(big-config) the new "big-config" for large setups

If you have many thousands of repos and users, neatly organised into
groups, etc., the normal gitolite fails.  (It actually runs out of
memory very fast while doing the "compile" when you push the config, due
to the number of combinations of repo/user being stored in the hash!)

This commit series will stop doing that if you set $GL_BIG_CONFIG = 1 in
the rc file.

Some notes:

  - deny rules will still work but somewhat differently -- now they must
    be placed all together in one place to work like before.  Ask me for
    details if you need to know before I get done with the docs

  - I've tested most of the important features, but not every single
    nuance

  - the update hook may be a tad less efficient now; we can try and
    tweak it later if needed but it shouldn't really hurt anything
    significantly even now

  - docs have not been written yet

conf/example.gitolite.rc

@@ -1,4 +1,5 @@
 # paths and configuration variables for gitolite
+$GL_BIG_CONFIG = 1;
 
 # please read comments before editing
 

hooks/common/update

@@ -34,27 +34,12 @@ exit 0 if exists $ENV{GL_BYPASS_UPDATE_HOOK};
 
 # we should already have the GL_RC env var set when we enter this hook
 die "parse $ENV{GL_RC} failed: "       . ($! or $@) unless do $ENV{GL_RC};
-# then "do" the compiled config file, whose name we now know.  Before doing
-# that we setup the creator etc from environment variables so that the parse
-# interpolates them.  We've minimised the duplication but this *does*
-# duplicate a bit of parse_acl from gitolite.pm; we don't want to include that
-# file here just for that little bit
-{
-    our $creator = $ENV{GL_CREATOR};
-    our $readers = $ENV{GL_READERS};
-    our $writers = $ENV{GL_WRITERS};
-    our $gl_user = $ENV{GL_USER};
 
-    die "parse $GL_CONF_COMPILED failed: " . ($! or $@) unless do $GL_CONF_COMPILED;
-
-    $repos{$ENV{GL_REPO}} = $repos{$ENV{GL_REPOPATT}} if ( $ENV{GL_REPOPATT} );
-}
-my $reported_repo = $ENV{GL_REPO} . ( $ENV{GL_REPOPATT} ? " ($ENV{GL_REPOPATT})" : "" );
-
-# we've started to need some common subs in what used to be a small, cute,
-# little script that barely spanned a few lines :(
 require "$ENV{GL_BINDIR}/gitolite.pm";
 
+my ($perm, $creator, $wild) = &repo_rights($ENV{GL_REPO});
+my $reported_repo = $ENV{GL_REPO} . ( $ENV{GL_REPOPATT} ? " ($ENV{GL_REPOPATT})" : "" );
+
 # ----------------------------------------------------------------------------
 #       start...
 # ----------------------------------------------------------------------------
@@ -71,17 +56,17 @@ chomp($merge_base = `git merge-base $oldsha $newsha`)
     unless $oldsha eq '0' x 40
         or $newsha eq '0' x 40;
 
-# what are you trying to do?  (is it 'W' or '+'?)
+# att_acc == attempted access -- what are you trying to do?  (is it 'W' or '+'?)
-my $perm = 'W';
+my $att_acc = 'W';
 # rewriting a tag is considered a rewind, in terms of permissions
-$perm = '+' if $ref =~ m(refs/tags/) and $oldsha ne ('0' x 40);
+$att_acc = '+' if $ref =~ m(refs/tags/) and $oldsha ne ('0' x 40);
 # non-ff push to ref
 # notice that ref delete looks like a rewind, as it should
-$perm = '+' if $oldsha ne $merge_base;
+$att_acc = '+' if $oldsha ne $merge_base;
 
