X-Git-Url: http://www.git.stargrave.org/?a=blobdiff_plain;f=lib%2FPublicInbox%2FSolverGit.pm;h=8629f0dafeb4216b41ac2c0951c866c8e76c2dcf;hb=55b707d788ce13696e4411389583e720ea6dab01;hp=a13ae9e51fba3d1ddfa13ced17806eb809234f21;hpb=65323f060a3db731bb9fafa004336eeb4bbb8f00;p=public-inbox.git diff --git a/lib/PublicInbox/SolverGit.pm b/lib/PublicInbox/SolverGit.pm index a13ae9e5..8629f0da 100644 --- a/lib/PublicInbox/SolverGit.pm +++ b/lib/PublicInbox/SolverGit.pm @@ -10,12 +10,13 @@ package PublicInbox::SolverGit; use strict; use warnings; -use File::Temp qw(); +use 5.010_001; +use File::Temp 0.19 (); # 0.19 for ->newdir use Fcntl qw(SEEK_SET); use PublicInbox::Git qw(git_unquote git_quote); -use PublicInbox::Spawn qw(spawn popen_rd); use PublicInbox::MsgIter qw(msg_iter msg_part_text); use PublicInbox::Qspawn; +use PublicInbox::Tmpfile; use URI::Escape qw(uri_escape_utf8); # POSIX requires _POSIX_ARG_MAX >= 4096, and xargs is required to @@ -38,38 +39,41 @@ my $MAX_PATCH = 9999; # oid_a => abbreviated pre-image oid, # oid_b => abbreviated post-image oid, # tmp => anonymous file handle with the diff, -# hdr_lines => arrayref of various header lines for mode information +# hdr_lines => string of various header lines for mode information # mode_a => original mode of oid_a (string, not integer), # ibx => PublicInbox::Inbox object containing the diff # smsg => PublicInbox::SearchMsg object containing diff # path_a => pre-image path # path_b => post-image path +# n => numeric path of the patch (relative to worktree) # } -# don't bother if somebody sends us a patch with these path components, -# it's junk at best, an attack attempt at worse: -my %bad_component = map { $_ => 1 } ('', '.', '..'); - sub dbg ($$) { print { $_[0]->{out} } $_[1], "\n" or ERR($_[0], "print(dbg): $!"); } +sub done ($$) { + my ($self, $res) = @_; + my $ucb = delete($self->{user_cb}) or return; + $ucb->($res, $self->{uarg}); +} + sub ERR ($$) { my ($self, $err) = @_; print { $self->{out} } $err, "\n"; - my $ucb = delete($self->{user_cb}); - eval { $ucb->($err) } if $ucb; + eval { done($self, $err) }; die $err; } -# look for existing blobs already in git repos +# look for existing objects already in git repos sub solve_existing ($$) { my ($self, $want) = @_; my $oid_b = $want->{oid_b}; + my $have_hints = scalar keys %$want > 1; my @ambiguous; # Array of [ git, $oids] foreach my $git (@{$self->{gits}}) { my ($oid_full, $type, $size) = $git->check($oid_b); - if (defined($type) && $type eq 'blob') { + if (defined($type) && (!$have_hints || $type eq 'blob')) { return [ $git, $oid_full, $type, int($size) ]; } @@ -84,21 +88,20 @@ sub solve_existing ($$) { # push @ambiguous, [ $git, @oids ]; dbg($self, "`$oid_b' ambiguous in " . - join("\n\t", $git->pub_urls) . "\n" . + join("\n\t", $git->pub_urls($self->{psgi_env})) + . "\n" . join('', map { "$_ blob\n" } @oids)); } scalar(@ambiguous) ? \@ambiguous : undef; } -sub extract_diff ($$$$$) { - my ($self, $p, $re, $ibx, $smsg) = @_; +sub extract_diff ($$) { + my ($p, $arg) = @_; + my ($self, $diffs, $pre, $post, $ibx, $smsg) = @$arg; my ($part) = @$p; # ignore $depth and @idx; - my $hdr_lines; # diff --git a/... b/... - my $tmp; my $ct = $part->content_type || 'text/plain'; my ($s, undef) = msg_part_text($part, $ct); defined $s or return; - my $di = {}; # Email::MIME::Encodings forces QP to be CRLF upon decoding, # change it back to LF: @@ -107,66 +110,88 @@ sub extract_diff ($$$$$) { $s =~ s/\r\n/\n/sg; } - foreach my $l (split(/^/m, $s)) { - if ($l =~ $re) { - $di->{oid_a} = $1; - $di->{oid_b} = $2; - if (defined($3)) { - my $mode_a = $3; - if ($mode_a =~ /\A(?:100644|120000|100755)\z/) { - $di->{mode_a} = $mode_a; - } - } - - - # start writing the diff out to a tempfile - my $pn = ++$self->{tot}; - open($tmp, '>', $self->{tmp}->dirname . "/$pn") or - die "open(tmp): $!"; - - push @$hdr_lines, $l; - $di->{hdr_lines} = $hdr_lines; - utf8::encode($_) for @$hdr_lines; - print $tmp @$hdr_lines or die "print(tmp): $!"; - - # for debugging/diagnostics: - $di->{ibx} = $ibx; - $di->{smsg} = $smsg; - } elsif ($l =~ m!\Adiff --git ("?[^/]+/.+) ("?[^/]+/.+)$!) { - last if $tmp; # got our blob, done! - - my ($path_a, $path_b) = ($1, $2); - - # diff header lines won't have \r because git - # will quote them, but Email::MIME gives CRLF - # for quoted-printable: - $path_b =~ tr/\r//d; - - # don't care for leading 'a/' and 'b/' - my (undef, @a) = split(m{/}, git_unquote($path_a)); - my (undef, @b) = split(m{/}, git_unquote($path_b)); - - # get rid of path-traversal attempts and junk patches: - foreach (@a, @b) { - return if $bad_component{$_}; - } - - $di->{path_a} = join('/', @a); - $di->{path_b} = join('/', @b); - $hdr_lines = [ $l ]; - } elsif ($tmp) { - utf8::encode($l); - print $tmp $l or die "print(tmp): $!"; - } elsif ($hdr_lines) { - push @$hdr_lines, $l; - if ($l =~ /\Anew file mode (100644|120000|100755)$/) { - $di->{mode_a} = $1; - } - } - } - return undef unless $tmp; + state $LF = qr!\r?\n!; + state $ANY = qr![^\r\n]+!; + state $MODE = '100644|120000|100755'; + state $FN = qr!(?:("?[^/\n]+/[^\r\n]+)|/dev/null)!; + + $s =~ m!( # $1 start header lines we save for debugging: + + # everything before ^index is optional, but we don't + # want to match ^(old|copy|rename|deleted|...) unless + # we match /^diff --git/ first: + (?: # begin optional stuff: + + # try to get the pre-and-post filenames as $2 and $3 + (?:^diff\x20--git\x20$FN\x20$FN$LF) + + (?:^(?: # pass all this to git-apply: + # old mode $4 + (?:old\x20mode\x20($MODE)) + | + # new mode (possibly new file) ($5) + (?:new\x20(?:file\x20)?mode\x20($MODE)) + | + (?:(?:copy|rename|deleted| + dissimilarity|similarity)$ANY) + )$LF)* + + )? # end of optional stuff, everything below is required + + # match the pre and post-image OIDs as $6 $7 + ^index\x20(${pre}[a-f0-9]*)\.\.(${post}[a-f0-9]*) + # mode if unchanged $8 + (?:\x20(100644|120000|100755))?