Loading...
--- libmalloc/libmalloc-792.1.1/src/malloc.c
+++ libmalloc/libmalloc-792.41.1/src/malloc.c
@@ -69,6 +69,9 @@
static malloc_zone_t *default_purgeable_zone;
static bool has_injected_zone0;
static bool malloc_xzone_enabled = MALLOC_XZONE_ENABLED_DEFAULT;
+#if CONFIG_XZONE_MALLOC && CONFIG_NANOZONE
+static bool malloc_nano_on_xzone = false;
+#endif // CONFIG_XZONE_MALLOC && CONFIG_NANOZONE
typedef enum {
MALLOC_XZONE_OVERRIDE_DEFAULT,
@@ -88,6 +91,11 @@
bool malloc_interposition_compat = false;
+#if CONFIG_MTE
+bool malloc_has_sec_transition = false;
+uint32_t malloc_sec_transition_policy = 0;
+bool malloc_sec_transition_early_malloc_support = false;
+#endif
#if CONFIG_MALLOC_PROCESS_IDENTITY
malloc_process_identity_t malloc_process_identity = MALLOC_PROCESS_NONE;
@@ -189,6 +197,15 @@
static const char zero_on_free_enabled_boot_arg[] = "malloc_zero_on_free_enabled";
+#if CONFIG_XZONE_MALLOC
+static const char secure_allocator_boot_arg[] = "malloc_secure_allocator";
+
+static
+bool xzm_create_mzones = true;
+
+static
+bool purgeable_zone_use_xzm = true;
+#endif // CONFIG_XZONE_MALLOC
#if CONFIG_MEDIUM_ALLOCATOR
static const char medium_enabled_boot_arg[] = "malloc_medium_zone";
@@ -354,6 +371,20 @@
}
}
+#if CONFIG_XZONE_MALLOC
+ flag = malloc_common_value_for_key_copy(bootargs,
+ secure_allocator_boot_arg, value_buf, sizeof(value_buf));
+ if (flag) {
+ const char *endp;
+ long value = malloc_common_convert_to_long(flag, &endp);
+ if (!*endp && (value == 0 || value == 1)) {
+ malloc_xzone_enabled = (bool)value;
+ } else {
+ malloc_report(ASL_LEVEL_ERR,
+ "malloc_secure_allocator must be 0 or 1 - ignored.\n");
+ }
+ }
+#endif // CONFIG_XZONE_MALLOC
#if CONFIG_MEDIUM_ALLOCATOR
#if TARGET_OS_OSX
@@ -1292,6 +1323,10 @@
/********* Creation and destruction ************/
static void set_flags_from_environment(void);
+#if CONFIG_MTE
+static bool _malloc_check_has_sec_transition(const char *apple[]);
+static uint32_t _malloc_get_sec_transition_policy(const char *apple[]);
+#endif
MALLOC_NOEXPORT void
malloc_zone_register_while_locked(malloc_zone_t *zone, bool make_default)
@@ -1310,12 +1345,28 @@
if (malloc_num_zones == malloc_num_zones_allocated) {
size_t malloc_zones_size = malloc_num_zones * sizeof(malloc_zone_t *);
- mach_vm_size_t alloc_size = round_page(malloc_zones_size + vm_page_size);
- mach_vm_address_t vm_addr;
-
- vm_addr = (mach_vm_address_t)mvm_allocate_plat(0, (size_t)alloc_size, 0,
- VM_FLAGS_ANYWHERE, MALLOC_GUARDED_METADATA,
- VM_MEMORY_MALLOC, NULL);
+
+ size_t new_malloc_num_zones_allocated = 0;
+ if (malloc_num_zones_allocated == 0) {
+ new_malloc_num_zones_allocated =
+ vm_page_size / sizeof(malloc_zone_t *);
+ } else if (malloc_num_zones_allocated < (1 << 17)) {
+ // Double up to 128K zones
+ new_malloc_num_zones_allocated = malloc_num_zones_allocated * 2;
+ } else {
+ // Increment in further batches of 128K zones
+ MALLOC_ASSERT(!os_add_overflow(malloc_num_zones_allocated, (1 << 17),
+ &new_malloc_num_zones_allocated));
+ }
+
+ mach_vm_size_t alloc_size = 0;
+ MALLOC_ASSERT(!os_mul_overflow(new_malloc_num_zones_allocated,
+ sizeof(malloc_zone_t *), &alloc_size));
+ alloc_size = round_page(alloc_size);
+
+ mach_vm_address_t vm_addr = (mach_vm_address_t)mvm_allocate_plat(