 # were any 'D' perms specified?  If they were, it means we have to separate
 # deletes from rewinds, so if the new sha is all 0's, change the '+' to a 'D'
-$perm = 'D' if ( $repos{$ENV{GL_REPO}}{DELETE_IS_D} or $repos{'@all'}{DELETE_IS_D} ) and $newsha eq '0' x 40;
+$att_acc = 'D' if ( $repos{$ENV{GL_REPO}}{DELETE_IS_D} or $repos{'@all'}{DELETE_IS_D} ) and $newsha eq '0' x 40;
 
 my @allowed_refs;
 # @all repos: see comments in similar code in check_access
@@ -113,12 +98,12 @@ if (exists $repos{$ENV{GL_REPO}}{NAME_LIMITS}) {
 # log is the *first* one (which is a *real* ref, like refs/heads/master),
 # while all the rest (if they exist) are like NAME/something.  So we do the
 # first one separately to capture it, then run the rest (if any)
-my $log_refex = check_ref(\@allowed_refs, $ENV{GL_REPO}, (shift @refs), $perm);
+my $log_refex = check_ref(\@allowed_refs, $ENV{GL_REPO}, (shift @refs), $att_acc);
-&check_ref               (\@allowed_refs, $ENV{GL_REPO}, $_           , $perm) for @refs;
+&check_ref               (\@allowed_refs, $ENV{GL_REPO}, $_           , $att_acc) for @refs;
 
 # if we returned at all, all the checks succeeded, so we log the action and exit 0
 
-&log_it("$ENV{GL_TS}  $perm\t" .
+&log_it("$ENV{GL_TS}  $att_acc\t" .
     substr($oldsha, 0, 14) . "\t" . substr($newsha, 0, 14) .
     "\t$reported_repo\t$ref\t$ENV{GL_USER}\t$log_refex\n");
 

src/gitolite.pm

@@ -1,4 +1,7 @@
 use strict;
+use Data::Dumper;
+$Data::Dumper::Deepcopy = 1;
+
 # this file is commonly used using "require".  It is not required to use "use"
 # (because it doesn't live in a different package)
 
@@ -34,8 +37,9 @@ our $USERNAME_PATT=qr(^\@?[0-9a-zA-Z][0-9a-zA-Z._\@+-]*$);  # very simple patter
 our $REPOPATT_PATT=qr(^\@?[0-9a-zA-Z][\\^.$|()[\]*+?{}0-9a-zA-Z._\@/-]*$);
 
 # these come from the RC file
-our ($REPO_UMASK, $GL_WILDREPOS, $GL_PACKAGE_CONF, $GL_PACKAGE_HOOKS, $REPO_BASE, $GL_CONF_COMPILED);
+our ($REPO_UMASK, $GL_WILDREPOS, $GL_PACKAGE_CONF, $GL_PACKAGE_HOOKS, $REPO_BASE, $GL_CONF_COMPILED, $GL_BIG_CONFIG);
 our %repos;
+our %groups;
 
 # ----------------------------------------------------------------------------
 #       convenience subs
@@ -253,24 +257,39 @@ sub parse_acl
     # want the config dumped as is, really
     return unless $repo;
 
-    # return with "no wildcard match" status if you found the actual repo in
+    my ($wild, @repo_plus, @user_plus);
-    # the config or if wild is unset
+    # expand $repo and $gl_user into all possible matching values
-    return $ENV{GL_REPOPATT} = "" if $repos{$repo} or not $GL_WILDREPOS;
+    ($wild, @repo_plus) = &get_memberships($repo,    1);
+    (       @user_plus) = &get_memberships($gl_user, 0);
+    # XXX testing notes: the above should return just one entry during
+    # non-BC usage, whether wild or not
+    die "assert 1 failed" if (@repo_plus > 1 and $repo_plus[-1] ne '@all'
+                          or  @repo_plus > 2) and not $GL_BIG_CONFIG;
 
-    # didn't find actual repo in %repos, and wild is set, so find the repo
+    # the old "convenience copy" thing.  Now on steroids :)
-    # pattern that matches the actual repo
-    my @matched = grep { $repo =~ /^$_$/ } sort keys %repos;
 