$LF + ) # end of header lines ($1) + ( # $9 is the patch body + # "--- a/foo.c" sets pre-filename ($10) in case + # $2 is missing + (?:^---\x20$FN$LF) + + # "+++ b/foo.c" sets post-filename ($11) in case + # $3 is missing + (?:^\+{3}\x20$FN$LF) + + # the meat of the diff, including "^\\No newline ..." + # We also allow for totally blank lines w/o leading spaces, + # because git-apply(1) handles that case, too + (?:^(?:[\@\+\x20\-\\][^\n]*|)$LF)+ + )!smx or return; + + my $di = { + hdr_lines => $1, + oid_a => $6, + oid_b => $7, + mode_a => $5 // $8 // $4, # new (file) // unchanged // old + }; + my $path_a = $2 // $10; + my $path_b = $3 // $11; + my $patch = $9; + + # don't care for leading 'a/' and 'b/' + my (undef, @a) = split(m{/}, git_unquote($path_a)); + my (undef, @b) = split(m{/}, git_unquote($path_b)); + + # get rid of path-traversal attempts and junk patches: + # it's junk at best, an attack attempt at worse: + state $bad_component = { map { $_ => 1 } ('', '.', '..') }; + foreach (@a, @b) { return if $bad_component->{$_} } + + $di->{path_a} = join('/', @a); + $di->{path_b} = join('/', @b); + + my $path = ++$self->{tot}; + $di->{n} = $path; + open(my $tmp, '>:utf8', $self->{tmp}->dirname . "/$path") or + die "open(tmp): $!"; + print $tmp $di->{hdr_lines}, $patch or die "print(tmp): $!"; close $tmp or die "close(tmp): $!"; - $di; + + # for debugging/diagnostics: + $di->{ibx} = $ibx; + $di->{smsg} = $smsg; + + push @$diffs, $di; } sub path_searchable ($) { defined($_[0]) && $_[0] =~ m!\A[\w/\. \-]+\z! } @@ -178,7 +203,7 @@ sub filename_query ($) { join('', map { qq( dfn:"$_") } split(/\.\./, $_[0])); } -sub find_extract_diff ($$$) { +sub find_extract_diffs ($$$) { my ($self, $ibx, $want) = @_; my $srch = $ibx->search or return; @@ -204,16 +229,25 @@ sub find_extract_diff ($$$) { } my $msgs = $srch->query($q, { relevance => 1 }); - my $re = qr/\Aindex ($pre[a-f0-9]*)\.\.($post[a-f0-9]*)(?: (\d+))?/; - my $di; + my $diffs = []; foreach my $smsg (@$msgs) { $ibx->smsg_mime($smsg) or next; - msg_iter(delete($smsg->{mime}), sub { - $di ||= extract_diff($self, $_[0], $re, $ibx, $smsg); - }); - return $di if $di; + my $mime = delete $smsg->{mime}; + msg_iter($mime, \&extract_diff, + [$self, $diffs, $pre, $post, $ibx, $smsg]); + } + @$diffs ? $diffs : undef; +} + +sub update_index_result ($$) { + my ($bref, $self) = @_; + my ($qsp, $msg) = delete @$self{qw(-qsp -msg)}; + if (my $err = $qsp->{err}) { + ERR($self, "git update-index error: $err"); } + dbg($self, $msg); + next_step($self); # onto do_git_apply } sub prepare_index ($) { @@ -232,26 +266,21 @@ sub prepare_index ($) { my $oid_full = $existing->[1]; my $path_a = $di->{path_a} or die "BUG: path_a missing for $oid_full"; - my $mode_a = $di->{mode_a} || extract_old_mode($di); + my $mode_a = $di->{mode_a} // '100644'; - open my $in, '+>', undef or die "open: $!"; + my $in = tmpfile("update-index.