Combine adjacent rules differing only in conntrack state match.

Signed-off-by: Tom Eastep <teastep@shorewall.net>
This commit is contained in:
Tom Eastep 2013-02-03 09:03:22 -08:00
parent 61c219ed3a
commit 014b4ddc50

View File

@ -3553,6 +3553,141 @@ sub delete_duplicates {
\@rules; \@rules;
} }
#
# Get the 'conntrack' state for the passed rule reference
#
sub get_conntrack( $ ) {
if ( my $states = $_[0]->{conntrack} ) {
unless ( reftype $states ) {
return $states if $states =~ s/--ctstate //;
}
}
'';
}
#
# Return an array of keys for the passed rule. 'conntrack' and 'comment' are omitted;
#
sub get_keys1( $ ) {
sort grep $_ ne 'conntrack' && $_ ne 'comment', keys %{$_[0]};
}
#
# The arguments are a list of rule references; function returns a similar list with adjacent compatible rules combined
#
# Adjacent rules are compatible if:
#
# - They all specify conntrack match
# - All of the rest of their members are identical with the possible exception of 'comment'.
#
# Adjacent distinct comments are combined, separated by ', '. Redundant adjacent comments are dropped.
#
sub combine_states {
my @rules;
my $rulenum = 1;
my $chainref = shift;
my $baseref = shift;
while ( $baseref ) {
{
my $ruleref;
my $conntrack;
my $basenum = $rulenum;
if ( my $conntrack1 = get_conntrack( $baseref ) ) {
my @keys1 = get_keys1( $baseref );
my @states = ( split ',', $conntrack1 );
my %states;
$states{$_} = 1 for @states;
my $origstates = @states;
my $comment = $baseref->{comment} || '';
my $lastcomment = $comment;
RULE:
while ( ( $ruleref = shift ) ) {
my $conntrack2;
$rulenum++;
if ( $conntrack2 = get_conntrack( $ruleref ) ) {
#
# We have a candidate
#
my $comment2 = $ruleref->{comment} || '';
last if $comment2 ne $lastcomment && length( $comment ) + length( $comment2 ) > 253;
my @keys2 = get_keys1( $ruleref );
last unless @keys1 == @keys2 ;
my $keynum = 0;
for my $key ( @keys1 ) {
last RULE unless $key eq $keys2[$keynum++];
last RULE unless compare_values( $baseref->{$key}, $ruleref->{$key} );
}
#
# The rules connection tracking states must be different; otherwise,
# the rules are identical
#
assert( $conntrack1 ne $conntrack2 );
if ( $comment2 ) {
if ( $comment ) {
$comment .= ", $comment2" unless $comment2 eq $lastcomment;
} else {
$comment = 'Others and ';
last if length( $comment ) + length( $comment2 ) > 255;
$comment .= $comment2;
}
$lastcomment = $comment2;
} else {
if ( $comment ) {
unless ( ( $comment2 = ' and others' ) eq $lastcomment ) {
last if length( $comment ) + length( $comment2 ) > 255;
$comment .= $comment2;
}
}
$lastcomment = $comment2;
}
for ( split ',', $conntrack2 ) {
unless ( $states{$_} ) {
push @states, $_;
$states{$_} = 1;
}
}
trace( $chainref, 'D', $rulenum, $ruleref ) if $debug;
} else {
last;
}
}
if ( @states > $origstates ) {
$baseref->{conntrack} = '--ctstate ' . join( ',', @states );
trace ( $chainref, 'R', $basenum, $baseref ) if $debug;
}
}
push @rules, $baseref;
$baseref = $ruleref ? $ruleref : shift;
}
}
\@rules;
}
sub optimize_level16( $$$ ) { sub optimize_level16( $$$ ) {
my ( $table, $tableref , $passes ) = @_; my ( $table, $tableref , $passes ) = @_;
my @chains = ( grep $_->{referenced}, values %{$tableref} ); my @chains = ( grep $_->{referenced}, values %{$tableref} );
@ -3572,6 +3707,13 @@ sub optimize_level16( $$$ ) {
} }
++$passes; ++$passes;
if ( have_capability 'CONNTRACK_MATCH' ) {
for my $chainref ( @chains ) {
$chainref->{rules} = combine_states( $chainref, @{$chainref->{rules}} );
}
}
} }
# #