-    # didn't find a match?  avoid leaking info to user about repo existence;
+    # note that when copying the @all entry, we retain the destination name as
-    # as before, pretend "no wildcard match" status
+    # @all; we dont change it to $repo or $gl_user
-    return $ENV{GL_REPOPATT} = "" unless @matched;
+    for my $r ('@all', @repo_plus) {
+        my $dr = $repo; $dr = '@all' if $r eq '@all';
+        $repos{$dr}{DELETE_IS_D} = 1 if $repos{$r}{DELETE_IS_D};
+        $repos{$dr}{NAME_LIMITS} = 1 if $repos{$r}{NAME_LIMITS};
 
-    die "$repo has multiple matches\n@matched\n" if @matched > 1;
+        for my $u ('@all', @user_plus) {
+            my $du = $gl_user; $du = '@all' if $u eq '@all';
+            $repos{$dr}{C}{$du} = 1 if $repos{$r}{C}{$u};
+            $repos{$dr}{R}{$du} = 1 if $repos{$r}{R}{$u};
+            $repos{$dr}{W}{$du} = 1 if $repos{$r}{W}{$u};
 
-    # found exactly one pattern that matched, copy its ACL for convenience
+            next if $r eq $dr and $u eq $du;    # no point duplicating those refexes
-    $repos{$repo} = $repos{$matched[0]};
+            push @{ $repos{$dr}{$du} }, @{ $repos{$r}{$u} }
-    # and return the pattern
+                if exists $repos{$r}{$u} and ref($repos{$r}{$u}) eq 'ARRAY';
-    return $ENV{GL_REPOPATT} = $matched[0];
+        }
+    }
+
+    $ENV{GL_REPOPATT} = "";
+    $ENV{GL_REPOPATT} = $wild if $wild and $GL_WILDREPOS;
+    return ($wild);
 }
 
 # ----------------------------------------------------------------------------
@@ -295,6 +314,11 @@ sub report_basic
     &report_version($GL_ADMINDIR, $user);
     print "\rthe gitolite config gives you the following access:\r\n";
     for my $r (sort keys %repos) {
+        if ($r =~ $REPONAME_PATT) {
+            &parse_acl($GL_CONF_COMPILED, $r, "NOBODY",      "NOBODY", "NOBODY");
+        } else {
+            &parse_acl($GL_CONF_COMPILED, $r, $ENV{GL_USER}, "NOBODY", "NOBODY");
+        }
         # @all repos; meaning of read/write flags:
         # @R => @all users are allowed access to this repo
         # #R => you're a super user and can see @all repos
@@ -331,7 +355,7 @@ sub expand_wild
         # actual_repo has to match the pattern being expanded
         next unless $actual_repo =~ /$repo/;
 
-        my($perm, $creator) = &repo_rights($actual_repo);
+        my($perm, $creator, $wild) = &repo_rights($actual_repo);
         next unless $perm =~ /\S/;
         print "$perm\t$creator\t$actual_repo\n";
     }
@@ -342,64 +366,67 @@ sub expand_wild
 # how/why).  Regardless of how we're called, we assume $ENV{GL_USER} is
 # already defined
 {
-    my %normal_repos;
+    my $last_repo = '';
-
     sub repo_rights {
         my $repo = shift;
         $repo =~ s/^\.\///;
         $repo =~ s/\.git$//;
 
+        return if $last_repo eq $repo;      # a wee bit o' caching, though not yet needed
+
         # we get passed an actual repo name.  It may be a normal
         # (non-wildcard) repo, in which case it is assumed to exist.  If it's
         # a wildrepo, it may or may not exist.  If it doesn't exist, the "C"
         # perms are also filled in, else that column is left blank
 
-        unless (%normal_repos) {
         unless ($REPO_BASE) {
-                # means we've been called from outside
+            # means we've been called from outside; see doc/admin-defined-commands.mkd
             &where_is_rc();
             die "parse $ENV{GL_RC} failed: "       . ($! or $@) unless do $ENV{GL_RC};
         }
 
-            &parse_acl($GL_CONF_COMPILED, "", "NOBODY", "NOBODY", "NOBODY");
-            %normal_repos = %repos;
-        }
-
-        my $creator;
         my $perm = '   ';
+        my $creator;
 
