +# returns true if there are pending "git cat-file" processes sub cleanup { my ($self) = @_; - _destroy($self, qw(in out pid)); - _destroy($self, qw(in_c out_c pid_c)); + local $in_cleanup = 1; + delete $self->{async_cat}; + async_wait_all($self); + delete $self->{inflight}; + delete $self->{inflight_c}; + _destroy($self, qw(cat_rbuf in out pid)); + _destroy($self, qw(chk_rbuf in_c out_c pid_c err_c)); + !!($self->{pid} || $self->{pid_c}); } + # assuming a well-maintained repo, this should be a somewhat # accurate estimation of its size # TODO: show this in the WWW UI as a hint to potential cloners sub packed_bytes { my ($self) = @_; my $n = 0; - foreach my $p (glob("$self->{git_dir}/objects/pack/*.pack")) { + my $pack_dir = git_path($self, 'objects/pack'); + foreach my $p (bsd_glob("$pack_dir/*.pack", GLOB_NOSORT)) { $n += -s $p; } $n @@ -200,6 +405,140 @@ sub packed_bytes { sub DESTROY { cleanup(@_) } +sub local_nick ($) { + my ($self) = @_; + my $ret = '???'; + # don't show full FS path, basename should be OK: + if ($self->{git_dir} =~ m!/([^/]+)(?:/\.git)?\z!) { + $ret = "/path/to/$1"; + } + wantarray ? ($ret) : $ret; +} + +sub host_prefix_url ($$) { + my ($env, $url) = @_; + return $url if index($url, '//') >= 0; + my $scheme = $env->{'psgi.url_scheme'}; + my $host_port = $env->{HTTP_HOST} // + "$env->{SERVER_NAME}:$env->{SERVER_PORT}"; + "$scheme://$host_port". ($env->{SCRIPT_NAME} || '/') . $url; +} + +sub pub_urls { + my ($self, $env) = @_; + if (my $urls = $self->{cgit_url}) { + return map { host_prefix_url($env, $_) } @$urls; + } + local_nick($self); +} + +sub cat_async_begin { + my ($self) = @_; + cleanup($self) if $self->alternates_changed; + $self->batch_prepare; + die 'BUG: already in async' if $self->{inflight}; + $self->{inflight} = []; +} + +sub cat_async ($$$;$) { + my ($self, $oid, $cb, $arg) = @_; + my $inflight = $self->{inflight} // cat_async_begin($self); + while (scalar(@$inflight) >= MAX_INFLIGHT) { + cat_async_step($self, $inflight); + } + print { $self->{out} } $oid, "\n" or $self->fail("write error: $!"); + push(@$inflight, $oid, $cb, $arg); +} + +sub async_prefetch { + my ($self, $oid, $cb, $arg) = @_; + if (my $inflight = $self->{inflight}) { + # we could use MAX_INFLIGHT here w/o the halving, + # but lets not allow one client to monopolize a git process + if (scalar(@$inflight) < int(MAX_INFLIGHT/2)) { + print { $self->{out} } $oid, "\n" or + $self->fail("write error: $!"); + return push(@$inflight, $oid, $cb, $arg); + } + } + undef; +} + +sub extract_cmt_time { + my ($bref, undef, undef, undef, $modified) = @_; + + if ($$bref =~ /^committer .*?> ([0-9]+) [\+\-]?[0-9]+/sm) { + my $cmt_time = $1 + 0; + $$modified = $cmt_time if $cmt_time > $$modified; + } +} + +# returns the modified time of a git repo, same as the "modified" field +# of a grokmirror manifest +sub modified ($) { + my ($self) = @_; + my $modified = 0; + my $fh = popen($self, qw(rev-parse --branches)); + local $/ = "\n"; + while (my $oid = <$fh>) { + chomp $oid; + cat_async($self, $oid, \&extract_cmt_time, \$modified); + } + cat_async_wait($self); + $modified || time; +} + +# for grokmirror, which doesn't read gitweb.description +# templates/hooks--update.sample and git-multimail in git.git +# only match "Unnamed repository", not the full contents of +# templates/this--description in git.git +sub manifest_entry { + my ($self, $epoch, $default_desc) = @_; + my ($fh, $pid) = $self->popen('show-ref'); + my $dig = Digest::SHA->new(1); + while (read($fh, my $buf, 65536)) { + $dig->add($buf); + } + close $fh; + waitpid($pid, 0); + return if $?; # empty, uninitialized git repo + my $git_dir = $self->{git_dir}; + my $ent = { + fingerprint => $dig->hexdigest, + reference => undef, + modified => modified($self), + }; + chomp(my $owner = $self->qx('config', 'gitweb.owner')); + utf8::decode($owner); + $ent->{owner} = $owner eq '' ? undef : $owner; + my $desc = ''; + if (open($fh, '<', "$git_dir/description")) { + local $/ = "\n"; + chomp($desc = <$fh>); + utf8::decode($desc); + } + $desc = 'Unnamed repository' if $desc eq ''; + if (defined $epoch && $desc =~ /\AUnnamed repository/) { + $desc = "$default_desc [epoch $epoch]"; + } + $ent->{description} = $desc; + if (open($fh, '<', "$git_dir/objects/info/alternates")) { + # n.b.: GitPython doesn't seem to handle comments or C-quoted + # strings like native git does; and we don't for now, either. + local $/ = "\n"; + chomp(my @alt = <$fh>); + + # grokmirror only supports 1 alternate for "reference", + if (scalar(@alt) == 1) { + my $objdir = "$git_dir/objects"; + my $ref = File::Spec->rel2abs($alt[0], $objdir); + $ref =~ s!/[^/]+/?\z!!; # basename + $ent->{reference} = $ref; + } + } + $ent; +} + 1; __END__ =pod