[clang] [analyzer] Fix nullptr dereference for symbols from pointer invalidation (PR #106568)

[via cfe-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-clang-static-analyzer-1

@llvm/pr-subscribers-clang

Author: Arseniy Zaostrovnykh (necto)

<details>
<summary>Changes</summary>

As reported in https://github.com/llvm/llvm-project/pull/105648#issuecomment-2317144635 commit 08ad8dc7154bf3ab79f750e6d5fb7df597c7601a
introduced a nullptr dereference in the case when store contains a binding to a symbol that has no origin region associated with it, such as the symbol generated when a pointer is passed to an opaque function.

---
Full diff: https://github.com/llvm/llvm-project/pull/106568.diff


2 Files Affected:

- (modified) clang/lib/StaticAnalyzer/Checkers/StackAddrEscapeChecker.cpp (+4-1) 
- (modified) clang/test/Analysis/stack-addr-ps.c (+19) 


``````````diff
diff --git a/clang/lib/StaticAnalyzer/Checkers/StackAddrEscapeChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/StackAddrEscapeChecker.cpp
index 20232405d572d2..d3b185541729d3 100644
--- a/clang/lib/StaticAnalyzer/Checkers/StackAddrEscapeChecker.cpp
+++ b/clang/lib/StaticAnalyzer/Checkers/StackAddrEscapeChecker.cpp
@@ -308,7 +308,10 @@ static const MemSpaceRegion *getStackOrGlobalSpaceRegion(const MemRegion *R) {
 const MemRegion *getOriginBaseRegion(const MemRegion *Reg) {
   Reg = Reg->getBaseRegion();
   while (const auto *SymReg = dyn_cast<SymbolicRegion>(Reg)) {
-    Reg = SymReg->getSymbol()->getOriginRegion()->getBaseRegion();
+    const auto* OriginReg = SymReg->getSymbol()->getOriginRegion();
+    if (!OriginReg)
+      break;
+    Reg = OriginReg->getBaseRegion();
   }
   return Reg;
 }
diff --git a/clang/test/Analysis/stack-addr-ps.c b/clang/test/Analysis/stack-addr-ps.c
index 138b8c16b02bde..f47529623a6f57 100644
--- a/clang/test/Analysis/stack-addr-ps.c
+++ b/clang/test/Analysis/stack-addr-ps.c
@@ -126,3 +126,22 @@ void caller_for_nested_leaking() {
   int *ptr = 0;
   caller_mid_for_nested_leaking(&ptr);
 }
+
+// This used to crash StackAddrEscapeChecker because
+// it features a symbol conj_$1{struct c *, LC1, S763, #1}
+// that has no origin region.
+// bbi-98571
+struct a {
+  int member;
+};
+
+struct c {
+  struct a *nested_ptr;
+};
+long global_var;
+void opaque(struct c*);
+void bbi_98571_no_crash() {
+  struct c *ptr = (struct c *)global_var;
+  opaque(ptr);
+  ptr->nested_ptr->member++;
+}

``````````

</details>


https://github.com/llvm/llvm-project/pull/106568