# Copyright (C) 2019-2021 all contributors
# License: AGPL-3.0+
# "Solve" blobs which don't exist in git code repositories by
# searching inboxes for post-image blobs.
# this emits a lot of debugging/tracing information which may be
# publicly viewed over HTTP(S). Be careful not to expose
# local filesystem layouts in the process.
package PublicInbox::SolverGit;
use strict;
use v5.10.1;
use File::Temp 0.19 (); # 0.19 for ->newdir
use Fcntl qw(SEEK_SET);
use PublicInbox::Git qw(git_unquote git_quote);
use PublicInbox::MsgIter qw(msg_part_text);
use PublicInbox::Qspawn;
use PublicInbox::Tmpfile;
use PublicInbox::GitAsyncCat;
use PublicInbox::Eml;
use URI::Escape qw(uri_escape_utf8);
# POSIX requires _POSIX_ARG_MAX >= 4096, and xargs is required to
# subtract 2048 bytes. We also don't factor in environment variable
# headroom into this.
use POSIX qw(sysconf _SC_ARG_MAX);
my $ARG_SIZE_MAX = (sysconf(_SC_ARG_MAX) || 4096) - 2048;
my $OID_MIN = 7;
# By default, "git format-patch" generates filenames with a four-digit
# prefix, so that means 9999 patch series are OK, right? :>
# Maybe we can make this configurable, main concern is disk space overhead
# for uncompressed patch fragments. Aside from space, public-inbox-httpd
# is otherwise unaffected by having many patches, here, as it can share
# work fairly. Other PSGI servers may have trouble, though.
my $MAX_PATCH = 9999;
my $LF = qr!\r?\n!;
my $ANY = qr![^\r\n]+!;
my $MODE = '100644|120000|100755';
my $FN = qr!(?:("?[^/\n]+/[^\r\n]+)|/dev/null)!;
my %BAD_COMPONENT = ('' => 1, '.' => 1, '..' => 1);
# di = diff info / a hashref with information about a diff ($di):
# {
# oid_a => abbreviated pre-image oid,
# oid_b => abbreviated post-image oid,
# tmp => anonymous file handle with the diff,
# 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::Smsg object containing diff
# path_a => pre-image path
# path_b => post-image path
# n => numeric path of the patch (relative to worktree)
# }
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 ($$) {
my ($self, $err) = @_;
print { $self->{out} } $err, "\n";
eval { done($self, $err) };
die $err;
}
# look for existing objects already in git repos, returns arrayref
# if found, number of remaining git coderepos to try if not.
sub solve_existing ($$) {
my ($self, $want) = @_;
my $try = $want->{try_gits} //= [ @{$self->{gits}} ]; # array copy
my $git = shift @$try or die 'BUG {try_gits} empty';
my $oid_b = $want->{oid_b};
my ($oid_full, $type, $size) = $git->check($oid_b);
# other than {oid_b, try_gits, try_ibxs}
my $have_hints = scalar keys %$want > 3;
if (defined($type) && (!$have_hints || $type eq 'blob')) {
delete $want->{try_gits};
return [ $git, $oid_full, $type, int($size) ]; # done, success
}
# TODO: deal with 40-char "abbreviations" with future SHA-256 git
return scalar(@$try) if length($oid_b) >= 40;
# parse stderr of "git cat-file --batch-check"
my $err = $git->last_check_err;
my (@oids) = ($err =~ /\b([a-f0-9]{40,})\s+blob\b/g);
return scalar(@$try) unless scalar(@oids);
# TODO: do something with the ambiguous array?
# push @ambiguous, [ $git, @oids ];
dbg($self, "`$oid_b' ambiguous in " .
join("\n\t", $git->pub_urls($self->{psgi_env}))
. "\n" .
