Loading...
--- libmalloc/libmalloc-317.100.9/src/magazine_inline.h
+++ libmalloc/libmalloc-646.0.13/src/magazine_inline.h
@@ -24,41 +24,10 @@
 #ifndef __MAGAZINE_INLINE_H
 #define __MAGAZINE_INLINE_H
 
+#include <malloc/_ptrcheck.h>
+__ptrcheck_abi_assume_single()
+
 extern unsigned int _os_cpu_number_override;
-
-/*
- * MALLOC_ABSOLUTE_MAX_SIZE - There are many instances of addition to a
- * user-specified size_t, which can cause overflow (and subsequent crashes)
- * for values near SIZE_T_MAX.  Rather than add extra "if" checks everywhere
- * this occurs, it is easier to just set an absolute maximum request size,
- * and immediately return an error if the requested size exceeds this maximum.
- * Of course, values less than this absolute max can fail later if the value
- * is still too large for the available memory.  The largest value added
- * seems to be large_vm_page_quanta_size (in the macro round_large_page_quanta()), so to be safe, we set
- * the maximum to be 2 * PAGE_SIZE less than SIZE_T_MAX.
- */
-#define MALLOC_ABSOLUTE_MAX_SIZE (SIZE_T_MAX - (2 * large_vm_page_quanta_size))
-
-// Gets the allocation size for a calloc(). Multiples size by num_items and adds
-// extra_size, storing the result in *total_size. Returns 0 on success, -1 (with
-// errno set to ENOMEM) on overflow.
-static int MALLOC_INLINE MALLOC_ALWAYS_INLINE
-calloc_get_size(size_t num_items, size_t size, size_t extra_size, size_t *total_size)
-{
-	size_t alloc_size = size;
-	if (num_items != 1 && (os_mul_overflow(num_items, size, &alloc_size)
-			|| alloc_size > MALLOC_ABSOLUTE_MAX_SIZE)) {
-		errno = ENOMEM;
-		return -1;
-	}
-	if (extra_size && (os_add_overflow(alloc_size, extra_size, &alloc_size)
-			|| alloc_size > MALLOC_ABSOLUTE_MAX_SIZE)) {
-		errno = ENOMEM;
-		return -1;
-	}
-	*total_size = alloc_size;
-	return 0;
-}
 
 /*********************	FREE LIST UTILITIES  ************************/
 
@@ -95,19 +64,6 @@
 static MALLOC_INLINE void recirc_list_splice_last(rack_t *rack, magazine_t *mag_ptr, region_trailer_t *node) MALLOC_ALWAYS_INLINE;
 static MALLOC_INLINE void recirc_list_splice_first(rack_t *rack, magazine_t *mag_ptr, region_trailer_t *node) MALLOC_ALWAYS_INLINE;
 
-static MALLOC_INLINE void
-yield(void)
-{
-	thread_switch(MACH_PORT_NULL, SWITCH_OPTION_DEPRESS, 1);
-}
-
-static MALLOC_INLINE kern_return_t
-_malloc_default_reader(task_t task, vm_address_t address, vm_size_t size, void **ptr)
-{
-	*ptr = (void *)address;
-	return 0;
-}
-
 #pragma mark helpers
 
 static MALLOC_INLINE MALLOC_ALWAYS_INLINE
@@ -197,23 +153,18 @@
 			"Corrupt value: %p\n", ptr, value);
 }
 
+// TODO: replace uses in small and medium with data PAC when possible
 static MALLOC_INLINE uintptr_t
 free_list_gen_checksum(uintptr_t ptr)
 {
-	uint8_t chk;
-
-	chk = (unsigned char)(ptr >> 0);
-	chk += (unsigned char)(ptr >> 8);
-	chk += (unsigned char)(ptr >> 16);
-	chk += (unsigned char)(ptr >> 24);
-#if __LP64__
-	chk += (unsigned char)(ptr >> 32);
-	chk += (unsigned char)(ptr >> 40);
-	chk += (unsigned char)(ptr >> 48);
-	chk += (unsigned char)(ptr >> 56);
+#if defined(__LP64__)
+	uint32_t level1 = (uint32_t)ptr + ((uint32_t)(ptr >> 32));
+#else
+	uint32_t level1 = (uint32_t)ptr;
 #endif
-
-	return chk;
+	uint16_t level2 = (uint16_t)level1 + ((uint16_t)(level1 >> 16));
+	uint8_t level3 = (uint8_t)level2 + ((uint8_t)(level2 >> 8));
+	return level3;
 }
 
 static unsigned
@@ -227,7 +178,7 @@
 	while (ptr.p) {
 		count++;
 		if (reader(task, (vm_address_t)ptr.inplace, sizeof(*ptr.inplace),
-				(void **)&mapped_inplace_free_entry)) {
+				__unsafe_forge_single(void **, &mapped_inplace_free_entry))) {
 			printer("** invalid pointer in free list: %p\n", ptr.inplace);
 			break;
 		}
@@ -236,8 +187,43 @@
 	return count;
 }
 
