[analyzer] MallocChecker: Add notes from NoOwnershipChangeVisitor only when a functio...
authorKristóf Umann <dkszelethus@gmail.com>
Wed, 25 Aug 2021 14:07:56 +0000 (16:07 +0200)
committerKristóf Umann <kristof.umann@ericsson.com>
Mon, 13 Sep 2021 13:01:20 +0000 (15:01 +0200)
D105819 Added NoOwnershipChangeVisitor, but it is only registered when an
off-by-default, hidden checker option was enabled. The reason behind this was
that it grossly overestimated the set of functions that really needed a note:

std::string getTrainName(const Train *T) {
  return T->name;
} // note: Retuning without changing the ownership of or deallocating memory
// Umm... I mean duh? Nor would I expect this function to do anything like that...

void foo() {
  Train *T = new Train("Land Plane");
  print(getTrainName(T)); // note: calling getTrainName / returning from getTrainName
} // warn: Memory leak

This patch adds a heuristic that guesses that any function that has an explicit
operator delete call could have be responsible for deallocating the memory that
ended up leaking. This is waaaay too conservative (see the TODOs in the new
function), but it safer to err on the side of too little than too much, and
would allow us to enable the option by default *now*, and add refinements
one-by-one.

Differential Revision: https://reviews.llvm.org/D108753

clang/include/clang/StaticAnalyzer/Checkers/Checkers.td
clang/lib/StaticAnalyzer/Checkers/MallocChecker.cpp
clang/test/Analysis/NewDeleteLeaks.cpp
clang/test/Analysis/analyzer-config.c

index 125ef85..86513a3 100644 (file)
@@ -493,8 +493,8 @@ def DynamicMemoryModeling: Checker<"DynamicMemoryModeling">,
                   "that neither deallocated it, or have taken responsibility "
                   "of the ownership are noted, similarly to "
                   "NoStoreFuncVisitor.",
-                  "false",
-                  InAlpha,
+                  "true",
+                  Released,
                   Hide>
   ]>,
   Dependencies<[CStringModeling]>,
index d97e8c3..9899298 100644 (file)
@@ -52,6 +52,8 @@
 #include "clang/AST/Expr.h"
 #include "clang/AST/ExprCXX.h"
 #include "clang/AST/ParentMap.h"
+#include "clang/ASTMatchers/ASTMatchFinder.h"
+#include "clang/ASTMatchers/ASTMatchers.h"
 #include "clang/Analysis/ProgramPoint.h"
 #include "clang/Basic/LLVM.h"
 #include "clang/Basic/SourceManager.h"
@@ -791,9 +793,28 @@ protected:
     return "";
   }
 
+  bool doesFnIntendToHandleOwnership(const Decl *Callee, ASTContext &ACtx) {
+    using namespace clang::ast_matchers;
+    const FunctionDecl *FD = dyn_cast<FunctionDecl>(Callee);
+    if (!FD)
+      return false;
+    // TODO: Operator delete is hardly the only deallocator -- Can we reuse
+    // isFreeingCall() or something thats already here?
+    auto Deallocations = match(
+        stmt(hasDescendant(cxxDeleteExpr().bind("delete"))
+             ), *FD->getBody(), ACtx);
+    // TODO: Ownership my change with an attempt to store the allocated memory.
+    return !Deallocations.empty();
+  }
+
   virtual bool
   wasModifiedInFunction(const ExplodedNode *CallEnterN,
                         const ExplodedNode *CallExitEndN) override {
+    if (!doesFnIntendToHandleOwnership(
+            CallExitEndN->getFirstPred()->getLocationContext()->getDecl(),
+            CallExitEndN->getState()->getAnalysisManager().getASTContext()))
+      return true;
+
     if (CallEnterN->getState()->get<RegionState>(Sym) !=
         CallExitEndN->getState()->get<RegionState>(Sym))
       return true;
index 28040d9..57c7e57 100644 (file)
 
 bool coin();
 
+// TODO: AST analysis of sink would reveal that it doesn't intent to free the
+// allocated memory, but in this instance, its also the only function with
+// the ability to do so, we should see a note here.
 namespace memory_allocated_in_fn_call {
 
 void sink(int *P) {
-} // ownership-note {{Returning without deallocating memory or storing the pointer for later deallocation}}
+}
 
 void foo() {
   sink(new int(5)); // expected-note {{Memory is allocated}}
-                    // ownership-note@-1 {{Calling 'sink'}}
-                    // ownership-note@-2 {{Returning from 'sink'}}
 } // expected-warning {{Potential memory leak [cplusplus.NewDeleteLeaks]}}
 // expected-note@-1 {{Potential memory leak}}
 
@@ -109,17 +110,14 @@ void foo() {
 
 } // namespace memory_shared_with_ptr_of_same_lifetime
 
-// TODO: We don't want a note here. sink() doesn't seem like a function that
-// even attempts to take care of any memory ownership problems.
 namespace memory_passed_into_fn_that_doesnt_intend_to_free {
 
 void sink(int *P) {
-} // ownership-note {{Returning without deallocating memory or storing the pointer for later deallocation}}
+}
 
 void foo() {
   int *ptr = new int(5); // expected-note {{Memory is allocated}}
-  sink(ptr);             // ownership-note {{Calling 'sink'}}
-                         // ownership-note@-1 {{Returning from 'sink'}}
+  sink(ptr);
 } // expected-warning {{Potential leak of memory pointed to by 'ptr' [cplusplus.NewDeleteLeaks]}}
 // expected-note@-1 {{Potential leak}}
 
index bf3335c..4cca1cb 100644 (file)
 // CHECK-NEXT: suppress-null-return-paths = true
 // CHECK-NEXT: track-conditions = true
 // CHECK-NEXT: track-conditions-debug = false
-// CHECK-NEXT: unix.DynamicMemoryModeling:AddNoOwnershipChangeNotes = false
+// CHECK-NEXT: unix.DynamicMemoryModeling:AddNoOwnershipChangeNotes = true
 // CHECK-NEXT: unix.DynamicMemoryModeling:Optimistic = false
 // CHECK-NEXT: unroll-loops = false
 // CHECK-NEXT: verbose-report-filename = false