-        # if repo is present "as is" in the config, those permissions will
+        # get basic info about the repo and fill %repos
-        # override anything inherited from a wildcard that may have matched
+        my $wild = '';
-        if ($normal_repos{$repo}) {
+        my $exists = -d "$ENV{GL_REPO_BASE_ABS}/$repo.git";
-            %repos = %normal_repos;
+        if ($exists) {
-            $creator = '<gitolite>';
+            # these will be empty if it's not a wildcard repo anyway
-        } elsif ( -d "$ENV{GL_REPO_BASE_ABS}/$repo.git" ) {
-            # must be a wildrepo, and it has already been created; find the
-            # creator and subsitute in repos
             my ($read, $write);
             ($creator, $read, $write) = &wild_repo_rights($ENV{GL_REPO_BASE_ABS}, $repo, $ENV{GL_USER});
             # get access list with these substitutions
-            &parse_acl($GL_CONF_COMPILED, $repo, $creator || "NOBODY", $read || "NOBODY", $write || "NOBODY");
+            $wild = &parse_acl($GL_CONF_COMPILED, $repo, $creator || "NOBODY", $read || "NOBODY", $write || "NOBODY");
+        } else {
+            $wild = &parse_acl($GL_CONF_COMPILED, $repo, $ENV{GL_USER}, "NOBODY", "NOBODY");
+        }
+
+        if ($exists and not $wild) {
+            $creator = '<gitolite>';
+        } elsif ($exists) {
+            # is a wildrepo, and it has already been created
             $creator = "($creator)";
         } else {
-            # repo didn't exist; C perms also need to be filled in after
+            # repo didn't exist; C perms need to be filled in
-            # getting access list with only creator filled in
-            &parse_acl($GL_CONF_COMPILED, $repo, $ENV{GL_USER}, "NOBODY", "NOBODY");
             $perm = ( $repos{$repo}{C}{'@all'} ? ' @C' : ( $repos{$repo}{C}{$ENV{GL_USER}} ? ' =C' : '   ' )) if $GL_WILDREPOS;
             # if you didn't have perms to create it, delete the "convenience"
             # copy of the ACL that parse_acl makes
             delete $repos{$repo} unless $perm =~ /C/;
-            $creator = "<repo_not_found>";
+            $creator = "<notfound>";
         }
         $perm .= ( $repos{$repo}{R}{'@all'} ? ' @R' : ( $repos{'@all'}{R}{$ENV{GL_USER}} ? ' #R' : ( $repos{$repo}{R}{$ENV{GL_USER}} ? '  R' : '   ' )));
         $perm .= ( $repos{$repo}{W}{'@all'} ? ' @W' : ( $repos{'@all'}{W}{$ENV{GL_USER}} ? ' #W' : ( $repos{$repo}{W}{$ENV{GL_USER}} ? '  W' : '   ' )));
-        return($perm, $creator);
+
+        # set up for caching %repos
+        $last_repo = $repo;
+
+        return($perm, $creator, $wild);
     }
 }
 
 # helper/convenience routine to get rights and ownership from a shell command
 sub cli_repo_rights {
-    my ($perm, $creator) = &repo_rights($_[0]);
+    my ($perm, $creator, $wild) = &repo_rights($_[0]);
     $perm =~ s/ /_/g;
     $creator =~ s/^\(|\)$//g;
     print "$perm $creator\n";
@@ -443,6 +470,78 @@ sub special_cmd
     }
 }
 