+#if __has_feature(ptrauth_calls) && defined(__arm64e__) && !TARGET_OS_SIMULATOR
+
+// We can use data PAC to protect the free list pointers
+static MALLOC_INLINE uintptr_t
+free_list_checksum_ptr(rack_t *rack, void *ptr)
+{
+	uintptr_t signed_ptr = (uintptr_t)ptrauth_sign_unauthenticated(ptr,
+			ptrauth_key_process_dependent_data,
+			ptrauth_blend_discriminator(rack,
+					ptrauth_string_discriminator("malloc freelist")));
+	return signed_ptr;
+}
+
+static MALLOC_INLINE void *
+free_list_unchecksum_ptr(rack_t *rack, inplace_union *ptr)
+{
+	void * __single stored_ptr = ptr->p;
+	// N.B. we don't use ptrauth_auth_data() because we want to be able to call
+	// free_list_checksum_botch() on failure, which prints a diagnostic first,
+	// rather than trapping directly
+	void * __single stripped_ptr = ptrauth_strip(stored_ptr, ptrauth_key_process_dependent_data);
+	uintptr_t resigned_ptr = free_list_checksum_ptr(rack, stripped_ptr);
+	if ((uintptr_t)stored_ptr != resigned_ptr) {
+		free_list_checksum_botch(rack, ptr, __unsafe_forge_single(void *, ptr->u));
+		__builtin_trap();
+	}
+	return stripped_ptr;
+}
+
+#else // __has_feature(ptrauth_calls) && defined(__arm64e__) && !TARGET_OS_SIMULATOR
+
+// We can't use data PAC so we manually calculate and store a checksum instead
+// TODO: use the high bits on LP64
+// TODO: this can likely still be faster
+
 #define NYBBLE 4
-#if __LP64__
+#if defined(__LP64__)
 #define ANTI_NYBBLE (64 - NYBBLE)
 #else
 #define ANTI_NYBBLE (32 - NYBBLE)
@@ -260,7 +246,7 @@
 	p.u = t & ~(uintptr_t)0xF;
 
 	if ((t ^ free_list_gen_checksum(p.u ^ rack->cookie)) & (uintptr_t)0xF) {
-		free_list_checksum_botch(rack, ptr, (void *)ptr->u);
+		free_list_checksum_botch(rack, ptr, __unsafe_forge_single(void *, ptr->u));
 		__builtin_trap();
 	}
 	return p.p;
@@ -268,6 +254,8 @@
 
 #undef ANTI_NYBBLE
 #undef NYBBLE
+
+#endif // __has_feature(ptrauth_calls) && defined(__arm64e__) && !TARGET_OS_SIMULATOR
 
 #pragma mark recirc helpers
 
@@ -339,7 +327,7 @@
  * cache would likely be a significant performance win here.
  */
 static MALLOC_INLINE rgnhdl_t
-hash_lookup_region_no_lock(region_t *regions, size_t num_entries, size_t shift, region_t r)
+hash_lookup_region_no_lock(region_t * __counted_by(num_entries) regions, size_t num_entries, size_t shift, region_t r)
 {
 	size_t index, hash_index;
 	rgnhdl_t entry;
@@ -351,7 +339,7 @@
 	// Multiplicative hash where the multiplier is a prime near (ULONG_MAX / phi). [phi = 1.618033...]
 	// Since the values of (((uintptr_t)r >> HASH_BLOCKS_ALIGN) are (roughly) an ascending sequence of integers,
 	// this hash works really well. See Knuth TAOCP, Vol. 3.
-#if __LP64__
+#if defined(__LP64__)
 	index = hash_index = (((uintptr_t)r >> HASH_BLOCKS_ALIGN) * 11400714819323198549ULL) >> (64 - shift);
 #else
 	index = hash_index = (((uintptr_t)r >> HASH_BLOCKS_ALIGN) * 2654435761UL) >> (32 - shift);
@@ -375,7 +363,7 @@
  * hash_region_insert_no_lock - Insert a region into the hash ring.
  */
 static void
-hash_region_insert_no_lock(region_t *regions, size_t num_entries, size_t shift, region_t r)
+hash_region_insert_no_lock(region_t * __counted_by(num_entries) regions, size_t num_entries, size_t shift, region_t r)
 {
 	size_t index, hash_index;
 	rgnhdl_t entry;
@@ -383,7 +371,7 @@
 	// Multiplicative hash where the multiplier is a prime near (ULONG_MAX / phi). [phi = 1.618033...]
 	// Since the values of (((uintptr_t)r >> HASH_BLOCKS_ALIGN) are (roughly) an ascending sequence of integers,
 	// this hash works really well. See Knuth TAOCP, Vol. 3.
-#if __LP64__
+#if defined(__LP64__)
 	index = hash_index = (((uintptr_t)r >> HASH_BLOCKS_ALIGN) * 11400714819323198549ULL) >> (64 - shift);
 #else
 	index = hash_index = (((uintptr_t)r >> HASH_BLOCKS_ALIGN) * 2654435761UL) >> (32 - shift);
@@ -419,7 +407,7 @@
  * the old entries since someone may still be allocating them.
  */
 static MALLOC_INLINE region_t *
-hash_regions_grow_no_lock(region_t *regions, size_t old_size, size_t *mutable_shift, size_t *new_size)
+hash_regions_grow_no_lock(region_t * __counted_by(old_size) regions, size_t old_size, size_t *mutable_shift, size_t *new_size)
 {
 	// double in size and allocate memory for the regions
 	*new_size = old_size + old_size;
@@ -439,15 +427,6 @@
 
