[clang][analyzer] Make ReturnPtrRange checker warn at negative index.

[balazske]

Summary

The checker did not work with negative array index at return.
Specially for -1 it did not produce a bug.
assumeInBound may not work correct with negative values so a pre-
check is added to the checker for negative array index.

[steakhal]

If assumeInBound() behaves unexpectedly, we should probably fix that instead.
This change would workaround the issue, instead of fixing it. By fixing the root cause all users of assumeInBound() would benefit.
I recommend you to have a deeper look at the insides of assumeInBound() investigating why it behaves that way.

[balazske]

assumeInBound should check if 0<=Idx<UpperBound. In this case it makes no sense of UpperBound is negative.
The problem appears in cases like this:

int *conjure();
int *m1() {
  int *p = conjure();
  return p - 1;
}

The extent of the region is symbolic and the index is a constant, so a constraint for the extent appears after assumeInBound.
For positive index (at p - 1 in the code) it looks correct, if p + 1 is used:

StOutBound:"program_state": {
  "store": { "pointer": "0x6acdd8", "items": [
    { "cluster": "GlobalInternalSpaceRegion", "pointer": "0x6acb98", "items": [
      { "kind": "Default", "offset": 0, "value": "conj_$0{int, LC1, S666, #1}" }
    ]},
    { "cluster": "GlobalSystemSpaceRegion", "pointer": "0x6accb8", "items": [
      { "kind": "Default", "offset": 0, "value": "conj_$1{int, LC1, S666, #1}" }
    ]}
  ]},
  "environment": { "pointer": "0x6ac010", "items": [
    { "lctx_id": 1, "location_context": "#0 Call", "calling": "conjured::b3", "location": null, "items": [
      { "stmt_id": 684, "pretty": "p + 1", "value": "&Element{SymRegion{conj_$2{int *, LC1, S666, #1}},1 S64b,int}" }
    ]}
  ]},
  "constraints": [
    { "symbol": "(extent_$3{SymRegion{conj_$2{int *, LC1, S666, #1}}}) / 4", "range": "{ [0, 1] }" }
  ],
  "equivalence_classes": null,
  "disequality_info": null,
  "dynamic_types": null,
  "dynamic_casts": null,
  "constructing_objects": null,
  "checker_messages": null
}
StInBound:"program_state": {
  "store": { "pointer": "0x6acdd8", "items": [
    { "cluster": "GlobalInternalSpaceRegion", "pointer": "0x6acb98", "items": [
      { "kind": "Default", "offset": 0, "value": "conj_$0{int, LC1, S666, #1}" }
    ]},
    { "cluster": "GlobalSystemSpaceRegion", "pointer": "0x6accb8", "items": [
      { "kind": "Default", "offset": 0, "value": "conj_$1{int, LC1, S666, #1}" }
    ]}
  ]},
  "environment": { "pointer": "0x6ac010", "items": [
    { "lctx_id": 1, "location_context": "#0 Call", "calling": "conjured::b3", "location": null, "items": [
      { "stmt_id": 684, "pretty": "p + 1", "value": "&Element{SymRegion{conj_$2{int *, LC1, S666, #1}},1 S64b,int}" }
    ]}
  ]},
  "constraints": [
    { "symbol": "(extent_$3{SymRegion{conj_$2{int *, LC1, S666, #1}}}) / 4", "range": "{ [-9223372036854775808, -1], [2, 9223372036854775807] }" }
  ],
  "equivalence_classes": null,
  "disequality_info": null,
  "dynamic_types": null,
  "dynamic_casts": null,
  "constructing_objects": null,
  "checker_messages": null
}

For p - 1 only the StOutBound exists:

OutBound:"program_state": {
  "store": { "pointer": "0x161bdd8", "items": [
    { "cluster": "GlobalInternalSpaceRegion", "pointer": "0x161bb98", "items": [
      { "kind": "Default", "offset": 0, "value": "conj_$0{int, LC1, S666, #1}" }
    ]},
    { "cluster": "GlobalSystemSpaceRegion", "pointer": "0x161bcb8", "items": [
      { "kind": "Default", "offset": 0, "value": "conj_$1{int, LC1, S666, #1}" }
    ]}
  ]},
  "environment": { "pointer": "0x161b010", "items": [
    { "lctx_id": 1, "location_context": "#0 Call", "calling": "conjured::b3", "location": null, "items": [
      { "stmt_id": 684, "pretty": "p - 1", "value": "&Element{SymRegion{conj_$2{int *, LC1, S666, #1}},-1 S64b,int}" }
    ]}
  ]},
  "constraints": [
    { "symbol": "(extent_$3{SymRegion{conj_$2{int *, LC1, S666, #1}}}) / 4", "range": "{ [-9223372036854775808, 9223372036854775807] }" }
  ],
  "equivalence_classes": null,
  "disequality_info": null,
  "dynamic_types": null,
  "dynamic_casts": null,
  "constructing_objects": null,
  "checker_messages": null
}

For p - 2 the StOutBound has constraint { [-9223372036854775808, -2], [0, 9223372036854775807] } and StInBound has { [-1, -1] }. Probably this is how the overflow situation is handled or it is not correct.

This checker can do anyway nothing with a negative index. It may be out of bound in a non-symbolic region but can be valid.

