/*
- * Copyright (c) 2000-2014 Apple Inc. All rights reserved.
+ * Copyright (c) 2000-2020 Apple Inc. All rights reserved.
*
* @APPLE_OSREFERENCE_LICENSE_HEADER_START@
- *
+ *
* This file contains Original Code and/or Modifications of Original Code
* as defined in and that are subject to the Apple Public Source License
* Version 2.0 (the 'License'). You may not use this file except in
* unlawful or unlicensed copies of an Apple operating system, or to
* circumvent, violate, or enable the circumvention or violation of, any
* terms of an Apple operating system software license agreement.
- *
+ *
* Please obtain a copy of the License at
* http://www.opensource.apple.com/apsl/ and read it before using this file.
- *
+ *
* The Original Code and all software distributed under the License are
* distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
* EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
* FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
* Please see the License for the specific language governing rights and
* limitations under the License.
- *
+ *
* @APPLE_OSREFERENCE_LICENSE_HEADER_END@
*/
/*
* @OSF_COPYRIGHT@
*/
-/*
+/*
* Mach Operating System
* Copyright (c) 1991,1990,1989,1988,1987 Carnegie Mellon University
* All Rights Reserved.
- *
+ *
* Permission to use, copy, modify and distribute this software and its
* documentation is hereby granted, provided that both the copyright
* notice and this permission notice appear in all copies of the
* software, derivative works or modified versions, and any portions
* thereof, and that both notices appear in supporting documentation.
- *
+ *
* CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
* CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND FOR
* ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
- *
+ *
* Carnegie Mellon requests users of this software to return to
- *
+ *
* Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
* School of Computer Science
* Carnegie Mellon University
* Pittsburgh PA 15213-3890
- *
+ *
* any improvements or extensions that they make and grant Carnegie Mellon
* the rights to redistribute these changes.
*/
* Zone-based memory allocator. A zone is a collection of fixed size
* data blocks for which quick allocation/deallocation is possible.
*/
-#include <zone_debug.h>
-#include <zone_alias_addr.h>
+#define ZALLOC_ALLOW_DEPRECATED 1
#include <mach/mach_types.h>
#include <mach/vm_param.h>
#include <mach/kern_return.h>
#include <mach/mach_host_server.h>
#include <mach/task_server.h>
#include <mach/machine/vm_types.h>
-#include <mach_debug/zone_info.h>
#include <mach/vm_map.h>
+#include <mach/sdt.h>
+#include <kern/bits.h>
+#include <kern/startup.h>
#include <kern/kern_types.h>
#include <kern/assert.h>
+#include <kern/backtrace.h>
#include <kern/host.h>
#include <kern/macro_help.h>
#include <kern/sched.h>
#include <kern/sched_prim.h>
#include <kern/misc_protos.h>
#include <kern/thread_call.h>
-#include <kern/zalloc.h>
+#include <kern/zalloc_internal.h>
#include <kern/kalloc.h>
-#include <kern/btlog.h>
+
+#include <prng/random.h>
#include <vm/pmap.h>
#include <vm/vm_map.h>
#include <vm/vm_kern.h>
#include <vm/vm_page.h>
+#include <vm/vm_compressor.h> /* C_SLOT_PACKED_PTR* */
#include <pexpert/pexpert.h>
#include <machine/machparam.h>
#include <machine/machine_routines.h> /* ml_cpu_get_info */
+#include <os/atomic.h>
+
#include <libkern/OSDebug.h>
#include <libkern/OSAtomic.h>
+#include <libkern/section_keywords.h>
#include <sys/kdebug.h>
+#include <san/kasan.h>
+
+#if KASAN_ZALLOC
+#define ZONE_ENABLE_LOGGING 0
+#elif DEBUG || DEVELOPMENT
+#define ZONE_ENABLE_LOGGING 1
+#else
+#define ZONE_ENABLE_LOGGING 0
+#endif
+
+extern void vm_pageout_garbage_collect(int collect);
+
+/* Returns pid of the task with the largest number of VM map entries. */
+extern pid_t find_largest_process_vm_map_entries(void);
+
/*
- * ZONE_ALIAS_ADDR
- *
- * With this option enabled, zones with alloc_size <= PAGE_SIZE allocate
- * a virtual page from the zone_map, but before zcram-ing the allocated memory
- * into the zone, the page is translated to use the alias address of the page
- * in the static kernel region. zone_gc reverses that translation when
- * scanning the freelist to collect free pages so that it can look up the page
- * in the zone_page_table, and free it to kmem_free.
- *
- * The static kernel region is a flat 1:1 mapping of physical memory passed
- * to xnu by the booter. It is mapped to the range:
- * [gVirtBase, gVirtBase + gPhysSize]
- *
- * Accessing memory via the static kernel region is faster due to the
- * entire region being mapped via large pages, cutting down
- * on TLB misses.
- *
- * zinit favors using PAGE_SIZE backing allocations for a zone unless it would
- * waste more than 10% space to use a single page, in order to take advantage
- * of the speed benefit for as many zones as possible.
- *
- * Zones with > PAGE_SIZE allocations can't take advantage of this
- * because kernel_memory_allocate doesn't give out physically contiguous pages.
- *
- * zone_virtual_addr()
- * - translates an address from the static kernel region to the zone_map
- * - returns the same address if it's not from the static kernel region
- * It relies on the fact that a physical page mapped to the
- * zone_map is not mapped anywhere else (except the static kernel region).
- *
- * zone_alias_addr()
- * - translates a virtual memory address from the zone_map to the
- * corresponding address in the static kernel region
+ * Callout to jetsam. If pid is -1, we wake up the memorystatus thread to do asynchronous kills.
+ * For any other pid we try to kill that process synchronously.
+ */
+extern boolean_t memorystatus_kill_on_zone_map_exhaustion(pid_t pid);
+
+extern zone_t vm_map_entry_zone;
+extern zone_t vm_object_zone;
+extern vm_offset_t kmapoff_kaddr;
+extern unsigned int kmapoff_pgcnt;
+extern unsigned int stack_total;
+extern unsigned long long stack_allocs;
+
+/*
+ * The max # of elements in a chunk should fit into
+ * zone_page_metadata.free_count (uint16_t).
*
+ * Update this if the type of free_count changes.
*/
+#define ZONE_CHUNK_MAXELEMENTS (UINT16_MAX)
-#if !ZONE_ALIAS_ADDR
-#define from_zone_map(addr, size) \
- ((vm_offset_t)(addr) >= zone_map_min_address && \
- ((vm_offset_t)(addr) + size - 1) < zone_map_max_address )
+#define ZONE_PAGECOUNT_BITS 14
+
+/* Zone elements must fit both a next pointer and a backup pointer */
+#define ZONE_MIN_ELEM_SIZE (2 * sizeof(vm_offset_t))
+#define ZONE_MAX_ALLOC_SIZE (32 * 1024)
+
+/* per-cpu zones are special because of counters */
+#define ZONE_MIN_PCPU_ELEM_SIZE (1 * sizeof(vm_offset_t))
+
+struct zone_map_range {
+ vm_offset_t min_address;
+ vm_offset_t max_address;
+};
+
+struct zone_page_metadata {
+ /* The index of the zone this metadata page belongs to */
+ zone_id_t zm_index;
+
+ /*
+ * zm_secondary_page == 0: number of pages in this run
+ * zm_secondary_page == 1: offset to the chunk start
+ */
+ uint16_t zm_page_count : ZONE_PAGECOUNT_BITS;
+
+ /* Whether this page is part of a chunk run */
+ uint16_t zm_percpu : 1;
+ uint16_t zm_secondary_page : 1;
+
+ /*
+ * The start of the freelist can be maintained as a 16-bit
+ * offset instead of a pointer because the free elements would
+ * be at max ZONE_MAX_ALLOC_SIZE bytes away from the start
+ * of the allocation chunk.
+ *
+ * Offset from start of the allocation chunk to free element
+ * list head.