+ 0, (size_t)alloc_size, 0, VM_FLAGS_ANYWHERE,
+ MALLOC_GUARDED_METADATA, VM_MEMORY_MALLOC, NULL);
if (vm_addr == 0) {
malloc_report(ASL_LEVEL_ERR, "malloc_zone_register allocation failed\n");
return;
@@ -1326,16 +1377,14 @@
* out of the previous array and into the new zones array */
if (malloc_zones) {
memcpy(new_zones, malloc_zones, malloc_zones_size);
- vm_addr = (mach_vm_address_t)malloc_zones;
- mach_vm_size_t dealloc_size = round_page(malloc_zones_size);
- mach_vm_deallocate(mach_task_self(), vm_addr, dealloc_size);
}
/* Update the malloc_zones pointer, which we leak if it was previously
* allocated, and the number of zones allocated */
protect_size = (size_t)alloc_size;
malloc_zones = new_zones;
- malloc_num_zones_allocated = (int32_t)(alloc_size / sizeof(malloc_zone_t *));
+ MALLOC_ASSERT(!os_convert_overflow(new_malloc_num_zones_allocated,
+ &malloc_num_zones_allocated));
} else {
/* If we don't need to reallocate zones, we need to briefly change the
* page protection the malloc zones to allow writes */
@@ -1447,6 +1496,19 @@
_malloc_detect_interposition();
+#if CONFIG_MTE
+ malloc_has_sec_transition = _malloc_check_has_sec_transition(apple);
+ // If the process has been spawned by setting has_sec_transition=1,
+ // enable the usage of xzone and the nano override by default.
+ // Note that we still allow overriding this behaviour through
+ // environment variables (whose values are loaded afterwards).
+ if (malloc_has_sec_transition) {
+ malloc_xzone_enabled = true;
+ malloc_xzone_nano_override = MALLOC_XZONE_OVERRIDE_ENABLED;
+ malloc_sec_transition_early_malloc_support = true;
+ malloc_sec_transition_policy = _malloc_get_sec_transition_policy(apple);
+ }
+#endif
set_flags_from_environment();
@@ -1485,6 +1547,101 @@
bool nano_on_xzone = false;
+#if CONFIG_XZONE_MALLOC
+#if CONFIG_NANOZONE
+ bool nano_xzone_enabled = false;
+#if CONFIG_FEATUREFLAGS_SIMPLE
+ nano_xzone_enabled = os_feature_enabled_simple(libmalloc,
+ SecureAllocator_Nano, MALLOC_XZONE_NANO_ENABLED_DEFAULT);
+#elif MALLOC_TARGET_DK_OSX
+ nano_xzone_enabled = true;
+#endif // CONFIG_FEATUREFLAGS_SIMPLE
+
+#if CONFIG_MALLOC_PROCESS_IDENTITY
+ if (malloc_process_is_security_critical(malloc_process_identity)) {
+ // This is load-bearing for MTLCompilerService
+ nano_xzone_enabled = true;
+ }
+#endif // CONFIG_MALLOC_PROCESS_IDENTITY
+
+ switch (malloc_xzone_nano_override) {
+ case MALLOC_XZONE_OVERRIDE_DISABLED:
+ nano_xzone_enabled = false;
+ break;
+ case MALLOC_XZONE_OVERRIDE_ENABLED:
+ nano_xzone_enabled = true;
+ break;
+ default:
+ break;
+ }
+
+ // If we're forcing nano-on-xzone via the environment, take that even if
+ // we'd otherwise take the nano config of xzone
+ if (malloc_nano_on_xzone_override == MALLOC_XZONE_OVERRIDE_ENABLED) {
+ nano_xzone_enabled = false;
+ }
+
+ if (malloc_xzone_enabled && _malloc_engaged_nano == NANO_V2 &&
+ !nano_xzone_enabled) {
+#if TARGET_OS_SIMULATOR
+ // Note: simulator processes get nano by default because
+ // NANOV2_DEFAULT_MODE is conditioned on MALLOC_TARGET_IOS, which
+ // excludes the simulator. So, simulator processes will not get xzone
+ // malloc until we either adjust this policy or enable it by default
+ // under nano.
