kernel/slab: Fix potential memory leak in HashedObjectCache.

If the resize-needed amount changed while we were allocating memory,
then we'd leak the new buffer.

In testing, I added an ASSERT() to check if this case actually happened,
and it didn't seem to fire when using the system (with a debug kernel
though, admittedly.)

Inspired by a change suggested on GitHub, but implemented in a
completely different way (that cleans up the code at the same time.)

src/system/kernel/slab/HashedObjectCache.cpp

@@ -173,22 +173,26 @@ void
 HashedObjectCache::_ResizeHashTableIfNeeded(uint32 flags)
 {
 	size_t hashSize = hash_table.ResizeNeeded();
-	if (hashSize != 0) {
+	if (hashSize == 0)
-		Unlock();
+		return;
-		void* buffer = slab_internal_alloc(hashSize, flags);
-		Lock();
 
-		if (buffer != NULL) {
+	Unlock();
-			if (hash_table.ResizeNeeded() == hashSize) {
+	void* buffer = slab_internal_alloc(hashSize, flags);
-				void* oldHash;
+	Lock();
-				hash_table.Resize(buffer, hashSize, true, &oldHash);
+
-				if (oldHash != NULL) {
+	if (buffer == NULL)
-					Unlock();
+		return;
-					slab_internal_free(oldHash, flags);
+
-					Lock();
+	if (hash_table.ResizeNeeded() == hashSize) {
-				}
+		void* oldHash = NULL;
-			}
+		hash_table.Resize(buffer, hashSize, true, &oldHash);
-		}
+		buffer = oldHash;
+	}
+
+	if (buffer != NULL) {
+		Unlock();
+		slab_internal_free(buffer, flags);
+		Lock();
 	}
 }