+ */
+ uint16_t zm_freelist_offs;
+
+ /*
+ * zm_secondary_page == 0: number of allocated elements in the chunk
+ * zm_secondary_page == 1: unused
+ *
+ * PAGE_METADATA_EMPTY_FREELIST indicates an empty freelist
+ */
+ uint16_t zm_alloc_count;
+#define PAGE_METADATA_EMPTY_FREELIST UINT16_MAX
+
+ zone_pva_t zm_page_next;
+ zone_pva_t zm_page_prev;
+
+ /*
+ * This is only for the sake of debuggers
+ */
+#define ZONE_FOREIGN_COOKIE 0x123456789abcdef
+ uint64_t zm_foreign_cookie[];
+};
+
+
+/* Align elements that use the zone page list to 32 byte boundaries. */
+#define ZONE_PAGE_FIRST_OFFSET(kind) ((kind) == ZONE_ADDR_NATIVE ? 0 : 32)
+
+static_assert(sizeof(struct zone_page_metadata) == 16, "validate packing");
+
+static __security_const_late struct {
+ struct zone_map_range zi_map_range;
+ struct zone_map_range zi_general_range;
+ struct zone_map_range zi_meta_range;
+ struct zone_map_range zi_foreign_range;
+
+ /*
+ * The metadata lives within the zi_meta_range address range.
+ *
+ * The correct formula to find a metadata index is:
+ * absolute_page_index - page_index(zi_meta_range.min_address)
+ *
+ * And then this index is used to dereference zi_meta_range.min_address
+ * as a `struct zone_page_metadata` array.
+ *
+ * To avoid doing that substraction all the time in the various fast-paths,
+ * zi_array_base is offset by `page_index(zi_meta_range.min_address)`
+ * to avoid redoing that math all the time.
+ */
+ struct zone_page_metadata *zi_array_base;
+} zone_info;
+
+/*
+ * The zone_locks_grp allows for collecting lock statistics.
+ * All locks are associated to this group in zinit.
+ * Look at tools/lockstat for debugging lock contention.
+ */
+LCK_GRP_DECLARE(zone_locks_grp, "zone_locks");
+LCK_MTX_EARLY_DECLARE(zone_metadata_region_lck, &zone_locks_grp);
+
+/*
+ * Exclude more than one concurrent garbage collection
+ */
+LCK_GRP_DECLARE(zone_gc_lck_grp, "zone_gc");
+LCK_MTX_EARLY_DECLARE(zone_gc_lock, &zone_gc_lck_grp);
+
+boolean_t panic_include_zprint = FALSE;
+mach_memory_info_t *panic_kext_memory_info = NULL;
+vm_size_t panic_kext_memory_size = 0;
+
+/*
+ * Protects zone_array, num_zones, num_zones_in_use, and
+ * zone_destroyed_bitmap
+ */
+static SIMPLE_LOCK_DECLARE(all_zones_lock, 0);
+static unsigned int num_zones_in_use;
+unsigned int _Atomic num_zones;
+SECURITY_READ_ONLY_LATE(unsigned int) zone_view_count;
+
+#if KASAN_ZALLOC
+#define MAX_ZONES 566
+#else /* !KASAN_ZALLOC */
+#define MAX_ZONES 402
+#endif/* !KASAN_ZALLOC */
+struct zone zone_array[MAX_ZONES];
+
+/* Initialized in zone_bootstrap(), how many "copies" the per-cpu system does */
+static SECURITY_READ_ONLY_LATE(unsigned) zpercpu_early_count;
+
+/* Used to keep track of destroyed slots in the zone_array */
+static bitmap_t zone_destroyed_bitmap[BITMAP_LEN(MAX_ZONES)];
+
+/* number of pages used by all zones */
+static long _Atomic zones_phys_page_count;
+
+/* number of zone mapped pages used by all zones */
+static long _Atomic zones_phys_page_mapped_count;
+
+/*
+ * Turn ZSECURITY_OPTIONS_STRICT_IOKIT_FREE off on x86 so as not
+ * not break third party kexts that haven't yet been recompiled
+ * to use the new iokit macros.
+ */
+#if XNU_TARGET_OS_OSX && __x86_64__
+#define ZSECURITY_OPTIONS_STRICT_IOKIT_FREE_DEFAULT 0
#else
-#define from_zone_map(addr, size) \
- ((vm_offset_t)(zone_virtual_addr((vm_map_address_t)(uintptr_t)addr)) >= zone_map_min_address && \
- ((vm_offset_t)(zone_virtual_addr((vm_map_address_t)(uintptr_t)addr)) + size -1) < zone_map_max_address )
+#define ZSECURITY_OPTIONS_STRICT_IOKIT_FREE_DEFAULT \
+ ZSECURITY_OPTIONS_STRICT_IOKIT_FREE
+#endif
+
+#define ZSECURITY_DEFAULT ( \
+ ZSECURITY_OPTIONS_SEQUESTER | \
+ ZSECURITY_OPTIONS_SUBMAP_USER_DATA | \
+ ZSECURITY_OPTIONS_SEQUESTER_KEXT_KALLOC | \
+ ZSECURITY_OPTIONS_STRICT_IOKIT_FREE_DEFAULT | \
+ 0)
+TUNABLE(zone_security_options_t, zsecurity_options, "zs", ZSECURITY_DEFAULT);
+
+#if VM_MAX_TAG_ZONES
+/* enable tags for zones that ask for it */
+TUNABLE(bool, zone_tagging_on, "-zt", false);
+#endif /* VM_MAX_TAG_ZONES */
+
+#if DEBUG || DEVELOPMENT
+TUNABLE(bool, zalloc_disable_copyio_check, "-no-copyio-zalloc-check", false);
+__options_decl(zalloc_debug_t, uint32_t, {
+ ZALLOC_DEBUG_ZONEGC = 0x00000001,
+ ZALLOC_DEBUG_ZCRAM = 0x00000002,
+});
+
+TUNABLE(zalloc_debug_t, zalloc_debug, "zalloc_debug", 0);
+#endif /* DEBUG || DEVELOPMENT */
+#if CONFIG_ZLEAKS
+/* Making pointer scanning leaks detection possible for all zones */
+TUNABLE(bool, zone_leaks_scan_enable, "-zl", false);
+#else
+#define zone_leaks_scan_enable false
#endif
+/*
+ * Async allocation of zones
+ * This mechanism allows for bootstrapping an empty zone which is setup with
+ * non-blocking flags. The first call to zalloc_noblock() will kick off a thread_call
+ * to zalloc_async. We perform a zalloc() (which may block) and then an immediate free.
+ * This will prime the zone for the next use.
+ *
+ * Currently the thread_callout function (zalloc_async) will loop through all zones
+ * looking for any zone with async_pending set and do the work for it.
+ *
+ * NOTE: If the calling thread for zalloc_noblock is lower priority than thread_call,
+ * then zalloc_noblock to an empty zone may succeed.
+ */
+static void zalloc_async(thread_call_param_t p0, thread_call_param_t p1);
+static thread_call_data_t call_async_alloc;
+static void zcram_and_lock(zone_t zone, vm_offset_t newmem, vm_size_t size);
+
/*
* Zone Corruption Debugging
*
- * We use three techniques to detect modification of a zone element
+ * We use four techniques to detect modification of a zone element
* after it's been freed.
*
* (1) Check the freelist next pointer for sanity.
* no part of the element has been modified while it was on the freelist.
* This will also help catch read-after-frees, as code will now dereference
* 0xdeadbeef instead of a valid but freed pointer.
+ * (4) If the zfree_clear_mem flag is set clear the element on free and
+ * assert that it is still clear when alloc-ed.
*
* (1) and (2) occur for every allocation and free to a zone.
* This is done to make it slightly more difficult for an attacker to
* manipulate the freelist to behave in a specific way.