+ malloc_xzone_enabled = false;
+#else // TARGET_OS_SIMULATOR
+#if CONFIG_FEATUREFLAGS_SIMPLE
+ nano_on_xzone = os_feature_enabled_simple(libmalloc,
+ SecureAllocator_NanoOnXzone, false);
+#endif // CONFIG_FEATUREFLAGS_SIMPLE
+
+ switch (malloc_nano_on_xzone_override) {
+ case MALLOC_XZONE_OVERRIDE_DISABLED:
+ nano_on_xzone = false;
+ break;
+ case MALLOC_XZONE_OVERRIDE_ENABLED:
+ nano_on_xzone = true;
+ break;
+ default:
+ break;
+ }
+
+ if (malloc_nano_on_xzone != nano_on_xzone) {
+ malloc_nano_on_xzone = nano_on_xzone;
+ }
+
+ if (!nano_on_xzone) {
+ malloc_xzone_enabled = false;
+ }
+#endif // TARGET_OS_SIMULATOR
+ }
+#endif // CONFIG_NANOZONE
+
+ if (malloc_xzone_enabled) {
+ mvm_guarded_range_init();
+
+ if (malloc_report_config && malloc_internal_security_policy) {
+ malloc_report(ASL_LEVEL_INFO,
+ "Guarded Range Config (base/size/carveout):"
+ " 0x%lx / 0x%lx / 0x%lx\n",
+ (unsigned long)malloc_guarded_range_config.base_address,
+ (unsigned long)malloc_guarded_range_config.size,
+ (unsigned long)malloc_guarded_range_config.carveout_address);
+ }
+
+ initial_xzone_zone = xzm_main_malloc_zone_create(malloc_debug_flags,
+ envp, apple, bootargs);
+
+ // Note: although it would be ideal if we could give this zone a unique
+ // name that draws attention to the fact that it will not behave like a
+ // nano or scalable zone, a lot of code looks specifically for
+ // "DefaultMallocZone" when looking for/at the default zone, so that's
+ // the name we'll need to continue to take.
+ malloc_set_zone_name(initial_xzone_zone, DEFAULT_MALLOC_ZONE_STRING);
+ malloc_zone_register_while_locked(initial_xzone_zone, /*make_default=*/true);
+ }
+#endif // CONFIG_XZONE_MALLOC
if (!initial_xzone_zone || nano_on_xzone) {
if (!initial_xzone_zone) {
@@ -2085,7 +2242,53 @@
}
}
-
+#if CONFIG_XZONE_MALLOC
+ flag = getenv("MallocSecureAllocatorCreateMzones");
+ if (flag && malloc_internal_security_policy) {
+ const char *endp;
+ long value = malloc_common_convert_to_long(flag, &endp);
+ if (!*endp && endp != flag && (value == 0 || value == 1)) {
+ xzm_create_mzones = value;
+ } else {
+ malloc_report(ASL_LEVEL_ERR,
+ "MallocSecureAllocatorCreateMzones must be 0 or 1.\n");
+ }
+ }
+ flag = getenv("MallocSecureAllocatorPurgeableZone");
+ if (flag && malloc_internal_security_policy) {
+ const char *endp;
+ long value = malloc_common_convert_to_long(flag, &endp);
+ if (!*endp && endp != flag && (value == 0 || value == 1)) {
+ purgeable_zone_use_xzm = value;
+ } else {
+ malloc_report(ASL_LEVEL_ERR,
+ "MallocSecureAllocatorPurgeableZone must be 0 or 1.\n");
+ }
+ }
+#endif // CONFIG_XZONE_MALLOC
+
+#if CONFIG_MTE
+ flag = getenv("MallocEarlyMallocSecTransitionSupport");
+ if (flag) {
+ const char *endp;
+ long value = malloc_common_convert_to_long(flag, &endp);
+ if (!*endp && endp != flag && (value == 0 || value == 1)) {
+ malloc_sec_transition_early_malloc_support = value;
+ } else {
+ malloc_report(ASL_LEVEL_ERR,
+ "MallocEarlyMallocSecTransitionSupport must be 0 or 1.\n");
+ }
+ }
+#endif // CONFIG_MTE
+