join('', map { "$_ blob\n" } @oids));
scalar(@$try);
}
sub extract_diff ($$) {
my ($p, $arg) = @_;
my ($self, $want, $smsg) = @$arg;
my ($part) = @$p; # ignore $depth and @idx;
my $ct = $part->content_type || 'text/plain';
my $post = $want->{oid_b};
my $pre = $want->{oid_a};
if (!defined($pre) || $pre !~ /\A[a-f0-9]+\z/) {
$pre = '[a-f0-9]{7}'; # for RE below
}
# Email::MIME::Encodings forces QP to be CRLF upon decoding,
# change it back to LF:
my $cte = $part->header('Content-Transfer-Encoding') || '';
my ($s, undef) = msg_part_text($part, $ct);
defined $s or return;
delete $part->{bdy};
if ($cte =~ /\bquoted-printable\b/i && $part->crlf eq "\n") {
$s =~ s/\r\n/\n/sg;
}
$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;
undef $s; # free memory
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)) if defined($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:
foreach (@a, @b) { return if $BAD_COMPONENT{$_} }
$di->{path_a} = join('/', @a) if @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): $!";
# for debugging/diagnostics:
$di->{ibx} = $want->{cur_ibx};
$di->{smsg} = $smsg;
push @{$self->{tmp_diffs}}, $di;
}
sub path_searchable ($) { defined($_[0]) && $_[0] =~ m!\A[\w/\. \-]+\z! }
# ".." appears in path names, which confuses Xapian into treating
# it as a range query. So we split on ".." since Xapian breaks
# on punctuation anyways:
sub filename_query ($) {
join('', map { qq( dfn:"$_") } split(/\.\./, $_[0]));
}
sub find_smsgs ($$$) {
my ($self, $ibx, $want) = @_;
my $srch = $ibx->isrch or return;
my $post = $want->{oid_b} or die 'BUG: no {oid_b}';
$post =~ /\A[a-f0-9]+\z/ or die "BUG: oid_b not hex: $post";
my $q = "dfpost:$post";
my $pre = $want->{oid_a};
if (defined $pre && $pre =~ /\A[a-f0-9]+\z/) {
$q .= " dfpre:$pre";
}
my $path_b = $want->{path_b};
if (path_searchable($path_b)) {
$q .= filename_query($path_b);
my $path_a = $want->{path_a};
if (path_searchable($path_a) && $path_a ne $path_b) {
$q .= filename_query($path_a);
}
}
my $mset = $srch->mset($q, { relevance => 1 });
$mset->size ? $srch->mset_to_smsg($ibx, $mset) : 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 ($) {
my ($self) = @_;
my $patches = $self->{patches};
$self->{nr} = 0;
my $di = $patches->[0] or die 'no patches';
my $oid_a = $di->{oid_a} or die '{oid_a} unset';
my $existing = $self->{found}->{$oid_a};
# no index creation for added files
$oid_a =~ /\A0+\z/ and return next_step($self);
die "BUG: $oid_a not found" unless $existing;
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} // '100644';
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, SEEK_SET) or die "seek: $!";
dbg($self, 'preparing index');
my $rdr = { 0 => $in };
my $cmd = [ qw(git update-index -z --index-info) ];
my $qsp = PublicInbox::Qspawn->new($cmd, $self->{git_env}, $rdr);
$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"
sub do_git_init ($) {
my ($self) = @_;
my $dir = $self->{tmp}->dirname;
my $git_dir = "$dir/git";
foreach ('', qw(objects refs objects/info refs/heads)) {
mkdir("$git_dir/$_") or die "mkdir $_: $!";
}
open my $fh, '>', "$git_dir/config" or die "open git/config: $!";
print $fh <<'EOF' or die "print git/config $!";
[core]
repositoryFormatVersion = 0
filemode = true
bare = false
fsyncObjectfiles = false
logAllRefUpdates = false
EOF
close $fh or die "close git/config: $!";
open $fh, '>', "$git_dir/HEAD" or die "open git/HEAD: $!";
print $fh "ref: refs/heads/master\n" or die "print git/HEAD: $!";
close $fh or die "close git/HEAD: $!";
my $f = 'objects/info/alternates';
open $fh, '>', "$git_dir/$f" or die "open: $f: $!";
foreach my $git (@{$self->{gits}}) {
print $fh $git->git_path('objects'),"\n" or die "print $f: $!";
}