*
- * Poisoning (3) occurs periodically for every N frees (counted per-zone)
- * and on every free for zones smaller than a cacheline. If -zp
- * is passed as a boot arg, poisoning occurs for every free.
+ * Poisoning (3) occurs periodically for every N frees (counted per-zone).
+ * If -zp is passed as a boot arg, poisoning occurs for every free.
+ *
+ * Zeroing (4) is done for those zones that pass the ZC_ZFREE_CLEARMEM
+ * flag on creation or if the element size is less than one cacheline.
*
* Performance slowdown is inversely proportional to the frequency of poisoning,
* with a 4-5% hit around N=1, down to ~0.3% at N=16 and just "noise" at N=32
*
*/
-/* Returns TRUE if we rolled over the counter at factor */
-static inline boolean_t
-sample_counter(volatile uint32_t * count_p, uint32_t factor)
-{
- uint32_t old_count, new_count;
- boolean_t rolled_over;
+#define ZP_DEFAULT_SAMPLING_FACTOR 16
+#define ZP_DEFAULT_SCALE_FACTOR 4
- do {
- new_count = old_count = *count_p;
+/*
+ * set by zp-factor=N boot arg
+ *
+ * A zp_factor of 0 indicates zone poisoning is disabled and can also be set by
+ * passing the -no-zp boot-arg.
+ *
+ * A zp_factor of 1 indicates zone poisoning is on for all elements and can be
+ * set by passing the -zp boot-arg.
+ */
+static TUNABLE(uint32_t, zp_factor, "zp-factor", ZP_DEFAULT_SAMPLING_FACTOR);
- if (++new_count >= factor) {
- rolled_over = TRUE;
- new_count = 0;
- } else {
- rolled_over = FALSE;
- }
+/* set by zp-scale=N boot arg, scales zp_factor by zone size */
+static TUNABLE(uint32_t, zp_scale, "zp-scale", ZP_DEFAULT_SCALE_FACTOR);
- } while (!OSCompareAndSwap(old_count, new_count, count_p));
+/* initialized to a per-boot random value in zp_bootstrap */
+static SECURITY_READ_ONLY_LATE(uintptr_t) zp_poisoned_cookie;
+static SECURITY_READ_ONLY_LATE(uintptr_t) zp_nopoison_cookie;
+static SECURITY_READ_ONLY_LATE(uintptr_t) zp_min_size;
+static SECURITY_READ_ONLY_LATE(uint64_t) zone_phys_mapped_max;
- return rolled_over;
-}
+static SECURITY_READ_ONLY_LATE(vm_map_t) zone_submaps[Z_SUBMAP_IDX_COUNT];
+static SECURITY_READ_ONLY_LATE(uint32_t) zone_last_submap_idx;
-#if defined(__LP64__)
-#define ZP_POISON 0xdeadbeefdeadbeef
-#else
-#define ZP_POISON 0xdeadbeef
-#endif
+static struct bool_gen zone_bool_gen;
+static zone_t zone_find_largest(void);
+static void zone_drop_free_elements(zone_t z);
-#define ZP_DEFAULT_SAMPLING_FACTOR 16
-#define ZP_DEFAULT_SCALE_FACTOR 4
+#define submap_for_zone(z) zone_submaps[(z)->submap_idx]
+#define MAX_SUBMAP_NAME 16
+
+/* Globals for random boolean generator for elements in free list */
+#define MAX_ENTROPY_PER_ZCRAM 4
+#if CONFIG_ZCACHE
/*
- * A zp_factor of 0 indicates zone poisoning is disabled,
- * however, we still poison zones smaller than zp_tiny_zone_limit (a cacheline).
- * Passing the -no-zp boot-arg disables even this behavior.
- * In all cases, we record and check the integrity of a backup pointer.
+ * Specifies a single zone to enable CPU caching for.
+ * Can be set using boot-args: zcc_enable_for_zone_name=<zone>
*/
+static char cache_zone_name[MAX_ZONE_NAME];
+static TUNABLE(bool, zcc_kalloc, "zcc_kalloc", false);
+
+__header_always_inline bool
+zone_caching_enabled(zone_t z)
+{
+ return z->zcache.zcc_depot != NULL;
+}
+#else
+__header_always_inline bool
+zone_caching_enabled(zone_t z __unused)
+{
+ return false;
+}
+#endif /* CONFIG_ZCACHE */
-/* set by zp-factor=N boot arg, zero indicates non-tiny poisoning disabled */
-uint32_t zp_factor = 0;
+#pragma mark Zone metadata
-/* set by zp-scale=N boot arg, scales zp_factor by zone size */
-uint32_t zp_scale = 0;
+__enum_closed_decl(zone_addr_kind_t, bool, {
+ ZONE_ADDR_NATIVE,
+ ZONE_ADDR_FOREIGN,
+});
-/* set in zp_init, zero indicates -no-zp boot-arg */
-vm_size_t zp_tiny_zone_limit = 0;
+static inline zone_id_t
+zone_index(zone_t z)
+{
+ return (zone_id_t)(z - zone_array);
+}
-/* initialized to a per-boot random value in zp_init */
-uintptr_t zp_poisoned_cookie = 0;
-uintptr_t zp_nopoison_cookie = 0;
+static inline bool
+zone_has_index(zone_t z, zone_id_t zid)
+{
+ return zone_array + zid == z;
+}
+static inline vm_size_t
+zone_elem_count(zone_t zone, vm_size_t alloc_size, zone_addr_kind_t kind)
+{
+ if (kind == ZONE_ADDR_NATIVE) {
+ if (zone->percpu) {
+ return PAGE_SIZE / zone_elem_size(zone);
+ }
+ return alloc_size / zone_elem_size(zone);
+ } else {
+ assert(alloc_size == PAGE_SIZE);
+ return (PAGE_SIZE - ZONE_PAGE_FIRST_OFFSET(kind)) / zone_elem_size(zone);
+ }
+}
-/*
- * initialize zone poisoning
- * called from zone_bootstrap before any allocations are made from zalloc
- */
-static inline void
-zp_init(void)
+__abortlike
+static void
+zone_metadata_corruption(zone_t zone, struct zone_page_metadata *meta,
+ const char *kind)
{
- char temp_buf[16];
+ panic("zone metadata corruption: %s (meta %p, zone %s%s)",
+ kind, meta, zone_heap_name(zone), zone->z_name);
+}
- /*
- * Initialize backup pointer random cookie for poisoned elements
- * Try not to call early_random() back to back, it may return
- * the same value if mach_absolute_time doesn't have sufficient time
- * to tick over between calls. <rdar://problem/11597395>
- * (This is only a problem on embedded devices)
- */
- zp_poisoned_cookie = (uintptr_t) early_random();
+__abortlike
+static void
+zone_invalid_element_addr_panic(zone_t zone, vm_offset_t addr)
+{
+ panic("zone element pointer validation failed (addr: %p, zone %s%s)",
+ (void *)addr, zone_heap_name(zone), zone->z_name);
+}
- /*
- * Always poison zones smaller than a cacheline,
- * because it's pretty close to free
- */
- ml_cpu_info_t cpu_info;
- ml_cpu_get_info(&cpu_info);
- zp_tiny_zone_limit = (vm_size_t) cpu_info.cache_line_size;
+__abortlike
+static void
+zone_page_metadata_index_confusion_panic(zone_t zone, vm_offset_t addr,
+ struct zone_page_metadata *meta)
+{
+ panic("%p not in the expected zone %s%s (%d != %d)",
+ (void *)addr, zone_heap_name(zone), zone->z_name,
+ meta->zm_index, zone_index(zone));
+}
- zp_factor = ZP_DEFAULT_SAMPLING_FACTOR;
- zp_scale = ZP_DEFAULT_SCALE_FACTOR;
+__abortlike
+static void
+zone_page_metadata_native_queue_corruption(zone_t zone, zone_pva_t *queue)
+{
+ panic("foreign metadata index %d enqueued in native head %p from zone %s%s",
+ queue->packed_address, queue, zone_heap_name(zone),
+ zone->z_name);
+}
- //TODO: Bigger permutation?
