Merge pull request #5578 from MathiasVP/ast-flow-smart-pointers

C++: AST dataflow through smart pointers

Author: jbj

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

@@ -15,6 +15,7 @@
  */
 
 private import cpp
+private import semmle.code.cpp.models.interfaces.PointerWrapper
 
 /**
  * Holds if `f` is an instantiation of the `std::move` or `std::forward`
@@ -94,6 +95,12 @@ private predicate pointerToPointerStep(Expr pointerIn, Expr pointerOut) {
 
 private predicate lvalueToReferenceStep(Expr lvalueIn, Expr referenceOut) {
   lvalueIn.getConversion() = referenceOut.(ReferenceToExpr)
+  or
+  exists(PointerWrapper wrapper, Call call | call = referenceOut |
+    referenceOut.getUnspecifiedType() instanceof ReferenceType and
+    call = wrapper.getAnUnwrapperFunction().getACallToThisFunction() and
+    lvalueIn = call.getQualifier().getFullyConverted()
+  )
 }
 
 private predicate referenceToLvalueStep(Expr referenceIn, Expr lvalueOut) {
@@ -106,6 +113,13 @@ private predicate referenceToPointerStep(Expr referenceIn, Expr pointerOut) {
       stdAddressOf(call.getTarget()) and
       referenceIn = call.getArgument(0).getFullyConverted()
     )
+  or
+  exists(CopyConstructor copy, Call call | call = pointerOut |
+    copy.getDeclaringType() instanceof PointerWrapper and
+    call.getTarget() = copy and
+    // The 0'th argument is the value being copied.
+    referenceIn = call.getArgument(0).getFullyConverted()
+  )
 }
 
 private predicate referenceToReferenceStep(Expr referenceIn, Expr referenceOut) {
@@ -190,6 +204,19 @@ private predicate pointerToUpdate(Expr pointer, Expr outer, ControlFlowNode node
         // See the `lvalueToUpdate` case for an explanation of this conjunct.
         call.getType().isDeeplyConstBelow()
       )
+      or
+      // Pointer wrappers behave as raw pointers for dataflow purposes.
+      outer = call.getAnArgument().getFullyConverted() and
+      exists(PointerWrapper wrapper | wrapper = outer.getType().stripTopLevelSpecifiers() |
+        not wrapper.pointsToConst()
+      )
+      or
+      outer = call.getQualifier().getFullyConverted() and
+      outer.getUnspecifiedType() instanceof PointerWrapper and
+      not (
+        call.getTarget().hasSpecifier("const") and
+        call.getType().isDeeplyConstBelow()
+      )
     )
     or
     exists(PointerFieldAccess fa |
@@ -218,7 +245,9 @@ private predicate referenceToUpdate(Expr reference, Expr outer, ControlFlowNode
     not stdIdentityFunction(call.getTarget()) and
     not stdAddressOf(call.getTarget()) and
     exists(ReferenceType rt | rt = outer.getType().stripTopLevelSpecifiers() |
-      not rt.getBaseType().isConst()
+      not rt.getBaseType().isConst() or
+      rt.getBaseType().getUnspecifiedType() =
+        any(PointerWrapper wrapper | not wrapper.pointsToConst())
     )
   ) and
   reference = outer

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

@@ -324,7 +324,7 @@ private class VariablePartialDefinitionNode extends PartialDefinitionNode {
  * A synthetic data flow node used for flow into a collection when an iterator
  * write occurs in a callee.
  */
-class IteratorPartialDefinitionNode extends PartialDefinitionNode {
+private class IteratorPartialDefinitionNode extends PartialDefinitionNode {
   override IteratorPartialDefinition pd;
 
   override Node getPreUpdateNode() { pd.definesExpressions(_, result.asExpr()) }

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

@@ -7,6 +7,7 @@ private import semmle.code.cpp.controlflow.SSA
 private import semmle.code.cpp.dataflow.internal.SubBasicBlocks
 private import semmle.code.cpp.dataflow.internal.AddressFlow
 private import semmle.code.cpp.models.implementations.Iterator
+private import semmle.code.cpp.models.interfaces.PointerWrapper
 
 /**
  * A conceptual variable that is assigned only once, like an SSA variable. This
@@ -158,18 +159,14 @@ private module PartialDefinitions {
     Expr innerDefinedExpr;
 
     IteratorPartialDefinition() {
-      exists(Expr convertedInner |
-        not this instanceof Conversion and
-        valueToUpdate(convertedInner, this.getFullyConverted(), node) and
-        innerDefinedExpr = convertedInner.getUnconverted() and
-        (
-          innerDefinedExpr.(Call).getQualifier() = getAnIteratorAccess(collection)
-          or
-          innerDefinedExpr.(Call).getQualifier() = collection.getAnAccess() and
-          collection instanceof IteratorParameter
-        ) and
-        innerDefinedExpr.(Call).getTarget() instanceof IteratorPointerDereferenceMemberOperator
-      )
+      innerDefinedExpr = getInnerDefinedExpr(this, node) and
+      (
+        innerDefinedExpr.(Call).getQualifier() = getAnIteratorAccess(collection)
+        or
+        innerDefinedExpr.(Call).getQualifier() = collection.getAnAccess() and
+        collection instanceof IteratorParameter
+      ) and
+      innerDefinedExpr.(Call).getTarget() instanceof IteratorPointerDereferenceMemberOperator
       or
       // iterators passed by value without a copy constructor
       exists(Call call |
@@ -207,16 +204,18 @@ private module PartialDefinitions {
     }
   }
 