close $fh or die "close: $f: $!";
my $tmp_git = $self->{tmp_git} = PublicInbox::Git->new($git_dir);
$tmp_git->{-tmp} = $self->{tmp};
$self->{git_env} = {
GIT_DIR => $git_dir,
GIT_INDEX_FILE => "$git_dir/index",
};
prepare_index($self);
}
sub do_finish ($) {
my ($self) = @_;
my ($found, $oid_want) = @$self{qw(found oid_want)};
if (my $exists = $found->{$oid_want}) {
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 done($self, $found->{$res[0]});
}
if (my $err = $tmp_git->last_check_err) {
dbg($self, $err);
}
done($self, undef);
}
sub event_step ($) {
my ($self) = @_;
eval {
# step 1: resolve blobs to patches in the todo queue
if (my $want = pop @{$self->{todo}}) {
# this populates {patches} and {todo}
resolve_patch($self, $want);
# step 2: then we instantiate a working tree once
# the todo queue is finally empty:
} elsif (!defined($self->{tmp_git})) {
do_git_init($self);
# step 3: apply each patch in the stack
} elsif (scalar @{$self->{patches}}) {
do_git_apply($self);
# step 4: execute the user-supplied callback with
# our result: (which may be undef)
# Other steps may call user_cb to terminate prematurely
# on error
} elsif (exists $self->{user_cb}) {
do_finish($self);
} else {
die 'about to call user_cb twice'; # Oops :x
}
}; # eval
my $err = $@;
if ($err) {
$err =~ s/^\s*Exception:\s*//; # bad word to show users :P
dbg($self, "E: $err");
eval { done($self, $err) };
}
}
sub next_step ($) {
my ($self) = @_;
# if outside of public-inbox-httpd, caller is expected to be
# looping event_step, anyways
my $async = $self->{psgi_env}->{'pi-httpd.async'} or return;
# PublicInbox::HTTPD::Async->new
$async->(undef, undef, $self);
}
sub mark_found ($$$) {
my ($self, $oid, $found_info) = @_;
my $found = $self->{found};
$found->{$oid} = $found_info;
my $oid_cur = $found_info->[1];
while ($oid_cur ne $oid && length($oid_cur) > $OID_MIN) {
$found->{$oid_cur} = $found_info;
chop($oid_cur);
}
}
sub parse_ls_files ($$) {
my ($self, $bref) = @_;
my ($qsp, $di) = delete @$self{qw(-qsp -cur_di)};
if (my $err = $qsp->{err}) {
die "git ls-files error: $err";
}
my ($line, @extra) = split(/\0/, $$bref);
scalar(@extra) and die "BUG: extra files in index: <",
join('> <', @extra), ">";
my ($info, $file) = split(/\t/, $line, 2);
my ($mode_b, $oid_b_full, $stage) = split(/ /, $info);
if ($file ne $di->{path_b}) {
die
"BUG: index mismatch: file=$file != path_b=$di->{path_b}";
}
my $tmp_git = $self->{tmp_git} or die 'no git working tree';
my (undef, undef, $size) = $tmp_git->check($oid_b_full);
defined($size) or die "check $oid_b_full failed";
dbg($self, "index at:\n$mode_b $oid_b_full\t$file");
my $created = [ $tmp_git, $oid_b_full, 'blob', $size, $di ];
mark_found($self, $di->{oid_b}, $created);
next_step($self); # onto the next patch
}
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, $self);
}
sub do_git_apply ($) {
my ($self) = @_;
my $dn = $self->{tmp}->dirname;
my $patches = $self->{patches};
# we need --ignore-whitespace because some patches are CRLF
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" . $di->{hdr_lines});
my $path = $di->{n};
$len += length($path) + 1;
push @cmd, $path;
$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 ($$) {
my ($self, $di) = @_;
# note: we don't pass the PSGI env unconditionally, here,
# different inboxes can have different HTTP_HOST on the same instance.
my $ibx = $di->{ibx};
my $env = $self->{psgi_env} if $ibx eq $self->{inboxes}->[0];
my $url = $ibx->base_url($env);
my $mid = $di->{smsg}->{mid};
defined($url) ? "$url$mid/" : "<$mid>";
}
sub retry_current {
my ($self, $want) = @_;
push @{$self->{todo}}, $want;
next_step($self); # retry solve_existing
}
sub try_harder ($$) {
my ($self, $want) = @_;