- /*
- * Permute the default factor +/- 1 to make it less predictable
- * This adds or subtracts ~4 poisoned objects per 1000 frees.
- */
- if (zp_factor != 0) {
- uint32_t rand_bits = early_random() & 0x3;
+__abortlike
+static void
+zone_page_metadata_list_corruption(zone_t zone, struct zone_page_metadata *meta)
+{
+ panic("metadata list corruption through element %p detected in zone %s%s",
+ meta, zone_heap_name(zone), zone->z_name);
+}
- if (rand_bits == 0x1)
- zp_factor += 1;
- else if (rand_bits == 0x2)
- zp_factor -= 1;
- /* if 0x0 or 0x3, leave it alone */
- }
+__abortlike
+static void
+zone_page_metadata_foreign_queue_corruption(zone_t zone, zone_pva_t *queue)
+{
+ panic("native metadata index %d enqueued in foreign head %p from zone %s%s",
+ queue->packed_address, queue, zone_heap_name(zone), zone->z_name);
+}
- /* -zp: enable poisoning for every alloc and free */
- if (PE_parse_boot_argn("-zp", temp_buf, sizeof(temp_buf))) {
- zp_factor = 1;
- }
+__abortlike
+static void
+zone_page_metadata_foreign_confusion_panic(zone_t zone, vm_offset_t addr)
+{
+ panic("manipulating foreign address %p in a native-only zone %s%s",
+ (void *)addr, zone_heap_name(zone), zone->z_name);
+}
- /* -no-zp: disable poisoning completely even for tiny zones */
- if (PE_parse_boot_argn("-no-zp", temp_buf, sizeof(temp_buf))) {
- zp_factor = 0;
- zp_tiny_zone_limit = 0;
- printf("Zone poisoning disabled\n");
- }
+__abortlike __unused
+static void
+zone_invalid_foreign_addr_panic(zone_t zone, vm_offset_t addr)
+{
+ panic("addr %p being freed to foreign zone %s%s not from foreign range",
+ (void *)addr, zone_heap_name(zone), zone->z_name);
+}
- /* zp-factor=XXXX: override how often to poison freed zone elements */
- if (PE_parse_boot_argn("zp-factor", &zp_factor, sizeof(zp_factor))) {
- printf("Zone poisoning factor override: %u\n", zp_factor);
- }
+__abortlike
+static void
+zone_page_meta_accounting_panic(zone_t zone, struct zone_page_metadata *meta,
+ const char *kind)
+{
+ panic("accounting mismatch (%s) for zone %s%s, meta %p", kind,
+ zone_heap_name(zone), zone->z_name, meta);
+}
- /* zp-scale=XXXX: override how much zone size scales zp-factor by */
- if (PE_parse_boot_argn("zp-scale", &zp_scale, sizeof(zp_scale))) {
- printf("Zone poisoning scale factor override: %u\n", zp_scale);
- }
+__abortlike
+static void
+zone_accounting_panic(zone_t zone, const char *kind)
+{
+ panic("accounting mismatch (%s) for zone %s%s", kind,
+ zone_heap_name(zone), zone->z_name);
+}
- /* Initialize backup pointer random cookie for unpoisoned elements */
- zp_nopoison_cookie = (uintptr_t) early_random();
+__abortlike
+static void
+zone_nofail_panic(zone_t zone)
+{
+ panic("zalloc(Z_NOFAIL) can't be satisfied for zone %s%s (potential leak)",
+ zone_heap_name(zone), zone->z_name);
+}
-#if MACH_ASSERT
- if (zp_poisoned_cookie == zp_nopoison_cookie)
- panic("early_random() is broken: %p and %p are not random\n",
- (void *) zp_poisoned_cookie, (void *) zp_nopoison_cookie);
+#if __arm64__
+// <rdar://problem/48304934> arm64 doesn't use ldp when I'd expect it to
+#define zone_range_load(r, rmin, rmax) \
+ asm("ldp %[rmin], %[rmax], [%[range]]" \
+ : [rmin] "=r"(rmin), [rmax] "=r"(rmax) \
+ : [range] "r"(r))
+#else
+#define zone_range_load(r, rmin, rmax) \
+ ({ rmin = (r)->min_address; rmax = (r)->max_address; })
#endif
- /*
- * Use the last bit in the backup pointer to hint poisoning state
- * to backup_ptr_mismatch_panic. Valid zone pointers are aligned, so
- * the low bits are zero.
- */
- zp_poisoned_cookie |= (uintptr_t)0x1ULL;
- zp_nopoison_cookie &= ~((uintptr_t)0x1ULL);
+__header_always_inline bool
+zone_range_contains(const struct zone_map_range *r, vm_offset_t addr, vm_offset_t size)
+{
+ vm_offset_t rmin, rmax;
-#if defined(__LP64__)
/*
- * Make backup pointers more obvious in GDB for 64 bit
- * by making OxFFFFFF... ^ cookie = 0xFACADE...
- * (0xFACADE = 0xFFFFFF ^ 0x053521)
- * (0xC0FFEE = 0xFFFFFF ^ 0x3f0011)
- * The high 3 bytes of a zone pointer are always 0xFFFFFF, and are checked
- * by the sanity check, so it's OK for that part of the cookie to be predictable.
- *
- * TODO: Use #defines, xors, and shifts
+ * The `&` is not a typo: we really expect the check to pass,
+ * so encourage the compiler to eagerly load and test without branches
*/
+ zone_range_load(r, rmin, rmax);
+ return (addr >= rmin) & (addr + size >= rmin) & (addr + size <= rmax);
+}
- zp_poisoned_cookie &= 0x000000FFFFFFFFFF;
- zp_poisoned_cookie |= 0x0535210000000000; /* 0xFACADE */
+__header_always_inline vm_size_t
+zone_range_size(const struct zone_map_range *r)
+{
+ vm_offset_t rmin, rmax;
- zp_nopoison_cookie &= 0x000000FFFFFFFFFF;
- zp_nopoison_cookie |= 0x3f00110000000000; /* 0xC0FFEE */
-#endif
+ zone_range_load(r, rmin, rmax);
+ return rmax - rmin;
}
-/* zone_map page count for page table structure */
-uint64_t zone_map_table_page_count = 0;
+#define from_zone_map(addr, size) \
+ zone_range_contains(&zone_info.zi_map_range, (vm_offset_t)(addr), size)
-/*
- * These macros are used to keep track of the number
- * of pages being used by the zone currently. The
- * z->page_count is protected by the zone lock.