+ flag = getenv("MallocEnableMSLAtLimitWarning");
+ if (flag) {
+ const char *endp;
+ long value = malloc_common_convert_to_long(flag, &endp);
+ if (!*endp && endp != flag && value == 1) {
+ malloc_memorystatus_mask_resource_exception_handling |= MALLOC_MEMORYSTATUS_MASK_RESOURCE_EXCEPTION_HANDLING;
+ }
+ }
if (getenv("MallocHelp")) {
malloc_report(ASL_LEVEL_INFO,
@@ -2113,6 +2316,37 @@
}
}
+#if CONFIG_MTE
+static bool
+_malloc_check_has_sec_transition(const char *apple[])
+{
+ const char *flag = _simple_getenv(apple, "has_sec_transition");
+ if (flag) {
+ const char *endp;
+ long value = malloc_common_convert_to_long(flag, &endp);
+ if (!*endp && endp != flag && (value == 0 || value == 1)) {
+ return (bool)value;
+ }
+ }
+ return false;
+}
+
+static uint32_t
+_malloc_get_sec_transition_policy(const char *apple[])
+{
+ const char *flag = _simple_getenv(apple, "sec_transition_policy");
+ uint32_t policy = 0;
+ if (flag) {
+ char *endp;
+ unsigned long value = strtoul(flag, &endp, 0);
+ if (!*endp && endp != flag) {
+ policy = (uint32_t)value;
+ }
+ }
+ return policy;
+}
+
+#endif // CONFIG_MTE
malloc_zone_t *
malloc_create_zone(vm_size_t start_size, unsigned flags)
@@ -2124,6 +2358,12 @@
return NULL;
}
+#if CONFIG_XZONE_MALLOC
+ if (initial_xzone_zone && xzm_create_mzones) {
+ zone = xzm_malloc_zone_create(flags | malloc_debug_flags,
+ (xzm_main_malloc_zone_t)initial_xzone_zone);
+ }
+#endif // CONFIG_XZONE_MALLOC
if (!zone) {
zone = create_scalable_zone(start_size, flags | malloc_debug_flags);
@@ -2784,6 +3024,11 @@
{
// TODO: save and restore permissions generally
bool mprotect_zone = true;
+#if CONFIG_XZONE_MALLOC
+ if (_xzm_malloc_zone_is_xzm(z)) {
+ mprotect_zone = false;
+ }
+#endif // CONFIG_XZONE_MALLOC
if (mprotect_zone) {
mprotect(z, sizeof(malloc_zone_t), PROT_READ | PROT_WRITE);
}
@@ -2854,6 +3099,20 @@
if ((malloc_debug_flags & (MALLOC_ABORT_ON_CORRUPTION | MALLOC_ABORT_ON_ERROR))) {
flags = MALLOC_REPORT_CRASH | MALLOC_REPORT_NOLOG;
}
+#if CONFIG_MTE
+ // When tagging is enabled, this will validate the MTE tag of the
+ // pointer and raise a fatal exception if it doesn't match the tag
+ // stored in memory.
+ //
+ // We need to do this here to cover the following deviations along the
+ // free() path:
+ // - when we are called directly by _free(), in the cases where we can't
+ // invoke zone->try_free_default()
+ // - in xzone, in the outlined path after the pointer lookup has failed,
+ // if we're in a try context
+ // - in xzone, from the slow path for try_free_default(), if the zone
+ // lookup fails when trying to determine the size of the allocation
+#endif
malloc_report_pointer_was_not_allocated(flags, ptr);
} else if (zone->version >= 6 && zone->free_definite_size) {
malloc_zone_free_definite_size(zone, ptr, size);
@@ -2953,6 +3212,15 @@
} else {
zone = _find_registered_zone(in_ptr, NULL, false);
if (!zone) {
+#if CONFIG_MTE
+ // This will validate the MTE tag of the pointer: this is required
+ // to make sure we raise a fatal exception when trying to reallocate
+ // a pointer with an invalid tag.