 #pragma mark mag index
 
-/*
- * These commpage routines provide fast access to the logical cpu number
- * of the calling processor assuming no pre-emption occurs.
- */
-
-extern unsigned int hyper_shift;
-extern unsigned int phys_ncpus;
-extern unsigned int logical_ncpus;
-
 static MALLOC_INLINE MALLOC_ALWAYS_INLINE
 unsigned int
 mag_max_magazines(void)
@@ -465,10 +444,10 @@
 #pragma mark mag lock
 
 static MALLOC_INLINE magazine_t *
-mag_lock_zine_for_region_trailer(magazine_t *magazines, region_trailer_t *trailer, mag_index_t mag_index)
+mag_lock_zine_for_region_trailer(magazine_t * __unsafe_indexable magazines, region_trailer_t *trailer, mag_index_t mag_index)
 {
 	mag_index_t refreshed_index;
-	magazine_t *mag_ptr = &(magazines[mag_index]);
+	magazine_t *mag_ptr = __unsafe_forge_single(magazine_t *, &(magazines[mag_index]));
 
 	// Take the lock  on entry.
 	SZONE_MAGAZINE_PTR_LOCK(mag_ptr);
@@ -481,7 +460,7 @@
 		SZONE_MAGAZINE_PTR_UNLOCK(mag_ptr);
 
 		mag_index = refreshed_index;
-		mag_ptr = &(magazines[mag_index]);
+		mag_ptr = __unsafe_forge_single(magazine_t *, &(magazines[mag_index]));
 		SZONE_MAGAZINE_PTR_LOCK(mag_ptr);
 	}
 
@@ -489,8 +468,6 @@
 }
 
 #pragma mark Region Cookie
-
-extern uint64_t malloc_entropy[2];
 
 static region_cookie_t
 region_cookie(void)
@@ -539,8 +516,8 @@
 static MALLOC_INLINE msize_t
 get_tiny_free_size_offset(const void *ptr, off_t mapped_offset)
 {
-	void *next_block = (void *)((uintptr_t)ptr + TINY_QUANTUM);
-	void *region_end = TINY_REGION_HEAP_END(TINY_REGION_FOR_PTR(ptr));
+	void * __single next_block = __unsafe_forge_single(void *, (uintptr_t)ptr + TINY_QUANTUM);
+	void * __single region_end = TINY_REGION_HEAP_END(TINY_REGION_FOR_PTR(ptr));
 
 	// check whether the next block is outside the tiny region or a block header
 	// if so, then the size of this block is one, and there is no stored size.
@@ -710,4 +687,28 @@
 	return r ? *r : r;
 }
 
+#pragma mark zero on free
+
+MALLOC_NOEXPORT
+extern unsigned malloc_zero_on_free_sample_period;
+
+static MALLOC_INLINE bool
+zero_on_free_should_sample(void)
+{
+	bool sample = false;
+	if (malloc_zero_on_free_sample_period != 0) {
+		uintptr_t value = (uintptr_t)_pthread_getspecific_direct(
+				__TSD_MALLOC_ZERO_CORRUPTION_COUNTER);
+		value++;
+		if (value == malloc_zero_on_free_sample_period) {
+			sample = true;
+			value = 0;
+		}
+		_pthread_setspecific_direct(__TSD_MALLOC_ZERO_CORRUPTION_COUNTER,
+				(void *)value);
+	}
+
+	return sample;
+}
+
 #endif // __MAGAZINE_INLINE_H