- */
-#define ZONE_PAGE_COUNT_INCR(z, count) \
-{ \
- OSAddAtomic64(count, &(z->page_count)); \
-}
+#define from_general_submap(addr, size) \
+ zone_range_contains(&zone_info.zi_general_range, (vm_offset_t)(addr), size)
-#define ZONE_PAGE_COUNT_DECR(z, count) \
-{ \
- OSAddAtomic64(-count, &(z->page_count)); \
-}
+#define from_foreign_range(addr, size) \
+ zone_range_contains(&zone_info.zi_foreign_range, (vm_offset_t)(addr), size)
-/* for is_sane_zone_element and garbage collection */
+#define from_native_meta_map(addr) \
+ zone_range_contains(&zone_info.zi_meta_range, (vm_offset_t)(addr), \
+ sizeof(struct zone_page_metadata))
-vm_offset_t zone_map_min_address = 0; /* initialized in zone_init */
-vm_offset_t zone_map_max_address = 0;
+#define zone_addr_kind(addr, size) \
+ (from_zone_map(addr, size) ? ZONE_ADDR_NATIVE : ZONE_ADDR_FOREIGN)
-/* Helpful for walking through a zone's free element list. */
-struct zone_free_element {
- struct zone_free_element *next;
- /* ... */
- /* void *backup_ptr; */
-};
+__header_always_inline bool
+zone_pva_is_null(zone_pva_t page)
+{
+ return page.packed_address == 0;
+}
-struct zone_page_metadata {
- queue_chain_t pages;
- struct zone_free_element *elements;
- zone_t zone;
- uint16_t alloc_count;
- uint16_t free_count;
-};
+__header_always_inline bool
+zone_pva_is_queue(zone_pva_t page)
+{
+ // actual kernel pages have the top bit set
+ return (int32_t)page.packed_address > 0;
+}
-/* The backup pointer is stored in the last pointer-sized location in an element. */
-static inline vm_offset_t *
-get_backup_ptr(vm_size_t elem_size,
- vm_offset_t *element)
+__header_always_inline bool
+zone_pva_is_equal(zone_pva_t pva1, zone_pva_t pva2)
{
- return (vm_offset_t *) ((vm_offset_t)element + elem_size - sizeof(vm_offset_t));
+ return pva1.packed_address == pva2.packed_address;
}
-static inline struct zone_page_metadata *
-get_zone_page_metadata(struct zone_free_element *element)
+__header_always_inline void
+zone_queue_set_head(zone_t z, zone_pva_t queue, zone_pva_t oldv,
+ struct zone_page_metadata *meta)
{
- return (struct zone_page_metadata *)(trunc_page((vm_offset_t)element));
+ zone_pva_t *queue_head = &((zone_pva_t *)zone_array)[queue.packed_address];
+
+ if (!zone_pva_is_equal(*queue_head, oldv)) {
+ zone_page_metadata_list_corruption(z, meta);
+ }
+ *queue_head = meta->zm_page_next;
}
-/*
- * Zone checking helper function.
- * A pointer that satisfies these conditions is OK to be a freelist next pointer
- * A pointer that doesn't satisfy these conditions indicates corruption
- */
-static inline boolean_t
-is_sane_zone_ptr(zone_t zone,
- vm_offset_t addr,
- size_t obj_size)
+__header_always_inline zone_pva_t
+zone_queue_encode(zone_pva_t *headp)
{
- /* Must be aligned to pointer boundary */
- if (__improbable((addr & (sizeof(vm_offset_t) - 1)) != 0))
- return FALSE;
+ return (zone_pva_t){ (uint32_t)(headp - (zone_pva_t *)zone_array) };
+}
- /* Must be a kernel address */
- if (__improbable(!pmap_kernel_va(addr)))
- return FALSE;
+__header_always_inline zone_pva_t
+zone_pva_from_addr(vm_address_t addr)
+{
+ // cannot use atop() because we want to maintain the sign bit
+ return (zone_pva_t){ (uint32_t)((intptr_t)addr >> PAGE_SHIFT) };
+}
- /* Must be from zone map if the zone only uses memory from the zone_map */
- /*
- * TODO: Remove the zone->collectable check when every
- * zone using foreign memory is properly tagged with allows_foreign
- */
- if (zone->collectable && !zone->allows_foreign) {
-#if ZONE_ALIAS_ADDR
- /*
- * If this address is in the static kernel region, it might be
- * the alias address of a valid zone element.
- * If we tried to find the zone_virtual_addr() of an invalid
- * address in the static kernel region, it will panic, so don't
- * check addresses in this region.
- *
- * TODO: Use a safe variant of zone_virtual_addr to
- * make this check more accurate
- *
- * The static kernel region is mapped at:
- * [gVirtBase, gVirtBase + gPhysSize]
- */
- if ((addr - gVirtBase) < gPhysSize)
- return TRUE;
-#endif
- /* check if addr is from zone map */
- if (addr >= zone_map_min_address &&
- (addr + obj_size - 1) < zone_map_max_address )
- return TRUE;
-
- return FALSE;
- }
-
- return TRUE;
+__header_always_inline vm_address_t
+zone_pva_to_addr(zone_pva_t page)
+{
+ // cause sign extension so that we end up with the right address
+ return (vm_offset_t)(int32_t)page.packed_address << PAGE_SHIFT;
}
-static inline boolean_t
-is_sane_zone_page_metadata(zone_t zone,
- vm_offset_t page_meta)
+__header_always_inline struct zone_page_metadata *
+zone_pva_to_meta(zone_pva_t page, zone_addr_kind_t kind)
{
- /* NULL page metadata structures are invalid */
- if (page_meta == 0)
- return FALSE;
- return is_sane_zone_ptr(zone, page_meta, sizeof(struct zone_page_metadata));
+ if (kind == ZONE_ADDR_NATIVE) {
+ return &zone_info.zi_array_base[page.packed_address];
+ } else {
+ return (struct zone_page_metadata *)zone_pva_to_addr(page);
+ }
}
-static inline boolean_t
-is_sane_zone_element(zone_t zone,
- vm_offset_t addr)
+__header_always_inline zone_pva_t
+zone_pva_from_meta(struct zone_page_metadata *meta, zone_addr_kind_t kind)
{
- /* NULL is OK because it indicates the tail of the list */
- if (addr == 0)
- return TRUE;
- return is_sane_zone_ptr(zone, addr, zone->elem_size);
-}
-
-/* Someone wrote to freed memory. */
-static inline void /* noreturn */
-zone_element_was_modified_panic(zone_t zone,
- vm_offset_t element,
- vm_offset_t found,
- vm_offset_t expected,
- vm_offset_t offset)
-{
- panic("a freed zone element has been modified in zone %s: expected %p but found %p, bits changed %p, at offset %d of %d in element %p, cookies %p %p",
- zone->zone_name,
- (void *) expected,
- (void *) found,
- (void *) (expected ^ found),
- (uint32_t) offset,
- (uint32_t) zone->elem_size,
- (void *) element,
- (void *) zp_nopoison_cookie,
- (void *) zp_poisoned_cookie);
+ if (kind == ZONE_ADDR_NATIVE) {
+ uint32_t index = (uint32_t)(meta - zone_info.zi_array_base);
+ return (zone_pva_t){ index };
+ } else {
+ return zone_pva_from_addr((vm_address_t)meta);
+ }
}
-/*
- * The primary and backup pointers don't match.
- * Determine which one was likely the corrupted pointer, find out what it
- * probably should have been, and panic.
- * I would like to mark this as noreturn, but panic() isn't marked noreturn.
- */
-static void /* noreturn */
-backup_ptr_mismatch_panic(zone_t zone,
- vm_offset_t element,
- vm_offset_t primary,
- vm_offset_t backup)
+__header_always_inline struct zone_page_metadata *
+zone_meta_from_addr(vm_offset_t addr, zone_addr_kind_t kind)
{
- vm_offset_t likely_backup;
-
- boolean_t sane_backup;
- boolean_t sane_primary = is_sane_zone_element(zone, primary);
- boolean_t element_was_poisoned = (backup & 0x1) ? TRUE : FALSE;
-
-#if defined(__LP64__)
- /* We can inspect the tag in the upper bits for additional confirmation */
- if ((backup & 0xFFFFFF0000000000) == 0xFACADE0000000000)
- element_was_poisoned = TRUE;
- else if ((backup & 0xFFFFFF0000000000) == 0xC0FFEE0000000000)
- element_was_poisoned = FALSE;
-#endif
-
- if (element_was_poisoned) {
- likely_backup = backup ^ zp_poisoned_cookie;
- sane_backup = is_sane_zone_element(zone, likely_backup);
+ if (kind == ZONE_ADDR_NATIVE) {
+ return zone_pva_to_meta(zone_pva_from_addr(addr), kind);
} else {
- likely_backup = backup ^ zp_nopoison_cookie;
- sane_backup = is_sane_zone_element(zone, likely_backup);
+ return (struct zone_page_metadata *)trunc_page(addr);
}
+}
- /* The primary is definitely the corrupted one */
- if (!sane_primary && sane_backup)
- zone_element_was_modified_panic(zone, element, primary, likely_backup, 0);
-
- /* The backup is definitely the corrupted one */
- if (sane_primary && !sane_backup)
- zone_element_was_modified_panic(zone, element, backup,
- (primary ^ (element_was_poisoned ? zp_poisoned_cookie : zp_nopoison_cookie)),
- zone->elem_size - sizeof(vm_offset_t));
-
- /*
- * Not sure which is the corrupted one.
