Reduce HT_MAX_SIZE to account for the max load factor of 0.5 (#10242)

zend_hash allocates a hash table twice as big as nTableSize
(HT_HASH_SIZE(HT_SIZE_TO_MASK(nTableSize)) == nTableSize*2), so HT_MAX_SIZE
must be half the max table size or less.

Fixes GH-10240

Author: arnaud-lb

Zend/zend_hash.c

@@ -166,6 +166,8 @@ static zend_always_inline void zend_hash_real_init_mixed_ex(HashTable *ht)
 	void *data;
 	uint32_t nSize = ht->nTableSize;
 
+	ZEND_ASSERT(HT_SIZE_TO_MASK(nSize));
+
 	if (UNEXPECTED(GC_FLAGS(ht) & IS_ARRAY_PERSISTENT)) {
 		data = pemalloc(HT_SIZE_EX(nSize, HT_SIZE_TO_MASK(nSize)), 1);
 	} else if (EXPECTED(nSize == HT_MIN_SIZE)) {
@@ -341,6 +343,8 @@ ZEND_API void ZEND_FASTCALL zend_hash_packed_to_hash(HashTable *ht)
 	Bucket *old_buckets = ht->arData;
 	uint32_t nSize = ht->nTableSize;
 
+	ZEND_ASSERT(HT_SIZE_TO_MASK(nSize));
+
 	HT_ASSERT_RC1(ht);
 	HT_FLAGS(ht) &= ~HASH_FLAG_PACKED;
 	new_data = pemalloc(HT_SIZE_EX(nSize, HT_SIZE_TO_MASK(nSize)), GC_FLAGS(ht) & IS_ARRAY_PERSISTENT);
@@ -369,7 +373,11 @@ ZEND_API void ZEND_FASTCALL zend_hash_to_packed(HashTable *ht)
 ZEND_API void ZEND_FASTCALL zend_hash_extend(HashTable *ht, uint32_t nSize, bool packed)
 {
 	HT_ASSERT_RC1(ht);
+
 	if (nSize == 0) return;
+
+	ZEND_ASSERT(HT_SIZE_TO_MASK(nSize));
+
 	if (UNEXPECTED(HT_FLAGS(ht) & HASH_FLAG_UNINITIALIZED)) {
 		if (nSize > ht->nTableSize) {
 			ht->nTableSize = zend_hash_check_size(nSize);
@@ -1207,6 +1215,8 @@ static void ZEND_FASTCALL zend_hash_do_resize(HashTable *ht)
 		uint32_t nSize = ht->nTableSize + ht->nTableSize;
 		Bucket *old_buckets = ht->arData;
 
+		ZEND_ASSERT(HT_SIZE_TO_MASK(nSize));
+
 		ht->nTableSize = nSize;
 		new_data = pemalloc(HT_SIZE_EX(nSize, HT_SIZE_TO_MASK(nSize)), GC_FLAGS(ht) & IS_ARRAY_PERSISTENT);
 		ht->nTableMask = HT_SIZE_TO_MASK(ht->nTableSize);

Zend/zend_types.h

@@ -400,16 +400,23 @@ struct _zend_array {
 #define HT_MIN_MASK ((uint32_t) -2)
 #define HT_MIN_SIZE 8
 
+/* HT_MAX_SIZE is chosen to satisfy the following constraints:
+ * - HT_SIZE_TO_MASK(HT_MAX_SIZE) != 0
+ * - HT_SIZE_EX(HT_MAX_SIZE, HT_SIZE_TO_MASK(HT_MAX_SIZE)) does not overflow or
+ *   wrapparound, and is <= the addressable space size
+ * - HT_MAX_SIZE must be a power of two:
+ *   (nTableSize<HT_MAX_SIZE ? nTableSize+nTableSize : nTableSize) <= HT_MAX_SIZE
+ */
 #if SIZEOF_SIZE_T == 4
-# define HT_MAX_SIZE 0x04000000 /* small enough to avoid overflow checks */
+# define HT_MAX_SIZE 0x02000000
 # define HT_HASH_TO_BUCKET_EX(data, idx) \
 	((Bucket*)((char*)(data) + (idx)))
 # define HT_IDX_TO_HASH(idx) \
 	((idx) * sizeof(Bucket))
 # define HT_HASH_TO_IDX(idx) \
 	((idx) / sizeof(Bucket))
 #elif SIZEOF_SIZE_T == 8
-# define HT_MAX_SIZE 0x80000000
+# define HT_MAX_SIZE 0x40000000
 # define HT_HASH_TO_BUCKET_EX(data, idx) \
 	((data) + (idx))
 # define HT_IDX_TO_HASH(idx) \
@@ -428,7 +435,7 @@ struct _zend_array {
 #define HT_SIZE_TO_MASK(nTableSize) \
 	((uint32_t)(-((nTableSize) + (nTableSize))))
 #define HT_HASH_SIZE(nTableMask) \
-	(((size_t)(uint32_t)-(int32_t)(nTableMask)) * sizeof(uint32_t))
+	(((size_t)-(uint32_t)(nTableMask)) * sizeof(uint32_t))
 #define HT_DATA_SIZE(nTableSize) \
 	((size_t)(nTableSize) * sizeof(Bucket))
 #define HT_SIZE_EX(nTableSize, nTableMask) \