$oid_full") or die "tmpfile: $!"; print $in "$mode_a $oid_full\t$path_a\0" or die "print: $!"; $in->flush or die "flush: $!"; sysseek($in, 0, 0) or die "seek: $!"; dbg($self, 'preparing index'); - my $rdr = { 0 => fileno($in) }; + my $rdr = { 0 => $in }; my $cmd = [ qw(git update-index -z --index-info) ]; my $qsp = PublicInbox::Qspawn->new($cmd, $self->{git_env}, $rdr); - $qsp->psgi_qx($self->{psgi_env}, undef, sub { - my ($bref) = @_; - if (my $err = $qsp->{err}) { - ERR($self, "git update-index error: $err"); - } - dbg($self, "index prepared:\n" . - "$mode_a $oid_full\t" . git_quote($path_a)); - next_step($self); # onto do_git_apply - }); + $path_a = git_quote($path_a); + $self->{-qsp} = $qsp; + $self->{-msg} = "index prepared:\n$mode_a $oid_full\t$path_a"; + $qsp->psgi_qx($self->{psgi_env}, undef, \&update_index_result, $self); } # pure Perl "git init" @@ -293,36 +322,26 @@ EOF prepare_index($self); } -sub extract_old_mode ($) { - my ($di) = @_; - if (join('', @{$di->{hdr_lines}}) =~ - /^old mode (100644|100755|120000)\b/) { - return $1; - } - '100644'; -} - -sub do_finish ($$) { - my ($self, $user_cb) = @_; - my $found = $self->{found}; - my $oid_want = $self->{oid_want}; +sub do_finish ($) { + my ($self) = @_; + my ($found, $oid_want) = @$self{qw(found oid_want)}; if (my $exists = $found->{$oid_want}) { - return $user_cb->($exists); + return done($self, $exists); } # let git disambiguate if oid_want was too short, # but long enough to be unambiguous: my $tmp_git = $self->{tmp_git}; if (my @res = $tmp_git->check($oid_want)) { - return $user_cb->($found->{$res[0]}); + return done($self, $found->{$res[0]}); } if (my $err = $tmp_git->last_check_err) { dbg($self, $err); } - $user_cb->(undef); + done($self, undef); } -sub do_step ($) { +sub event_step ($) { my ($self) = @_; eval { # step 1: resolve blobs to patches in the todo queue @@ -343,8 +362,8 @@ sub do_step ($) { # our result: (which may be undef) # Other steps may call user_cb to terminate prematurely # on error - } elsif (my $user_cb = delete($self->{user_cb})) { - do_finish($self, $user_cb); + } elsif (exists $self->{user_cb}) { + do_finish($self); } else { die 'about to call user_cb twice'; # Oops :x } @@ -353,23 +372,17 @@ sub do_step ($) { if ($err) { $err =~ s/^\s*Exception:\s*//; # bad word to show users :P dbg($self, "E: $err"); - my $ucb = delete($self->{user_cb}); - eval { $ucb->($err) } if $ucb; + eval { done($self, $err) }; } } -sub step_cb ($) { - my ($self) = @_; - sub { do_step($self) }; -} - sub next_step ($) { my ($self) = @_; # if outside of public-inbox-httpd, caller is expected to be - # looping step_cb, anyways + # looping event_step, anyways my $async = $self->{psgi_env}->{'pi-httpd.async'} or return; # PublicInbox::HTTPD::Async->new - $async->(undef, step_cb($self)); + $async->(undef, undef, $self); } sub mark_found ($$$) { @@ -383,8 +396,9 @@ sub mark_found ($$$) { } } -sub parse_ls_files ($$$$) { - my ($self, $qsp, $bref, $di) = @_; +sub parse_ls_files ($$) { + my ($self, $bref) = @_; + my ($qsp, $di) = delete @$self{qw(-qsp -cur_di)}; if (my $err = $qsp->{err}) { die "git ls-files error: $err"; } @@ -410,15 +424,53 @@ sub parse_ls_files ($$$$) { next_step($self); # onto the next patch } -sub start_ls_files ($$) { - my ($self, $di) = @_; - my $cmd = [qw(git ls-files -s -z)]; - my $qsp = PublicInbox::Qspawn->new($cmd, $self->{git_env}); - $qsp->psgi_qx($self->{psgi_env}, undef, sub { - my ($bref) = @_; - eval { parse_ls_files($self, $qsp, $bref, $di) }; - ERR($self, $@) if $@; - }); +sub ls_files_result { + my ($bref, $self) = @_; + eval { parse_ls_files($self, $bref) }; + ERR($self, $@) if $@; +} + +sub oids_same_ish ($$) { + (index($_[0], $_[1]) == 0) || (index($_[1], $_[0]) == 0); +} + +sub skip_identical ($$$) { + my ($self, $patches, $cur_oid_b) = @_; + while (my $nxt = $patches->[0]) { + if (oids_same_ish($cur_oid_b, $nxt->{oid_b})) { + dbg($self, 'skipping '.di_url($self, $nxt). + " for $cur_oid_b"); + shift @$patches; + } else { + return; + } + } +} + +sub apply_result ($$) { + my ($bref, $self) = @_; + my ($qsp, $di) = delete @$self{qw(-qsp -cur_di)}; + dbg($self, $$bref); + my $patches = $self->{patches}; + if (my $err = $qsp->{err}) { + my $msg = "git apply error: $err"; + my $nxt = $patches->[0]; + if ($nxt && oids_same_ish($nxt->{oid_b}, $di->{oid_b})) { + dbg($self, $msg); + dbg($self, 'trying '.di_url($self, $nxt)); + return do_git_apply($self); + } else { + ERR($self, $msg); + } + } else { + skip_identical($self, $patches, $di->{oid_b}); + } + + my @cmd = qw(git ls-files -s -z); + $qsp = PublicInbox::Qspawn->new(\@cmd, $self->{git_env}); + $self->{-cur_di} = $di; + $self->{-qsp} = $qsp; + $qsp->psgi_qx($self->{psgi_env}, undef, \&ls_files_result, $self); } sub do_git_apply ($) { @@ -427,33 +479,30 @@ sub do_git_apply ($) { my $patches = $self->{patches}; # we need --ignore-whitespace because some patches are CRLF - my @cmd = (qw(git -C), $dn, qw(apply --cached --ignore-whitespace - --whitespace=warn --verbose)); + my @cmd = (qw(git apply --cached --ignore-whitespace + --unidiff-zero --whitespace=warn --verbose)); my $len = length(join(' ', @cmd)); my $total = $self->{tot}; my $di; # keep track of the last one for "git ls-files" + my $prv_oid_b; do { my $i = ++$self->{nr}; $di = shift @$patches; dbg($self, "\napplying [$i/$total] " . di_url($self, $di) . - "\n" . join('', @{$di->{hdr_lines}})); - my $path = $total + 1 - $i; + "\n" . $di->{hdr_lines}); + my $path = $di->{n}; $len += length($path) + 1; push @cmd, $path; - } while (@$patches && $len < $ARG_SIZE_MAX); - - my $rdr = { 2 => 1 }; - my $qsp = PublicInbox::Qspawn->new(\@cmd, $self->{git_env}, $rdr); - $qsp->psgi_qx($self->{psgi_env}, undef, sub { - my ($bref) = @_; - dbg($self, $$bref); - if (my $err = $qsp->{err}) { - ERR($self, "git apply error: $err"); - } - eval { start_ls_files($self, $di) }; - ERR($self, $@) if $@; - }); + $prv_oid_b = $di->{oid_b}; + } while (@$patches && $len < $ARG_SIZE_MAX && + !oids_same_ish($patches->[0]->{oid_b}, $prv_oid_b)); + + my $opt = { 2 => 1, -C => $dn, quiet => 1 }; + my $qsp = PublicInbox::Qspawn->new(\@cmd, $self->{git_env}, $opt); + $self->{-cur_di} = $di; + $self->{-qsp} = $qsp; + $qsp->psgi_qx($self->{psgi_env}, undef, \&apply_result, $self); } sub di_url ($$) { @@ -480,11 +529,12 @@ sub resolve_patch ($$) { die "Loop detected solving $cur_want\n"; } if (my $existing = solve_existing($self, $want)) { + my ($found_git, undef, $type, undef) = @$existing; dbg($self, "found $cur_want in " . - join("\n", $existing->[0]->pub_urls)); + join("\n", $found_git->pub_urls($self->{psgi_env}))); - if ($cur_want eq $self->{oid_want}) { # all done! - eval { delete($self->{user_cb})->($existing) }; + if ($cur_want eq $self->{oid_want} || $type ne 'blob') { + eval { done($self, $existing) }; die "E: $@" if $@; return; } @@ -494,15 +544,16 @@ sub resolve_patch ($$) { # scan through inboxes to look for emails which results in # the oid we want: - my $di; foreach my $ibx (@{$self->{inboxes}}) { - $di = find_extract_diff($self, $ibx, $want) or next; + my $diffs = find_extract_diffs($self, $ibx, $want) or next; - unshift @{$self->{patches}}, $di; - dbg($self, "found $cur_want in ".di_url($self, $di)); + unshift @{$self->{patches}}, @$diffs; + dbg($self, "found $cur_want in ". + join("\n\t", map { di_url($self, $_) } @$diffs)); # good, we can find a path to the oid we $want, now # lets see if we need to apply more patches: + my $di = $diffs->[0]; my $src = $di->{oid_a}; unless ($src =~ /\A0+\z/) { @@ -521,18 +572,20 @@ sub resolve_patch ($$) { } dbg($self, "could not find $cur_want"); - eval { delete($self->{user_cb})->(undef) }; # not found! :< + eval { done($self, undef) }; die "E: $@" if $@; } # this API is designed to avoid creating self-referential structures; # so user_cb never references the SolverGit object sub new { - my ($class, $ibx, $user_cb) = @_; + my ($class, $ibx, $user_cb, $uarg) = @_; bless { gits => $ibx->{-repo_objs}, user_cb => $user_cb, + uarg => $uarg, + # -cur_di, -qsp, -msg => temporary fields for Qspawn callbacks # TODO: config option for searching related inboxes inboxes => [ $ibx ], @@ -540,6 +593,7 @@ sub new { } # recreate $oid_want using $hints +# hints keys: path_a, path_b, oid_a # Calls {user_cb} with: [ ::Git object, oid_full, type, size, di (diff_info) ] # with found object, or undef if nothing was found # Calls {user_cb} with a string error on fatal errors @@ -548,7 +602,7 @@ sub solve ($$$$$) { # should we even get here? Probably not, but somebody # could be manually typing URLs: - return (delete $self->{user_cb})->(undef) if $oid_want =~ /\A0+\z/; + return done($self, undef) if $oid_want =~ /\A0+\z/; $self->{oid_want} = $oid_want; $self->{out} = $out; @@ -558,15 +612,14 @@ sub solve ($$$$$) { $self->{todo} = [ { %$hints, oid_b => $oid_want } ]; $self->{patches} = []; # [ $di, $di, ... ] $self->{found} = {}; # { abbr => [ ::Git, oid, type, size, $di ] } - $self->{tmp} = File::Temp->newdir('solver.tmp-XXXXXXXX', TMPDIR => 1); + $self->{tmp} = File::Temp->newdir("solver.$oid_want-XXXXXXXX", TMPDIR => 1); dbg($self, "solving $oid_want ..."); - my $step_cb = step_cb($self); if (my $async = $env->{'pi-httpd.async'}) { # PublicInbox::HTTPD::Async->new - $async->(undef, $step_cb); + $async->(undef, undef, $self); } else { - $step_cb->() while $self->{user_cb}; + event_step($self) while $self->{user_cb}; } }