+# ----------------------------------------------------------------------------
+#       get memberships
+# ----------------------------------------------------------------------------
+
+# given a plain reponame or username, return:
+# - the name itself, plus all the groups it belongs to if $GL_BIG_CONFIG is
+#   set
+# OR
+# - (for repos) if the name itself doesn't exist in the config, a wildcard
+#   matching it, plus all the groups that wildcard belongs to (again if
+#   $GL_BIG_CONFIG is set)
+
+# A name can normally appear (repo example) (user example)
+# - directly (repo foo) (RW = bar)
+# - (only for repos) as a direct wildcard (repo foo/.*)
+# but if $GL_BIG_CONFIG is set, it can also appear:
+# - indirectly (@g = foo; repo @g) (@ug = bar; RW = @ug))
+# - (only for repos) as an indirect wildcard (@g = foo/.*; repo @g).
+# things that may not be obvious from the above:
+# - the wildcard stuff does not apply to username memberships
+# - for repos, wildcard appearances are TOTALLY ignored if a non-wild
+#   appearance (direct or indirect) exists
+
+sub get_memberships {
+    my $base = shift;   # reponame or username
+    my $is_repo = shift;    # some true value means a repo name has been passed
+
+    my $wild = '';
+    my (@ret, @ret_w);      # maintain wild matches separately from non-wild
+
+    # direct
+    push @ret, $base if not $is_repo or exists $repos{$base};
+    if ($is_repo and $GL_WILDREPOS and not @ret) {
+        for my $i (sort keys %repos) {
+            if ($base =~ /^$i$/) {
+                die "$ABRT $base matches $wild AND $i\n" if $wild and $wild ne $i;
+                $wild = $i;
+                # direct wildcard
+                push @ret_w, $i;
+            }
+        }
+    }
+
+    if ($GL_BIG_CONFIG) {
+        for my $g (sort keys %groups) {
+            for my $i (sort keys %{ $groups{$g} }) {
+                if ($base eq $i) {
+                    # indirect
+                    push @ret, $g;
+                } elsif ($is_repo and $GL_WILDREPOS and not @ret and $base =~ /^$i$/) {
+                    die "$ABRT $base matches $wild AND $i\n" if $wild and $wild ne $i;
+                    $wild = $i;
+                    # indirect wildcard
+                    push @ret_w, $g;
+                }
+            }
+        }
+    }
+
+    # deal with returning user info first
+    return (@ret) unless $is_repo;
+
+    # enforce the rule about ignoring all wildcard matches if a non-wild match
+    # exists while returning.  (The @ret gating above does not adequately
+    # ensure this, it is only an optimisation).
+    #
+    # Also note that there is an extra return value when called for repos
+    # (compared to usernames)
+
+    return ((@ret ? '' : $wild), (@ret ? @ret : @ret_w));
+}
+
 # ----------------------------------------------------------------------------
 #       generic check access routine
 # ----------------------------------------------------------------------------

src/gl-auth-command

@@ -28,6 +28,7 @@ our ($GL_LOGT, $GL_CONF_COMPILED, $REPO_BASE, $GIT_PATH, $REPO_UMASK, $GL_ADMIND
 # and these are set by gitolite.pm
 our ($R_COMMANDS, $W_COMMANDS, $REPONAME_PATT, $REPOPATT_PATT);
 our %repos;
+our %groups;
 
 # the common setup module is in the same directory as this running program is
 my $bindir = $0;
@@ -55,6 +56,10 @@ my $repo_base_abs = $ENV{GL_REPO_BASE_ABS} = ( $REPO_BASE =~ m(^/) ? $REPO_BASE
 #       start...
 # ----------------------------------------------------------------------------
 
+# no arguments given?  default user is $USER (fedorahosted works like this,
+# and it is harmless for others)
+@ARGV = ($ENV{USER}) unless @ARGV;
+
 # if the first argument is a "-s", this user is allowed to get a shell using
 # this key
 my $shell_allowed = 0;
@@ -184,7 +189,7 @@ $ENV{GL_REPO}=$repo;
 #       first level permissions check
 # ----------------------------------------------------------------------------
 
-my ($perm, $creator) = &repo_rights($repo);
+my ($perm, $creator, $wild) = &repo_rights($repo);
 if ($perm =~ /C/) {
     # it was missing, and you have create perms
     wrap_chdir("$repo_base_abs");

src/gl-compile-conf

@@ -52,7 +52,7 @@ $Data::Dumper::Sortkeys = 1;
 open STDOUT, ">", "/dev/null" if (@ARGV and shift eq '-q');
 
