[compiler-rt] [scudo] Add hooks to mark the range of realloc (PR #73883)

[via llvm-commits]

================
@@ -27,6 +27,51 @@ static void reportDeallocation(void *ptr) {
     if (__scudo_deallocate_hook)
       __scudo_deallocate_hook(ptr);
 }
+static void reportReallocBegin(void *old_ptr) {
+  if (SCUDO_ENABLE_HOOKS)
+    if (__scudo_realloc_begin_hook)
+      __scudo_realloc_begin_hook(old_ptr);
+}
+static void reportReallocEnd(void *old_ptr) {
+  if (SCUDO_ENABLE_HOOKS)
+    if (__scudo_realloc_end_hook)
+      __scudo_realloc_end_hook(old_ptr);
+}
+
+static void *reallocImpl(void *ptr, size_t size) {
+  if (!ptr) {
+    void *Ptr = SCUDO_ALLOCATOR.allocate(size, scudo::Chunk::Origin::Malloc,
+                                         SCUDO_MALLOC_ALIGNMENT);
+    reportAllocation(Ptr, size);
+    return Ptr;
+  }
+  if (size == 0) {
+    reportDeallocation(ptr);
+    SCUDO_ALLOCATOR.deallocate(ptr, scudo::Chunk::Origin::Malloc);
+    return nullptr;
+  }
+
+  // Given that the reporting of deallocation and allocation are not atomic, we
+  // always pretend the old pointer will be released so that the user don't need
+  // to worry about the false double-use case from the view of hooks.
+  // For example, before the reporting all the events in the `realloc` has
+  // finished, another thread may have got the pointer released by the
----------------
ChiaHungDuan wrote:

In this case, I try to say that another thread gets the old pointer released by realloc.
Reword it a little bit. Let me know if it's more clear

https://github.com/llvm/llvm-project/pull/73883