detect-vuln: test on expanded pattern space

src/detect/detect-vuln.pl

@@ -57,48 +57,65 @@
 my $limitTime = (defined $query->{timeLimit}) ? "timeout $query->{timeLimit}s" : "";
 my $ulimitMemory = (defined $query->{memoryLimit}) ? "ulimit -m $memoryLimitInBytes; ulimit -v $memoryLimitInBytes;" : "";
 
-# Run each detector. Can re-use the input file.
+my @patternsToTry = &expandPatternSpaceForDetectors($query->{pattern});
+
+# This will contain N_DETECTORS * scalar(@patternsToTry) opinions.
 my @detectorOpinions;
-&log("Applying detectors to pattern /$query->{pattern}/");
-for my $d (@DETECTORS) {
-  &log("Querying detector $d->{name}");
-  my $t0 = [gettimeofday];
-  my $stderrFile = "/tmp/detect-vuln-$$-stderr";
-  my ($rc, $out) = &cmd("$ulimitMemory $limitTime $d->{driver} $patternFile 2>$stderrFile");
-  my $elapsed = tv_interval($t0);
-  chomp $out;
-
-  # Clean up in case there was a timeout.
-  my $stderr = &readFile("file"=>$stderrFile);
-  my @filesToClean = ($stderr =~ m/CLEANUP: (\S+)/g);
-  &log("Cleaning up @filesToClean");
-  unlink @filesToClean;
-  unlink $stderrFile;
-
-  my $opinion = { "name"         => $d->{name},
-                  "secToDecide" => sprintf("%.4f", $elapsed),
-                };
-
-  if ($rc eq 124) {
-    &log("Detector $d->{name} timed out");
-    $opinion->{hasOpinion} = 0;
-    $opinion->{opinion} = "TIMEOUT"; 
-  }
-  elsif ($rc) {
-    &log("Detector $d->{name} said rc $rc");
-    $opinion->{hasOpinion} = 0;
-    $opinion->{opinion} = "INTERNAL-ERROR"; 
-  }
-  else {
-    &log("Detector $d->{name} said: $out");
-    my $result = decode_json($out);
-    # Extract the details needed to make the summary.
-    # Otherwise we repeat ourselves too much.
-    $opinion->{hasOpinion} = 1;
-    $opinion->{opinion} = $result->{opinion};
+# Try each pattern.
+for my $pattern (@patternsToTry) {
+  &log("Applying detectors to pattern /$pattern/");
+
+  # Craft query file.
+  my $newQuery = decode_json(encode_json($query));
+  $newQuery->{pattern} = $pattern;
+  my $tmpPatternFile = &makeQueryFile($newQuery);
+
+  # Ask each detector.
+  for my $d (@DETECTORS) {
+    &log("Querying detector $d->{name}");
+    my $t0 = [gettimeofday];
+    my $stderrFile = "/tmp/detect-vuln-$$-stderr";
+    my ($rc, $out) = &cmd("$ulimitMemory $limitTime $d->{driver} $tmpPatternFile 2>$stderrFile");
+    my $elapsed = tv_interval($t0);
+    chomp $out;
+
+    # Clean up in case there was a timeout.
+    my $stderr = &readFile("file"=>$stderrFile);
+    my @filesToClean = ($stderr =~ m/CLEANUP: (\S+)/g);
+    &log("Cleaning up @filesToClean");
+    unlink @filesToClean;
+    unlink $stderrFile;
+
+    my $opinion = { "name"         => $d->{name},
+                    "secToDecide" => sprintf("%.4f", $elapsed),
+                  };
+
+    if ($rc eq 124) {
+      &log("Detector $d->{name} timed out");
+      $opinion->{hasOpinion} = 0;
+      $opinion->{opinion} = "TIMEOUT";
+    }
+    elsif ($rc) {
+      &log("Detector $d->{name} said rc $rc");
+      $opinion->{hasOpinion} = 0;
+      $opinion->{opinion} = "INTERNAL-ERROR";
+    }
+    else {
+      &log("Detector $d->{name} said: $out");
+      my $result = decode_json($out);
+      # Extract the details needed to make the summary.
+      # Otherwise we repeat ourselves too much.
+      $opinion->{hasOpinion} = 1;
+      $opinion->{opinion} = $result->{opinion};
+
+      # Note the pattern we queried about, so we can distinguish from the original.
+      $opinion->{patternVariant} = $pattern;
+    }
+
+    push @detectorOpinions, $opinion;
   }
 
-  push @detectorOpinions, $opinion;
+  unlink $tmpPatternFile;
 }
 
 $query->{detectorOpinions} = \@detectorOpinions;
@@ -149,6 +166,13 @@ sub getDetectors {
   return @detectors;
 }
 
+sub makeQueryFile {
+  my ($query) = @_;
+  my $tmpFile = "/tmp/detect-vuln-$$.json";
+  &writeToFile("file"=>$tmpFile, "contents"=>encode_json($query));
+  return $tmpFile;
+}
+
 # input: (\@list, $e)
 # output: true if $e is in @list, else false
 sub listContains {
@@ -173,3 +197,50 @@ sub readFile {
 
   return $contents;
 }
+
+# input: %args: keys: file contents
+# output: $file
+sub writeToFile {
+  my %args = @_;
+
+	open(my $fh, '>', $args{file});
+	print $fh $args{contents};
+	close $fh;
+
+  return $args{file};
+}
+
+sub expandPatternSpaceForDetectors {
+  my ($pattern) = @_;
+
+  my @patternsToTry = ($pattern);
+
+  # If pattern is unanchored, a backtracking regex engine will run the loop:
+  #   for (1 .. n):
+  #     _match(regex, substr)
+  # This means that if each match is linear-time, the worst-case behavior is quadratic.
+  # For example, /a+$/ is quadratic in Node.js.
+  # The detectors don't seem to acknowledge this loop.
+  # We can simulate it by prefixing un-anchored regexes with '^(.*?)'.
+  # This is also how a linear-time engine scans all starting indices in parallel; see Cox's writings.
+  if (substr($query->{pattern}, 0, 1) ne "^") {
+    my $anchoredPattern = "^(.*?)$query->{pattern}";
+    push @patternsToTry, $anchoredPattern;
+  }
+
+  # If pattern contains curlies "{\d*,\d*}", the detectors may time out due to graph expansion.
+  # We can try a more general pattern with "*" and "+" instead.
+  # The detectors might give false positives but that's OK, that's what the validate stage is for.
+  # I'm not being careful about escaped curly braces, so let's hope there are no meta-regexes here.
+  my $genericCurlies = $query->{pattern};
+  # {0, and {, both mean "0 or more"
+  $genericCurlies =~ s/{0,\d*}/\*/g;
+  $genericCurlies =~ s/{,\d*}/\*/g;
+  # {[1-9] means "1 or more"
+  $genericCurlies =~ s/{[1-9]\d*,\d*}/\+/g;
+  if ($genericCurlies ne $pattern) {
+    push @patternsToTry, $genericCurlies;
+  }
+
+  return @patternsToTry;
+}