# do we have more inboxes to try?
return retry_current($self, $want) if scalar @{$want->{try_ibxs}};
my $cur_want = $want->{oid_b};
if (length($cur_want) > $OID_MIN) { # maybe a shorter OID will work
delete $want->{try_ibxs}; # drop empty arrayref
chop($cur_want);
dbg($self, "retrying $want->{oid_b} as $cur_want");
$want->{oid_b} = $cur_want;
return retry_current($self, $want); # retry with shorter abbrev
}
dbg($self, "could not find $cur_want");
eval { done($self, undef) };
die "E: $@" if $@;
}
sub extract_diffs_done {
my ($self, $want) = @_;
delete $want->{try_smsgs};
delete $want->{cur_ibx};
my $diffs = delete $self->{tmp_diffs};
if (scalar @$diffs) {
unshift @{$self->{patches}}, @$diffs;
dbg($self, "found $want->{oid_b} in " . join(" ||\n\t",
map { di_url($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/) {
# we have to solve it using another oid, fine:
my $job = { oid_b => $src, path_b => $di->{path_a} };
push @{$self->{todo}}, $job;
}
return next_step($self); # onto the next todo item
}
try_harder($self, $want);
}
sub extract_diff_async {
my ($bref, $oid, $type, $size, $x) = @_;
my ($self, $want, $smsg) = @$x;
if (defined($oid)) {
$smsg->{blob} eq $oid or
ERR($self, "BUG: $smsg->{blob} != $oid");
PublicInbox::Eml->new($bref)->each_part(\&extract_diff, $x, 1);
}
scalar(@{$want->{try_smsgs}}) ? retry_current($self, $want)
: extract_diffs_done($self, $want);
}
sub resolve_patch ($$) {
my ($self, $want) = @_;
my $cur_want = $want->{oid_b};
if (scalar(@{$self->{patches}}) > $MAX_PATCH) {
die "Aborting, too many steps to $self->{oid_want}";
}
if (my $msgs = $want->{try_smsgs}) {
my $smsg = shift @$msgs;
if ($self->{psgi_env}->{'pi-httpd.async'}) {
return ibx_async_cat($want->{cur_ibx}, $smsg->{blob},
\&extract_diff_async,
[$self, $want, $smsg]);
} else {
if (my $eml = $want->{cur_ibx}->smsg_eml($smsg)) {
$eml->each_part(\&extract_diff,
[ $self, $want, $smsg ], 1);
}
}
return scalar(@$msgs) ? retry_current($self, $want)
: extract_diffs_done($self, $want);
}
# see if we can find the blob in an existing git repo:
if (!$want->{try_ibxs} && $self->{seen_oid}->{$cur_want}++) {
die "Loop detected solving $cur_want\n";
}
$want->{try_ibxs} //= [ @{$self->{inboxes}} ]; # array copy
my $existing = solve_existing($self, $want);
if (ref $existing) {
my ($found_git, undef, $type, undef) = @$existing;
dbg($self, "found $cur_want in " .
join(" ||\n\t",
$found_git->pub_urls($self->{psgi_env})));
if ($cur_want eq $self->{oid_want} || $type ne 'blob') {
eval { done($self, $existing) };
die "E: $@" if $@;
return;
}
mark_found($self, $cur_want, $existing);
return next_step($self); # onto patch application
} elsif ($existing > 0) {
return retry_current($self, $want);
} else { # $existing == 0: we may retry if inbox scan (below) fails
delete $want->{try_gits};
}
# scan through inboxes to look for emails which results in
# the oid we want:
my $ibx = shift(@{$want->{try_ibxs}}) or die 'BUG: {try_ibxs} empty';
if (my $msgs = find_smsgs($self, $ibx, $want)) {
$want->{try_smsgs} = $msgs;
$want->{cur_ibx} = $ibx;
$self->{tmp_diffs} = [];
return retry_current($self, $want);
}
try_harder($self, $want);
}
# 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, $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 ],
}, $class;
}
# recreate $oid_want using $hints
# hints keys: path_a, path_b, oid_a (note: `oid_b' is NOT a hint)
# 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
sub solve ($$$$$) {
my ($self, $env, $out, $oid_want, $hints) = @_;
# should we even get here? Probably not, but somebody
# could be manually typing URLs:
return done($self, undef) if $oid_want =~ /\A0+\z/;
$self->{oid_want} = $oid_want;
$self->{out} = $out;
$self->{seen_oid} = {};
$self->{tot} = 0;
$self->{psgi_env} = $env;
$self->{todo} = [ { %$hints, oid_b => $oid_want } ];
$self->{patches} = []; # [ $di, $di, ... ]
$self->{found} = {}; # { abbr => [ ::Git, oid, type, size, $di ] }
$self->{tmp} = File::Temp->newdir("solver.$oid_want-XXXX", TMPDIR => 1);
dbg($self, "solving $oid_want ...");
if (my $async = $env->{'pi-httpd.async'}) {
# PublicInbox::HTTPD::Async->new
$async->(undef, undef, $self);
} else {
event_step($self) while $self->{user_cb};
}
}
1;