 # these are set by the "rc" file
-our ($GL_ADMINDIR, $GL_CONF, $GL_KEYDIR, $GL_CONF_COMPILED, $REPO_BASE, $REPO_UMASK, $PROJECTS_LIST, $GIT_PATH, $GL_WILDREPOS, $GL_GITCONFIG_KEYS, $GL_PACKAGE_HOOKS, $GL_SETPERMS_OVERRIDES_CONFIG);
+our ($GL_ADMINDIR, $GL_CONF, $GL_KEYDIR, $GL_CONF_COMPILED, $REPO_BASE, $REPO_UMASK, $PROJECTS_LIST, $GIT_PATH, $GL_WILDREPOS, $GL_GITCONFIG_KEYS, $GL_PACKAGE_HOOKS, $GL_SETPERMS_OVERRIDES_CONFIG, $GL_BIG_CONFIG);
 # and these are set by gitolite.pm
 our ($REPONAME_PATT, $REPOPATT_PATT, $USERNAME_PATT, $AUTH_COMMAND, $AUTH_OPTIONS, $ABRT, $WARN);
 
@@ -181,10 +181,12 @@ sub parse_conf_file
         # repo(s)
         elsif (/^repo (.*)/)
         {
-            # grab the list and expand any @stuff in it
+            # grab the list...
             @repos = split ' ', $1;
             unless (@repos == 1 and $repos[0] eq '@all') {
-                @repos = expand_list ( @repos );
+                # ...expand groups in the default case
+                @repos = expand_list ( @repos ) unless $GL_BIG_CONFIG;
+                # ...sanity check
                 for (@repos) {
                     die "$ABRT bad reponame $_\n"
                         if ($GL_WILDREPOS and $_ !~ $REPOPATT_PATT);
@@ -214,7 +216,7 @@ sub parse_conf_file
 
             # expand the user list, unless it is just "@all"
             @users = expand_list ( @users )
-                unless (@users == 1 and $users[0] eq '@all');
+                unless ($GL_BIG_CONFIG or (@users == 1 and $users[0] eq '@all'));
             do  { die "$ABRT bad username $_\n" unless $_ =~ $USERNAME_PATT } for @users;
 
             s/\bCREAT[EO]R\b/~\$creator/g for @users;
@@ -370,6 +372,7 @@ my $dumped_data = Data::Dumper->Dump([\%repos], [qw(*repos)]);
 # much...
 $dumped_data =~ s/'(?=[^']*\$(?:creator|readers|writers|gl_user))~*(.*?)'/"$1"/g;
 print $compiled_fh $dumped_data;
+print $compiled_fh Data::Dumper->Dump([\%groups], [qw(*groups)]) if $GL_BIG_CONFIG and %groups;
 close $compiled_fh or die "$ABRT close compiled-conf failed: $!\n";
 
 # ----------------------------------------------------------------------------
@@ -406,8 +409,7 @@ wrap_chdir("$repo_base_abs");
 
 for my $repo (sort keys %repos) {
     next unless $repo =~ $REPONAME_PATT;
-    next if $repo =~ m(^EXTCMD/);   # these are not real repos
+    next if $repo =~ m(^\@|EXTCMD/);    # these are not real repos
-    next if $repo eq '@all';
     unless (-d "$repo.git") {
         print STDERR "creating $repo...\n";
         new_repo($repo, "$GL_ADMINDIR/hooks/common");
@@ -457,6 +459,7 @@ wrap_chdir("$repo_base_abs");
 # daemons first...
 for my $repo (sort keys %repos) {
     next unless $repo =~ $REPONAME_PATT;
+    next if $repo =~ m(^\@|EXTCMD/);    # these are not real repos
     my $export_ok = "$repo.git/git-daemon-export-ok";
     if ($repos{$repo}{'R'}{'daemon'}) {
         system("touch $export_ok");
@@ -469,6 +472,7 @@ my %projlist = ();
 # ...then gitwebs
 for my $repo (sort keys %repos) {
     next unless $repo =~ $REPONAME_PATT;
+    next if $repo =~ m(^\@|EXTCMD/);    # these are not real repos
     my $desc_file = "$repo.git/description";
     # note: having a description also counts as enabling gitweb
     if ($repos{$repo}{'R'}{'gitweb'} or $desc{"$repo.git"}) {