+  private Expr getInnerDefinedExpr(Expr e, ControlFlowNode node) {
+    not e instanceof Conversion and
+    exists(Expr convertedInner |
+      valueToUpdate(convertedInner, e.getFullyConverted(), node) and
+      result = convertedInner.getUnconverted()
+    )
+  }
+
   class VariablePartialDefinition extends PartialDefinition {
     Expr innerDefinedExpr;
 
-    VariablePartialDefinition() {
-      not this instanceof Conversion and
-      exists(Expr convertedInner |
-        valueToUpdate(convertedInner, this.getFullyConverted(), node) and
-        innerDefinedExpr = convertedInner.getUnconverted()
-      )
-    }
+    VariablePartialDefinition() { innerDefinedExpr = getInnerDefinedExpr(this, node) }
 
     deprecated override predicate partiallyDefines(Variable v) {
       innerDefinedExpr = v.getAnAccess()
@@ -296,7 +295,8 @@ module FlowVar_internal {
     // treating them as immutable, but for data flow it gives better results in
     // practice to make the variable synonymous with its contents.
     not v.getUnspecifiedType() instanceof ReferenceType and
-    not v instanceof IteratorParameter
+    not v instanceof IteratorParameter and
+    not v instanceof PointerWrapperParameter
   }
 
   /**
@@ -644,10 +644,19 @@ module FlowVar_internal {
   predicate parameterIsNonConstReference(Parameter p) {
     exists(ReferenceType refType |
       refType = p.getUnderlyingType() and
-      not refType.getBaseType().isConst()
+      (
+        not refType.getBaseType().isConst()
+        or
+        // A field of a parameter of type `const std::shared_ptr<A>& p` can still be changed even though
+        // the base type of the reference is `const`.
+        refType.getBaseType().getUnspecifiedType() =
+          any(PointerWrapper wrapper | not wrapper.pointsToConst())
+      )
     )
     or
     p instanceof IteratorParameter
+    or
+    p instanceof PointerWrapperParameter
   }
 
   /**
@@ -836,6 +845,10 @@ module FlowVar_internal {
     IteratorParameter() { this.getUnspecifiedType() instanceof Iterator }
   }
 
+  class PointerWrapperParameter extends Parameter {
+    PointerWrapperParameter() { this.getUnspecifiedType() instanceof PointerWrapper }
+  }
+
   /**
    * Holds if `v` is initialized to have value `assignedExpr`.
    */

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

@@ -11,6 +11,7 @@
 private import semmle.code.cpp.models.interfaces.DataFlow
 private import semmle.code.cpp.models.interfaces.Taint
 private import semmle.code.cpp.models.interfaces.Iterator
+private import semmle.code.cpp.models.interfaces.PointerWrapper
 
 private module DataFlow {
   import semmle.code.cpp.dataflow.internal.DataFlowUtil
@@ -141,7 +142,10 @@ private predicate noFlowFromChildExpr(Expr e) {
   or
   e instanceof LogicalOrExpr
   or
-  e instanceof Call
+  // Allow taint from `operator*` on smart pointers.
+  exists(Call call | e = call |
+    not call.getTarget() = any(PointerWrapper wrapper).getAnUnwrapperFunction()
+  )
   or
   e instanceof SizeofOperator
   or

cpp/ql/src/semmle/code/cpp/exprs/Call.qll

@@ -314,6 +314,7 @@ class OverloadedPointerDereferenceFunction extends Function {
  * T1 operator*(const T2 &);
  * T1 a; T2 b;
  * a = *b;
+ * ```
  */
 class OverloadedPointerDereferenceExpr extends FunctionCall {
   OverloadedPointerDereferenceExpr() {

cpp/ql/src/semmle/code/cpp/models/implementations/SmartPointer.qll

@@ -13,6 +13,8 @@ private class UniqueOrSharedPtr extends Class, PointerWrapper {
     or
     result.getClassAndName(["operator->", "get"]) = this
   }
+
+  override predicate pointsToConst() { this.getTemplateArgument(0).(Type).isConst() }
 }
 
 /** Any function that unwraps a pointer wrapper class to reveal the underlying pointer. */

cpp/ql/src/semmle/code/cpp/models/interfaces/PointerWrapper.qll

@@ -11,4 +11,7 @@ abstract class PointerWrapper extends Class {
    * that return a reference to the pointed-to object.
    */
   abstract MemberFunction getAnUnwrapperFunction();
+
+  /** Holds if the type of the data that is pointed to by this pointer wrapper is `const`. */
+  abstract predicate pointsToConst();
 }