+ // Note that this is the only place where we handle this in the
+ // realloc path: for malloc_type_realloc, this means that we only
+ // validate the MTE tag when the zone lookup fails in the fast path,
+ // or in the slow path only if the symbol has not been interposed.
+#endif
int flags = MALLOC_REPORT_DEBUG | MALLOC_REPORT_NOLOG;
const int abort_flags =
(MALLOC_ABORT_ON_CORRUPTION | MALLOC_ABORT_ON_ERROR);
@@ -3171,8 +3439,7 @@
// descriptor work here. That's okay, as it's uncommon and SPI, so its
// callers should be built with TMO.
const malloc_zone_malloc_options_t known_options =
- MALLOC_ZONE_MALLOC_OPTION_CLEAR
- ;
+ MALLOC_ZONE_MALLOC_OPTION_CLEAR | MALLOC_NP_OPTION_CANONICAL_TAG;
if (options & ~known_options) {
malloc_zone_error(MALLOC_ABORT_ON_ERROR, true,
"malloc_zone_malloc_with_options: unsupported options 0x%llx\n",
@@ -3180,6 +3447,20 @@
__builtin_trap();
}
+#if CONFIG_MTE
+ // rdar://140822174
+ // When dyld interposition or a wrapper zone that does not support
+ // forwarding malloc options is enabled, we need to set a flag in the
+ // TSD to preserve the semantics of canonical tagging.
+ malloc_thread_options_t opts;
+ if ((options & MALLOC_ZONE_MALLOC_OPTION_CANONICAL_TAG) &&
+ malloc_has_sec_transition) {
+ opts = malloc_get_thread_options();
+ malloc_thread_options_t newopts = opts;
+ newopts.ReservedFlag = true;
+ _malloc_set_thread_options(newopts);
+ }
+#endif // CONFIG_MTE
if (align > MALLOC_ZONE_MALLOC_DEFAULT_ALIGN) {
ptr = malloc_zone_memalign(zone, align, size);
@@ -3192,6 +3473,13 @@
ptr = malloc_zone_malloc(zone, size);
}
+#if CONFIG_MTE
+ // Restore the saved TSD flags
+ if ((options & MALLOC_ZONE_MALLOC_OPTION_CANONICAL_TAG) &&
+ malloc_has_sec_transition) {
+ _malloc_set_thread_options(opts);
+ }
+#endif // CONFIG_MTE
} else {
MALLOC_TRACE(TRACE_malloc_options | DBG_FUNC_START, (uintptr_t)zone,
align, size, 0);
@@ -3227,8 +3515,9 @@
malloc_zone_malloc_with_options(malloc_zone_t *zone, size_t align,
size_t size, malloc_zone_malloc_options_t options)
{
- if (os_unlikely((align != 0) && (!powerof2(align) ||
- ((size & (align-1)) != 0)))) { // equivalent to (size % align != 0)
+ if (align != MALLOC_ZONE_MALLOC_DEFAULT_ALIGN &&
+ (os_unlikely((align != 0) && (!powerof2(align) ||
+ ((size & (align-1)) != 0))))) { // equivalent to (size % align != 0)
return NULL;
}
@@ -3262,6 +3551,17 @@
static void
_malloc_create_purgeable_zone(void * __unused ctx)
{
+#if CONFIG_XZONE_MALLOC
+ if (initial_xzone_zone && purgeable_zone_use_xzm) {
+ unsigned flags = malloc_debug_flags | MALLOC_PURGEABLE;
+ default_purgeable_zone = xzm_malloc_zone_create(flags,
+ (xzm_main_malloc_zone_t)initial_xzone_zone);
+ malloc_zone_register(default_purgeable_zone);
+ malloc_set_zone_name(default_purgeable_zone,
+ DEFAULT_PUREGEABLE_ZONE_STRING);
+ return;
+ }
+#endif // CONFIG_XZONE_MALLOC
//
// PR_7288598: Must pass a *scalable* zone (szone) as the helper for create_purgeable_zone().
@@ -4044,6 +4344,11 @@
register_msl_dylib(void *dylib)
{
#if TARGET_OS_DRIVERKIT
+ // rdar://153673319
+ if (!msl.copy_msl_lite_hooks) {
+ return;
+ }
+
set_msl_lite_hooks(msl.copy_msl_lite_hooks);
#else
if (!dylib) {