[martong]

Would it make sense to tweak assumeInBound in a way to handle symbolic regions differently than concrete regions?
I mean

static int arr[10];
return arr - 1; // we want a warning here
...

int *p = conjure();
return p - 1;  // no warning here

[balazske]

With the current main branch the new test passes.

clang/lib/StaticAnalyzer/Checkers/ReturnPointerRangeChecker.cpp

void ReturnPointerRangeChecker::checkPreStmt(const ReturnStmt *RS,
   DefinedOrUnknownSVal ElementCount = getDynamicElementCount(
       state, ER->getSuperRegion(), C.getSValBuilder(), ER->getValueType());
 
   // We assume that the location after the last element in the array is used as
   // end() iterator. Reporting on these would return too many false positives.
   if (Idx == ElementCount)
     return;
 
+  auto ElementCountNL = ElementCount.getAs<NonLoc>();
+  assert(ElementCountNL && "Array size should not be a pointer");
+  auto IdxNegativeVal =
+      C.getSValBuilder()
+          .evalBinOpNN(state, BO_LT, *ElementCountNL,
+                       C.getSValBuilder().makeZeroArrayIndex(),
+                       C.getSValBuilder().getConditionType())
+          .castAs<DefinedOrUnknownSVal>();
+  ProgramStateRef IdxNegative, IdxNonNegative;
+  std::tie(IdxNegative, IdxNonNegative) = state->assume(IdxNegativeVal);
+
+  ProgramStateRef StError;
+  if (IdxNegative && !IdxNonNegative) {
+    StError = IdxNegative;
+  } else {
     ProgramStateRef StInBound = state->assumeInBound(Idx, ElementCount, true);
     ProgramStateRef StOutBound = state->assumeInBound(Idx, ElementCount, false);
-  if (StOutBound && !StInBound) {
-    ExplodedNode *N = C.generateErrorNode(StOutBound);
+    if (StOutBound && !StInBound)
+      StError = StOutBound;
+    else
+      return;
+  }
+
+  ExplodedNode *N = C.generateErrorNode(StError);
 
-    if (!N)
-      return;
 
-    // FIXME: This bug correspond to CWE-466.  Eventually we should have bug
-    // types explicitly reference such exploit categories (when applicable).
-    if (!BT)
-      BT.reset(new BuiltinBug(
-          this, "Buffer overflow",
-          "Returned pointer value points outside the original object "
-          "(potential buffer overflow)"));
 
-    // Generate a report for this bug.
-    auto Report =
-        std::make_unique<PathSensitiveBugReport>(*BT, BT->getDescription(), N);
-    Report->addRange(RetE->getSourceRange());
 
-    const auto ConcreteElementCount = ElementCount.getAs<nonloc::ConcreteInt>();
-    const auto ConcreteIdx = Idx.getAs<nonloc::ConcreteInt>();
 
-    const auto *DeclR = ER->getSuperRegion()->getAs<DeclRegion>();
 
-    if (DeclR)
-      Report->addNote("Original object declared here",
-                      {DeclR->getDecl(), C.getSourceManager()});
 
-    if (ConcreteElementCount) {
-      SmallString<128> SBuf;
-      llvm::raw_svector_ostream OS(SBuf);
-      OS << "Original object ";
-      if (DeclR) {
-        OS << "'";
-        DeclR->getDecl()->printName(OS);
-        OS << "' ";
-      }
-      OS << "is an array of " << ConcreteElementCount->getValue() << " '";
-      ER->getValueType().print(OS,
-                               PrintingPolicy(C.getASTContext().getLangOpts()));
-      OS << "' objects";
-      if (ConcreteIdx) {
-        OS << ", returned pointer points at index " << ConcreteIdx->getValue();
-      }
 
-      Report->addNote(SBuf,
-                      {RetE, C.getSourceManager(), C.getLocationContext()});
-    }
+    Report->addNote(SBuf, {RetE, C.getSourceManager(), C.getLocationContext()});
 
-    bugreporter::trackExpressionValue(N, RetE, *Report);
 
-    C.emitReport(std::move(Report));
-  }
-}
 
 void ento::registerReturnPointerRangeChecker(CheckerManager &mgr) {
   mgr.registerChecker<ReturnPointerRangeChecker>();
 }
 
 bool ento::shouldRegisterReturnPointerRangeChecker(const CheckerManager &mgr) {
   return true;
 }

clang/test/Analysis/return-ptr-range.cpp

if (I != 11) // expected-note{{Assuming 'I' is equal to 11}}
     return DataArr;
   return DataArr + I; // expected-warning{{Returned pointer value points outside the original object}}
                       // expected-note@-1{{Returned pointer value points outside the original object}}
                       // expected-note@-2{{Original object 'DataArr' is an array of 10 'test_array_of_struct::Data' objects, returned pointer points at index 11}}
 }
 
 }
 
+int *test_negative_index(int I) {
+  static int arr[10]; // expected-note{{Original object declared here}}
+                      // expected-note@-1{{'arr' initialized here}}
+                      // expected-note@-2{{Original object declared here}}
+                      // expected-note@-3{{'arr' initialized here}}
+  if (I == -1)        // expected-note{{Assuming the condition is true}}
+                      // expected-note@-1{{Taking true branch}}
+                      // expected-note@-2{{Assuming the condition is false}}
+                      // expected-note@-3{{Taking false branch}}
+    return arr + I;   // expected-warning{{Returned pointer value points outside the original object}}
+                      // expected-note@-1{{Returned pointer value points outside the original object}}
+                      // expected-note@-2{{Original object 'arr' is an array of 10 'int' objects, returned pointer points at index -1}}
+  if (I == -2)        // expected-note{{Assuming the condition is true}}
+                      // expected-note@-1{{Taking true branch}}
+    return arr + I;   // expected-warning{{Returned pointer value points outside the original object}}
+                      // expected-note@-1{{Returned pointer value points outside the original object}}
+                      // expected-note@-2{{Original object 'arr' is an array of 10 'int' objects, returned pointer points at index -2}}
+  if (I == 0)
+    return arr + I;
+  return arr + I;
+}