- * It's less likely that the backup pointer was overwritten with
- * ( (sane address) ^ (valid cookie) ), so we'll guess that the
- * primary pointer has been overwritten with a sane but incorrect address.
- */
- if (sane_primary && sane_backup)
- zone_element_was_modified_panic(zone, element, primary, likely_backup, 0);
+#define zone_native_meta_from_addr(addr) \
+ zone_meta_from_addr((vm_offset_t)(addr), ZONE_ADDR_NATIVE)
- /* Neither are sane, so just guess. */
- zone_element_was_modified_panic(zone, element, primary, likely_backup, 0);
+__header_always_inline vm_offset_t
+zone_meta_to_addr(struct zone_page_metadata *meta, zone_addr_kind_t kind)
+{
+ if (kind == ZONE_ADDR_NATIVE) {
+ return ptoa((int)(meta - zone_info.zi_array_base));
+ } else {
+ return (vm_offset_t)meta;
+ }
}
-/*
- * Sets the next element of tail to elem.
- * elem can be NULL.
- * Preserves the poisoning state of the element.
- */
-static inline void
-append_zone_element(zone_t zone,
- struct zone_free_element *tail,
- struct zone_free_element *elem)
+__header_always_inline void
+zone_meta_queue_push(zone_t z, zone_pva_t *headp,
+ struct zone_page_metadata *meta, zone_addr_kind_t kind)
{
- vm_offset_t *backup = get_backup_ptr(zone->elem_size, (vm_offset_t *) tail);
+ zone_pva_t head = *headp;
+ zone_pva_t queue_pva = zone_queue_encode(headp);
+ struct zone_page_metadata *tmp;
+
+ meta->zm_page_next = head;
+ if (!zone_pva_is_null(head)) {
+ tmp = zone_pva_to_meta(head, kind);
+ if (!zone_pva_is_equal(tmp->zm_page_prev, queue_pva)) {
+ zone_page_metadata_list_corruption(z, meta);
+ }
+ tmp->zm_page_prev = zone_pva_from_meta(meta, kind);
+ }
+ meta->zm_page_prev = queue_pva;
+ *headp = zone_pva_from_meta(meta, kind);
+}
- vm_offset_t old_backup = *backup;
+__header_always_inline struct zone_page_metadata *
+zone_meta_queue_pop(zone_t z, zone_pva_t *headp, zone_addr_kind_t kind,
+ vm_offset_t *page_addrp)
+{
+ zone_pva_t head = *headp;
+ struct zone_page_metadata *meta = zone_pva_to_meta(head, kind);
+ vm_offset_t page_addr = zone_pva_to_addr(head);
+ struct zone_page_metadata *tmp;
- vm_offset_t old_next = (vm_offset_t) tail->next;
- vm_offset_t new_next = (vm_offset_t) elem;
+ if (kind == ZONE_ADDR_NATIVE && !from_native_meta_map(meta)) {
+ zone_page_metadata_native_queue_corruption(z, headp);
+ }
+ if (kind == ZONE_ADDR_FOREIGN && from_zone_map(meta, sizeof(*meta))) {
+ zone_page_metadata_foreign_queue_corruption(z, headp);
+ }
- if (old_next == (old_backup ^ zp_nopoison_cookie))
- *backup = new_next ^ zp_nopoison_cookie;
- else if (old_next == (old_backup ^ zp_poisoned_cookie))
- *backup = new_next ^ zp_poisoned_cookie;
- else
- backup_ptr_mismatch_panic(zone,
- (vm_offset_t) tail,
- old_next,
- old_backup);
+ if (!zone_pva_is_null(meta->zm_page_next)) {
+ tmp = zone_pva_to_meta(meta->zm_page_next, kind);
+ if (!zone_pva_is_equal(tmp->zm_page_prev, head)) {
+ zone_page_metadata_list_corruption(z, meta);
+ }
+ tmp->zm_page_prev = meta->zm_page_prev;
+ }
+ *headp = meta->zm_page_next;
- tail->next = elem;
+ *page_addrp = page_addr;
+ return meta;
}
-
-/*
- * Insert a linked list of elements (delineated by head and tail) at the head of
- * the zone free list. Every element in the list being added has already gone
- * through append_zone_element, so their backup pointers are already
- * set properly.
- * Precondition: There should be no elements after tail
- */
-static inline void
-add_list_to_zone(zone_t zone,
- struct zone_free_element *head,
- struct zone_free_element *tail)
+__header_always_inline void
+zone_meta_requeue(zone_t z, zone_pva_t *headp,
+ struct zone_page_metadata *meta, zone_addr_kind_t kind)
{
- assert(tail->next == NULL);
- assert(!zone->use_page_list);
+ zone_pva_t meta_pva = zone_pva_from_meta(meta, kind);
+ struct zone_page_metadata *tmp;
- append_zone_element(zone, tail, zone->free_elements);
+ if (!zone_pva_is_null(meta->zm_page_next)) {
+ tmp = zone_pva_to_meta(meta->zm_page_next, kind);
+ if (!zone_pva_is_equal(tmp->zm_page_prev, meta_pva)) {
+ zone_page_metadata_list_corruption(z, meta);
+ }
+ tmp->zm_page_prev = meta->zm_page_prev;
+ }
+ if (zone_pva_is_queue(meta->zm_page_prev)) {
+ zone_queue_set_head(z, meta->zm_page_prev, meta_pva, meta);
+ } else {
+ tmp = zone_pva_to_meta(meta->zm_page_prev, kind);
+ if (!zone_pva_is_equal(tmp->zm_page_next, meta_pva)) {
+ zone_page_metadata_list_corruption(z, meta);
+ }
+ tmp->zm_page_next = meta->zm_page_next;
+ }
- zone->free_elements = head;
+ zone_meta_queue_push(z, headp, meta, kind);
}
-
/*
- * Adds the element to the head of the zone's free list
- * Keeps a backup next-pointer at the end of the element
+ * Routine to populate a page backing metadata in the zone_metadata_region.
+ * Must be called without the zone lock held as it might potentially block.
*/
-static inline void
-free_to_zone(zone_t zone,
- vm_offset_t element,
- boolean_t poison)
+static void
+zone_meta_populate(struct zone_page_metadata *from, struct zone_page_metadata *to)
{
- vm_offset_t old_head;
- struct zone_page_metadata *page_meta;
-
- vm_offset_t *primary = (vm_offset_t *) element;
- vm_offset_t *backup = get_backup_ptr(zone->elem_size, primary);
+ vm_offset_t page_addr = trunc_page(from);
- if (zone->use_page_list) {
- page_meta = get_zone_page_metadata((struct zone_free_element *)element);
- assert(page_meta->zone == zone);
- old_head = (vm_offset_t)page_meta->elements;
- } else {
- old_head = (vm_offset_t)zone->free_elements;
- }
-
-#if MACH_ASSERT
- if (__improbable(!is_sane_zone_element(zone, old_head)))
- panic("zfree: invalid head pointer %p for freelist of zone %s\n",
- (void *) old_head, zone->zone_name);
+ for (; page_addr < (vm_offset_t)to; page_addr += PAGE_SIZE) {
+#if !KASAN_ZALLOC
+ /*
+ * This can race with another thread doing a populate on the same metadata
+ * page, where we see an updated pmap but unmapped KASan shadow, causing a
+ * fault in the shadow when we first access the metadata page. Avoid this
+ * by always synchronizing on the zone_metadata_region lock with KASan.
