C++: Support std::addressof

I didn't add this support in `AddressConstantExpression.qll` since I
think it would require extra work and testing to get the constexprness
right. My long-term plan for `AddressConstantExpression.qll` is to move
its functionality to the extractor.

cpp/ql/src/semmle/code/cpp/dataflow/EscapesTree.qll

@@ -19,15 +19,13 @@ private import cpp
  * template functions, these functions are essentially casts, so we treat them
  * as such.
  */
-private predicate stdIdentityFunction(Function f) {
+private predicate stdIdentityFunction(Function f) { f.hasQualifiedName("std", ["move", "forward"]) }
-  f.getNamespace().getParentNamespace() instanceof GlobalNamespace and
+
-  f.getNamespace().getName() = "std" and
+/**
-  (
+ * Holds if `f` is an instantiation of `std::addressof`, which effectively
-    f.getName() = "move"
+ * converts a reference to a pointer.
-    or
+ */
-    f.getName() = "forward"
+private predicate stdAddressOf(Function f) { f.hasQualifiedName("std", "addressof") }
-  )
-}
 
 private predicate lvalueToLvalueStepPure(Expr lvalueIn, Expr lvalueOut) {
   lvalueIn = lvalueOut.(DotFieldAccess).getQualifier().getFullyConverted()
@@ -99,12 +97,17 @@ private predicate lvalueToReferenceStep(Expr lvalueIn, Expr referenceOut) {
 }
 
 private predicate referenceToLvalueStep(Expr referenceIn, Expr lvalueOut) {
-  // This probably cannot happen. It would require an expression to be
-  // converted to a reference and back again without an intermediate variable
-  // assignment.
   referenceIn.getConversion() = lvalueOut.(ReferenceDereferenceExpr)
 }
 
+private predicate referenceToPointerStep(Expr referenceIn, Expr pointerOut) {
+  pointerOut =
+    any(FunctionCall call |
+      stdAddressOf(call.getTarget()) and
+      referenceIn = call.getArgument(0).getFullyConverted()
+    )
+}
+
 private predicate referenceToReferenceStep(Expr referenceIn, Expr referenceOut) {
   referenceOut =
     any(FunctionCall call |
@@ -153,6 +156,12 @@ private predicate pointerFromVariableAccess(VariableAccess va, Expr pointer) {
     pointerToPointerStep(prev, pointer)
   )
   or
+  // reference -> pointer
+  exists(Expr prev |
+    referenceFromVariableAccess(va, prev) and
+    referenceToPointerStep(prev, pointer)
+  )
+  or
   // lvalue -> pointer
   exists(Expr prev |
     lvalueFromVariableAccess(va, prev) and
@@ -177,7 +186,8 @@ private predicate referenceFromVariableAccess(VariableAccess va, Expr reference)
 private predicate valueMayEscapeAt(Expr e) {
   exists(Call call |
     e = call.getAnArgument().getFullyConverted() and
-    not stdIdentityFunction(call.getTarget())
+    not stdIdentityFunction(call.getTarget()) and
+    not stdAddressOf(call.getTarget())
   )
   or
   exists(AssignExpr assign | e = assign.getRValue().getFullyConverted())

cpp/ql/src/semmle/code/cpp/dataflow/internal/AddressFlow.qll

@@ -21,15 +21,13 @@ private import cpp
  * template functions, these functions are essentially casts, so we treat them
  * as such.
  */
-private predicate stdIdentityFunction(Function f) {
+private predicate stdIdentityFunction(Function f) { f.hasQualifiedName("std", ["move", "forward"]) }
-  f.getNamespace().getParentNamespace() instanceof GlobalNamespace and
+
-  f.getNamespace().getName() = "std" and
+/**
-  (
+ * Holds if `f` is an instantiation of `std::addressof`, which effectively
-    f.getName() = "move"
+ * converts a reference to a pointer.
-    or
+ */
-    f.getName() = "forward"
+private predicate stdAddressOf(Function f) { f.hasQualifiedName("std", "addressof") }
-  )
-}
 
 private predicate lvalueToLvalueStep(Expr lvalueIn, Expr lvalueOut) {
   lvalueIn.getConversion() = lvalueOut.(ParenthesisExpr)
@@ -96,6 +94,14 @@ private predicate referenceToLvalueStep(Expr referenceIn, Expr lvalueOut) {
   referenceIn.getConversion() = lvalueOut.(ReferenceDereferenceExpr)
 }
 
+private predicate referenceToPointerStep(Expr referenceIn, Expr pointerOut) {
+  pointerOut =
+    any(FunctionCall call |
+      stdAddressOf(call.getTarget()) and
+      referenceIn = call.getArgument(0).getFullyConverted()
+    )
+}
+
 private predicate referenceToReferenceStep(Expr referenceIn, Expr referenceOut) {
   referenceOut =
     any(FunctionCall call |
@@ -185,6 +191,7 @@ private predicate referenceToUpdate(Expr reference, Expr outer, ControlFlowNode
     node = call and
     outer = call.getAnArgument().getFullyConverted() and
     not stdIdentityFunction(call.getTarget()) and
+    not stdAddressOf(call.getTarget()) and
     exists(ReferenceType rt | rt = outer.getType().stripTopLevelSpecifiers() |
       not rt.getBaseType().isConst()
     )
@@ -196,6 +203,11 @@ private predicate referenceToUpdate(Expr reference, Expr outer, ControlFlowNode
     lvalueToUpdate(lvalueMid, outer, node)
   )
   or
+  exists(Expr pointerMid |
+    referenceToPointerStep(reference, pointerMid) and
+    pointerToUpdate(pointerMid, outer, node)
+  )
+  or
   exists(Expr referenceMid |
     referenceToReferenceStep(reference, referenceMid) and
     referenceToUpdate(referenceMid, outer, node)

cpp/ql/test/library-tests/defuse/addressOf.cpp

@@ -63,3 +63,18 @@ void nonexamples(int *ptr, int &ref) {
     nonexamples(&*ptr, ref);
   }
 }
+
+
+namespace std {
+  template<typename T>
+  constexpr T *addressof(T &obj) noexcept {
+    return __builtin_addressof(obj);
+  }
+}
+
+void use_std_addressof() {
+  int x = 0;
+  int *y = std::addressof(x) + *std::addressof(x);
+}
+
+// semmle-extractor-options: --clang

cpp/ql/test/library-tests/defuse/isAddressOfAccess.expected

@@ -25,6 +25,10 @@
 | addressOf.cpp:62:11:62:13 | ptr |  |
 | addressOf.cpp:63:19:63:21 | ptr |  |
 | addressOf.cpp:63:24:63:26 | ref | non-const address |
+| addressOf.cpp:71:32:71:34 | obj |  |
+| addressOf.cpp:71:32:71:34 | obj |  |
+| addressOf.cpp:77:27:77:27 | x | non-const address |
+| addressOf.cpp:77:48:77:48 | x |  |
 | file://:0:0:0:0 | captured |  |
 | file://:0:0:0:0 | captured |  |
 | file://:0:0:0:0 | captured | non-const address |