add stringConcatenatedWith feature to help the model learn that string concatenation leaves are usually not sinks

javascript/ql/experimental/adaptivethreatmodeling/lib/experimental/adaptivethreatmodeling/EndpointFeatures.qll

@@ -237,7 +237,8 @@ private newtype TEndpointFeature =
   TInputArgumentIndex() or
   TContextFunctionInterfaces() or
   TContextSurroundingFunctionParameters() or
-  TAssignedToPropName()
+  TAssignedToPropName() or
+  TStringConcatenatedWith()
 
 /**
  * An implementation of an endpoint feature: produces feature names and values for used in ML.
@@ -492,6 +493,51 @@ class AssignedToPropName extends EndpointFeature, TAssignedToPropName {
   }
 }
 
+/**
+ * The feature that shows the text an endpoint is being concatenated with.class
+ *
+ * ### Example
+ *
+ * ```javascript
+ * const x = 'foo' + endpoint + 'bar'; // feature value is `'foo' -endpoint- 'bar'`
+ */
+class StringConcatenatedWith extends EndpointFeature, TStringConcatenatedWith {
+  override string getName() { result = "stringConcatenatedWith" }
+
+  override string getValue(DataFlow::Node endpoint) {
+    exists(StringOps::ConcatenationRoot root |
+      root.getALeaf() = endpoint and
+      result =
+        concat(StringOps::ConcatenationLeaf p |
+            p.getRoot() = root and
+            (
+              p.getStartLine() < endpoint.getStartLine()
+              or
+              p.getStartLine() = endpoint.getStartLine() and
+              p.getStartColumn() < endpoint.getStartColumn()
+            )
+          |
+            SyntacticUtilities::renderStringConcatOperand(p), " + "
+            order by
+              p.getStartLine(), p.getStartColumn()
+          ) + " -endpoint- " +
+          concat(StringOps::ConcatenationLeaf p |
+            p.getRoot() = root and
+            (
+              p.getStartLine() > endpoint.getStartLine()
+              or
+              p.getStartLine() = endpoint.getStartLine() and
+              p.getStartColumn() > endpoint.getStartColumn()
+            )
+          |
+            SyntacticUtilities::renderStringConcatOperand(p), " + "
+            order by
+              p.getStartLine(), p.getStartColumn()
+          )
+    )
+  }
+}
+
 /**
  * The feature for the imports used in the callee of an invocation.
  *
@@ -555,6 +601,23 @@ class ContextFunctionInterfaces extends EndpointFeature, TContextFunctionInterfa
  * Syntactic utilities for feature value computation.
  */
 private module SyntacticUtilities {
+  bindingset[start, end]
+  string renderStringConcatOperands(DataFlow::Node root, int start, int end) {
+    result =
+      concat(int i, string operand |
+        i = [start .. end] and
+        operand = renderStringConcatOperand(StringConcatenation::getOperand(root, i))
+      |
+        operand, " + " order by i
+      )
+  }
+
+  string renderStringConcatOperand(DataFlow::Node operand) {
+    if exists(unique(string v | operand.mayHaveStringValue(v)))
+    then result = "'" + any(string v | operand.mayHaveStringValue(v)) + "'"
+    else result = getSimpleAccessPath(operand)
+  }
+
   /** Gets all the imports defined in the file containing the endpoint. */
   string getImportPathsForFile(File file) {
     result =

javascript/ql/experimental/adaptivethreatmodeling/test/generic_feature_testing/FeatureValue.expected

@@ -188,3 +188,15 @@
 | test.js:22:21:22:28 | endpoint | enclosingFunctionBody | f endpoint 12 f p endpoint f p q endpoint o m endpoint o m p endpoint o m p q endpoint F endpoint o m m m endpoint f endpoint o x m endpoint o m x p m endpoint p endpoint foo bar baz endpoint foo bar endpoint f f o m endpoint |
 | test.js:22:21:22:28 | endpoint | enclosingFunctionName |  |
 | test.js:22:21:22:28 | endpoint | fileImports | foo lib1 lib2 lib3 |
+| test.js:33:50:33:57 | endpoint | calleeAccessPath |  |
+| test.js:33:50:33:57 | endpoint | calleeAccessPathWithStructuralInfo |  |
+| test.js:33:50:33:57 | endpoint | contextFunctionInterfaces | f(?)\nfoo()\ng()\nm() |
+| test.js:33:50:33:57 | endpoint | contextSurroundingFunctionParameters |  |
+| test.js:33:50:33:57 | endpoint | fileImports | foo lib1 lib2 lib3 |
+| test.js:33:50:33:57 | endpoint | stringConcatenatedWith | f() + '<a target="_blank" href="' -endpoint- '"></a>' |
+| test.js:35:18:35:25 | endpoint | calleeAccessPath |  |
+| test.js:35:18:35:25 | endpoint | calleeAccessPathWithStructuralInfo |  |
+| test.js:35:18:35:25 | endpoint | contextFunctionInterfaces | f(?)\nfoo()\ng()\nm() |
+| test.js:35:18:35:25 | endpoint | contextSurroundingFunctionParameters |  |
+| test.js:35:18:35:25 | endpoint | fileImports | foo lib1 lib2 lib3 |
+| test.js:35:18:35:25 | endpoint | stringConcatenatedWith | 'foo' -endpoint- 'bar' |

javascript/ql/experimental/adaptivethreatmodeling/test/generic_feature_testing/test.js

@@ -27,3 +27,9 @@ function f({ endpoint }) {}
 const g = async () => undefined;
 
 const o = { m: () => undefined }
+
+const url = f();
+
+const x = f() + "<a target=\"_blank\" href=\"" + endpoint + "\"></a>";
+
+const y = "foo"+ endpoint + "bar";