+ */
+ if (pmap_find_phys(kernel_pmap, page_addr)) {
+ continue;
+ }
#endif
- if (__improbable(!is_sane_zone_element(zone, element)))
- panic("zfree: freeing invalid pointer %p to zone %s\n",
- (void *) element, zone->zone_name);
+ for (;;) {
+ kern_return_t ret = KERN_SUCCESS;
- /*
- * Always write a redundant next pointer
- * So that it is more difficult to forge, xor it with a random cookie
- * A poisoned element is indicated by using zp_poisoned_cookie
- * instead of zp_nopoison_cookie
- */
-
- *backup = old_head ^ (poison ? zp_poisoned_cookie : zp_nopoison_cookie);
+ /* All updates to the zone_metadata_region are done under the zone_metadata_region_lck */
+ lck_mtx_lock(&zone_metadata_region_lck);
+ if (0 == pmap_find_phys(kernel_pmap, page_addr)) {
+ ret = kernel_memory_populate(kernel_map, page_addr,
+ PAGE_SIZE, KMA_NOPAGEWAIT | KMA_KOBJECT | KMA_ZERO,
+ VM_KERN_MEMORY_OSFMK);
+ }
+ lck_mtx_unlock(&zone_metadata_region_lck);
- /* Insert this element at the head of the free list */
- *primary = old_head;
- if (zone->use_page_list) {
- page_meta->elements = (struct zone_free_element *)element;
- page_meta->free_count++;
- if (zone->allows_foreign && !from_zone_map(element, zone->elem_size)) {
- if (page_meta->free_count == 1) {
- /* first foreign element freed on page, move from all_used */
- remqueue((queue_entry_t)page_meta);
- enqueue_tail(&zone->pages.any_free_foreign, (queue_entry_t)page_meta);
- } else {
- /* no other list transitions */
+ if (ret == KERN_SUCCESS) {
+ break;
}
- } else if (page_meta->free_count == page_meta->alloc_count) {
- /* whether the page was on the intermediate or all_used, queue, move it to free */
- remqueue((queue_entry_t)page_meta);
- enqueue_tail(&zone->pages.all_free, (queue_entry_t)page_meta);
- } else if (page_meta->free_count == 1) {
- /* first free element on page, move from all_used */
- remqueue((queue_entry_t)page_meta);
- enqueue_tail(&zone->pages.intermediate, (queue_entry_t)page_meta);
+
+ /*
+ * We can't pass KMA_NOPAGEWAIT under a global lock as it leads
+ * to bad system deadlocks, so if the allocation failed,
+ * we need to do the VM_PAGE_WAIT() outside of the lock.
+ */
+ VM_PAGE_WAIT();
}
- } else {
- zone->free_elements = (struct zone_free_element *)element;
}
- zone->count--;
- zone->countfree++;
}
-
-/*
- * Removes an element from the zone's free list, returning 0 if the free list is empty.
- * Verifies that the next-pointer and backup next-pointer are intact,
- * and verifies that a poisoned element hasn't been modified.
- */
-static inline vm_offset_t
-try_alloc_from_zone(zone_t zone,
- boolean_t* check_poison)
+static inline bool
+zone_allocated_element_offset_is_valid(zone_t zone, vm_offset_t addr,
+ vm_offset_t page, zone_addr_kind_t kind)
{
- vm_offset_t element;
- struct zone_page_metadata *page_meta;
+ vm_offset_t offs = addr - page - ZONE_PAGE_FIRST_OFFSET(kind);
+ vm_offset_t esize = zone_elem_size(zone);
- *check_poison = FALSE;
-
- /* if zone is empty, bail */
- if (zone->use_page_list) {
- if (zone->allows_foreign && !queue_empty(&zone->pages.any_free_foreign))
- page_meta = (struct zone_page_metadata *)queue_first(&zone->pages.any_free_foreign);
- else if (!queue_empty(&zone->pages.intermediate))
- page_meta = (struct zone_page_metadata *)queue_first(&zone->pages.intermediate);
- else if (!queue_empty(&zone->pages.all_free))
- page_meta = (struct zone_page_metadata *)queue_first(&zone->pages.all_free);
- else {
- return 0;
- }
-
- /* Check if page_meta passes is_sane_zone_element */
- if (__improbable(!is_sane_zone_page_metadata(zone, (vm_offset_t)page_meta)))
- panic("zalloc: invalid metadata structure %p for freelist of zone %s\n",
- (void *) page_meta, zone->zone_name);
- assert(page_meta->zone == zone);
- element = (vm_offset_t)page_meta->elements;
+ if (esize & (esize - 1)) { /* not a power of 2 */
+ return (offs % esize) == 0;
} else {
- if (zone->free_elements == NULL)
- return 0;
-
- element = (vm_offset_t)zone->free_elements;
+ return (offs & (esize - 1)) == 0;
}
+}
-#if MACH_ASSERT
- if (__improbable(!is_sane_zone_element(zone, element)))
- panic("zfree: invalid head pointer %p for freelist of zone %s\n",
- (void *) element, zone->zone_name);
+__attribute__((always_inline))
+static struct zone_page_metadata *
+zone_allocated_element_resolve(zone_t zone, vm_offset_t addr,
+ vm_offset_t *pagep, zone_addr_kind_t *kindp)
+{
+ struct zone_page_metadata *meta;
+ zone_addr_kind_t kind;
+ vm_offset_t page;
+ vm_offset_t esize = zone_elem_size(zone);
+
+ kind = zone_addr_kind(addr, esize);
+ page = trunc_page(addr);
+ meta = zone_meta_from_addr(addr, kind);
+
+ if (kind == ZONE_ADDR_NATIVE) {
+ if (meta->zm_secondary_page) {
+ if (meta->zm_percpu) {
+ zone_invalid_element_addr_panic(zone, addr);
+ }
+ page -= ptoa(meta->zm_page_count);
+ meta -= meta->zm_page_count;
+ }
+ } else if (!zone->allows_foreign) {
+ zone_page_metadata_foreign_confusion_panic(zone, addr);
+#if __LP64__
+ } else if (!from_foreign_range(addr, esize)) {
+ zone_invalid_foreign_addr_panic(zone, addr);
+#else
+ } else if (!pmap_kernel_va(addr)) {
+ zone_invalid_element_addr_panic(zone, addr);
#endif
+ }
- vm_offset_t *primary = (vm_offset_t *) element;
- vm_offset_t *backup = get_backup_ptr(zone->elem_size, primary);
+ if (!zone_allocated_element_offset_is_valid(zone, addr, page, kind)) {
+ zone_invalid_element_addr_panic(zone, addr);
+ }
- vm_offset_t next_element = *primary;
- vm_offset_t next_element_backup = *backup;
+ if (!zone_has_index(zone, meta->zm_index)) {
+ zone_page_metadata_index_confusion_panic(zone, addr, meta);
+ }
- /*
- * backup_ptr_mismatch_panic will determine what next_element
- * should have been, and print it appropriately
- */
- if (__improbable(!is_sane_zone_element(zone, next_element)))
- backup_ptr_mismatch_panic(zone, element, next_element, next_element_backup);
+ if (kindp) {
+ *kindp = kind;
+ }
+ if (pagep) {
+ *pagep = page;
+ }
+ return meta;
+}
- /* Check the backup pointer for the regular cookie */
- if (__improbable(next_element != (next_element_backup ^ zp_nopoison_cookie))) {
+__attribute__((always_inline))
+void
+zone_allocated_element_validate(zone_t zone, vm_offset_t addr)
+{
+ zone_allocated_element_resolve(zone, addr, NULL, NULL);
+}
- /* Check for the poisoned cookie instead */
- if (__improbable(next_element != (next_element_backup ^ zp_poisoned_cookie)))
- /* Neither cookie is valid, corruption has occurred */
- backup_ptr_mismatch_panic(zone, element, next_element, next_element_backup);
+__header_always_inline vm_offset_t
+zone_page_meta_get_freelist(zone_t zone, struct zone_page_metadata *meta,
+ vm_offset_t page)
+{
+ assert(!meta->zm_secondary_page);
+ if (meta->zm_freelist_offs == PAGE_METADATA_EMPTY_FREELIST) {
+ return 0;
+ }
- /*
- * Element was marked as poisoned, so check its integrity before using it.
