fetch: support running as root

The "-w" perlop always succeeds as root, so we need to check
st_mode for writability bits to detect directories we shouldn't
write to.

Reported-by: Konstantin Ryabitsev <konstantin@linuxfoundation.org>
Link: https://public-inbox.org/meta/20210927124056.kj5okiefvs4ztk27@meerkat.local/

diff --git a/lib/PublicInbox/Fetch.pm b/lib/PublicInbox/Fetch.pm
index 7881b402e3f6c2b7d870a115b50f30286a029ab4..5ada1f49e4dc28ba39e2c0278208869575e4456a 100644 (file)
--- a/lib/PublicInbox/Fetch.pm
+++ b/lib/PublicInbox/Fetch.pm
@@ -6,7 +6,7 @@ use strict;
 use v5.10.1;
 use parent qw(PublicInbox::IPC);
 use URI ();
-use PublicInbox::Spawn qw(popen_rd run_die);
+use PublicInbox::Spawn qw(popen_rd run_die spawn);
 use PublicInbox::Admin;
 use PublicInbox::LEI;
 use PublicInbox::LeiCurl;
@@ -95,6 +95,13 @@ sub get_fingerprint2 {
        Digest::SHA::sha256(do { local $/; <$rd> });
 }
 
+sub writable_dir ($) {
+       my ($dir) = @_;
+       return unless -d $dir && -w _;
+       my @st = stat($dir);
+       $st[2] & 0222; # any writable bits set? (in case of root)
+}
+
 sub do_fetch { # main entry point
        my ($cls, $lei, $cd) = @_;
        my $ibx_ver;
@@ -112,7 +119,7 @@ sub do_fetch { # main entry point
                my ($git_url, $epoch);
                for my $nr (@epochs) { # try newest epoch, first
                        my $edir = "$dir/git/$nr.git";
-                       unless (-d $edir && -w _) { # must be writable dir
+                       if (!writable_dir($edir)) {
                                $skip->{$nr} = 1;
                                next;
                        }
@@ -122,6 +129,10 @@ sub do_fetch { # main entry point
                                $epoch = $nr;
                        } else {
                                warn "W: $edir missing remote.origin.url\n";
+                               my $pid = spawn([qw(git config -l)], undef,
+                                       { 1 => $lei->{2}, 2 => $lei->{2} });
+                               waitpid($pid, 0);
+                               $lei->child_error($?) if $?;
                        }
                }
                @epochs = grep { !$skip->{$_} } @epochs if $skip;

diff --git a/t/v2mirror.t b/t/v2mirror.t
index 63d17ebfebba9e09c6af0207802d0fdf354bc803..37d64e83e53bff264bbc04a5eb4d5f22889b5d1c 100644 (file)
--- a/t/v2mirror.t
+++ b/t/v2mirror.t
@@ -5,6 +5,7 @@ use v5.10.1;
 use PublicInbox::TestCommon;
 use File::Path qw(remove_tree make_path);
 use Cwd qw(abs_path);
+use Carp ();
 use PublicInbox::Spawn qw(which);
 require_git(2.6);
 require_cmd('curl');
@@ -102,7 +103,9 @@ my @new_epochs;
 my $fetch_each_epoch = sub {
        my %before = map { $_ => 1 } glob("$tmpdir/m/git/*");
        run_script([qw(-fetch --exit-code -q)], undef, {-C => "$tmpdir/m"}) or
-               xbail '-fetch fail';
+               xbail('-fetch fail ',
+                       [ xqx([which('find'), "$tmpdir/m", qw(-type f -ls) ]) ],
+                       Carp::longmess());
        is($?, 0, '--exit-code 0 after fetch updated');
        my @after = grep { !$before{$_} } glob("$tmpdir/m/git/*");
        push @new_epochs, @after;
@@ -273,6 +276,10 @@ if ('test read-only epoch dirs') {
        my @g = glob("$dst/git/*.git");
        my @w = grep { -w $_ } @g;
        my @r = grep { ! -w $_ } @g;
+       if ($> == 0) {
+               @w = grep { (stat($_))[2] & 0200 } @g;
+               @r = grep { !((stat($_))[2] & 0200) } @g;
+       }
        is(scalar(@w), 1, 'one writable directory');
        my ($w) = ($w[0] =~ m!/([0-9]+)\.git\z!);
        is((grep {
@@ -287,7 +294,7 @@ if ('test read-only epoch dirs') {
                        "http://$host:$port/v2/", $dst]);
        my @g2 = glob("$dst/git/*.git") ;
        is_deeply(\@g2, \@g, 'cloned again');
-       is(scalar(grep { -w $_ } @g2), scalar(@w) + 1,
+       is(scalar(grep { (stat($_))[2] & 0200 } @g2), scalar(@w) + 1,
                'got one more cloned epoch');
 
        # make 0.git writable and fetch into it, relies on culled manifest
@@ -377,7 +384,7 @@ EOM
        @cmd = (qw(-clone -q --epoch=~0), "http://$host:$port/v2", $dst);
        run_script(\@cmd, undef, { 2 => \($err = '') });
        is($?, 0, 'partial scraping clone on old PublicInbox::WWW');
-       my @g_last = grep { -w $_ } glob("$dst/git/*.git");
+       my @g_last = grep { (stat($_))[2] & 0200 } glob("$dst/git/*.git");
        is_deeply(\@g_last, [ $g_all[-1] ], 'partial clone of ~0 worked');
 
        chmod(0755, $g_all[0]) or xbail "chmod $!";