- */
- *check_poison = TRUE;
+ vm_size_t size = ptoa(meta->zm_percpu ? 1 : meta->zm_page_count);
+ if (meta->zm_freelist_offs + zone_elem_size(zone) > size) {
+ zone_metadata_corruption(zone, meta, "freelist corruption");
}
- if (zone->use_page_list) {
-
- /* Make sure the page_meta is at the correct offset from the start of page */
- if (__improbable(page_meta != get_zone_page_metadata((struct zone_free_element *)element)))
- panic("zalloc: metadata located at incorrect location on page of zone %s\n",
- zone->zone_name);
+ return page + meta->zm_freelist_offs;
+}
- /* Make sure next_element belongs to the same page as page_meta */
- if (next_element) {
- if (__improbable(page_meta != get_zone_page_metadata((struct zone_free_element *)next_element)))
- panic("zalloc: next element pointer %p for element %p points to invalid element for zone %s\n",
- (void *)next_element, (void *)element, zone->zone_name);
- }
+__header_always_inline void
+zone_page_meta_set_freelist(struct zone_page_metadata *meta,
+ vm_offset_t page, vm_offset_t addr)
+{
+ assert(!meta->zm_secondary_page);
+ if (addr) {
+ meta->zm_freelist_offs = (uint16_t)(addr - page);
+ } else {
+ meta->zm_freelist_offs = PAGE_METADATA_EMPTY_FREELIST;
}
+}
- /* Remove this element from the free list */
- if (zone->use_page_list) {
-
- page_meta->elements = (struct zone_free_element *)next_element;
- page_meta->free_count--;
+static bool
+zone_page_meta_is_sane_element(zone_t zone, struct zone_page_metadata *meta,
+ vm_offset_t page, vm_offset_t element, zone_addr_kind_t kind)
+{
+ if (element == 0) {
+ /* ends of the freelist are NULL */
+ return true;
+ }
+ if (element < page + ZONE_PAGE_FIRST_OFFSET(kind)) {
+ return false;
+ }
+ vm_size_t size = ptoa(meta->zm_percpu ? 1 : meta->zm_page_count);
+ if (element > page + size - zone_elem_size(zone)) {
+ return false;
+ }
+ return true;
+}
- if (zone->allows_foreign && !from_zone_map(element, zone->elem_size)) {
- if (page_meta->free_count == 0) {
- /* move to all used */
- remqueue((queue_entry_t)page_meta);
- enqueue_tail(&zone->pages.all_used, (queue_entry_t)page_meta);
- } else {
- /* no other list transitions */
- }
- } else if (page_meta->free_count == 0) {
- /* remove from intermediate or free, move to all_used */
- remqueue((queue_entry_t)page_meta);
- enqueue_tail(&zone->pages.all_used, (queue_entry_t)page_meta);
- } else if (page_meta->alloc_count == page_meta->free_count + 1) {
- /* remove from free, move to intermediate */
- remqueue((queue_entry_t)page_meta);
- enqueue_tail(&zone->pages.intermediate, (queue_entry_t)page_meta);
+/* Routine to get the size of a zone allocated address.
+ * If the address doesnt belong to the zone maps, returns 0.
+ */
+vm_size_t
+zone_element_size(void *addr, zone_t *z)
+{
+ struct zone_page_metadata *meta;
+ struct zone *src_zone;
+
+ if (from_zone_map(addr, sizeof(void *))) {
+ meta = zone_native_meta_from_addr(addr);
+ src_zone = &zone_array[meta->zm_index];
+ if (z) {
+ *z = src_zone;
+ }
+ return zone_elem_size(src_zone);
+ }
+#if CONFIG_GZALLOC
+ if (__improbable(gzalloc_enabled())) {
+ vm_size_t gzsize;
+ if (gzalloc_element_size(addr, z, &gzsize)) {
+ return gzsize;
}
- } else {
- zone->free_elements = (struct zone_free_element *)next_element;
}
- zone->countfree--;
- zone->count++;
- zone->sum_count++;
+#endif /* CONFIG_GZALLOC */
- return element;
+ return 0;
}
+/* This function just formats the reason for the panics by redoing the checks */
+__abortlike
+static void
+zone_require_panic(zone_t zone, void *addr)
+{
+ uint32_t zindex;
+ zone_t other;
-/*
- * End of zone poisoning
- */
-
-/*
- * Fake zones for things that want to report via zprint but are not actually zones.
- */
-struct fake_zone_info {
- const char* name;
- void (*init)(int);
- void (*query)(int *,
- vm_size_t *, vm_size_t *, vm_size_t *, vm_size_t *,
- uint64_t *, int *, int *, int *);
-};
+ if (!from_zone_map(addr, zone_elem_size(zone))) {
+ panic("zone_require failed: address not in a zone (addr: %p)", addr);
+ }
-static const struct fake_zone_info fake_zones[] = {
-};
-static const unsigned int num_fake_zones =
- sizeof (fake_zones) / sizeof (fake_zones[0]);
+ zindex = zone_native_meta_from_addr(addr)->zm_index;
+ other = &zone_array[zindex];
+ if (zindex >= os_atomic_load(&num_zones, relaxed) || !other->z_self) {
+ panic("zone_require failed: invalid zone index %d "
+ "(addr: %p, expected: %s%s)", zindex,
+ addr, zone_heap_name(zone), zone->z_name);
+ } else {
+ panic("zone_require failed: address in unexpected zone id %d (%s%s) "
+ "(addr: %p, expected: %s%s)",
+ zindex, zone_heap_name(other), other->z_name,
+ addr, zone_heap_name(zone), zone->z_name);
+ }
+}
-/*
- * Zone info options
- */
-boolean_t zinfo_per_task = FALSE; /* enabled by -zinfop in boot-args */
-#define ZINFO_SLOTS 200 /* for now */
-#define ZONES_MAX (ZINFO_SLOTS - num_fake_zones - 1)
+__abortlike
+static void
+zone_id_require_panic(zone_id_t zid, void *addr)
+{
+ zone_require_panic(&zone_array[zid], addr);
+}
/*
- * Support for garbage collection of unused zone pages
+ * Routines to panic if a pointer is not mapped to an expected zone.
+ * This can be used as a means of pinning an object to the zone it is expected
+ * to be a part of. Causes a panic if the address does not belong to any
+ * specified zone, does not belong to any zone, has been freed and therefore
+ * unmapped from the zone, or the pointer contains an uninitialized value that
+ * does not belong to any zone.
*
- * The kernel virtually allocates the "zone map" submap of the kernel
- * map. When an individual zone needs more storage, memory is allocated
- * out of the zone map, and the two-level "zone_page_table" is
- * on-demand expanded so that it has entries for those pages.
- * zone_page_init()/zone_page_alloc() initialize "alloc_count"
- * to the number of zone elements that occupy the zone page (which may
- * be a minimum of 1, including if a zone element spans multiple
- * pages).
- *
- * Asynchronously, the zone_gc() logic attempts to walk zone free
- * lists to see if all the elements on a zone page are free. If
- * "collect_count" (which it increments during the scan) matches
- * "alloc_count", the zone page is a candidate for collection and the
- * physical page is returned to the VM system. During this process, the
- * first word of the zone page is re-used to maintain a linked list of
- * to-be-collected zone pages.
+ * Note that this can only work with collectable zones without foreign pages.
*/
-typedef uint32_t zone_page_index_t;
-#define ZONE_PAGE_INDEX_INVALID ((zone_page_index_t)0xFFFFFFFFU)
+void
+zone_require(zone_t zone, void *addr)
+{
+ if (__probable(from_general_submap(addr, zone_elem_size(zone)) &&
+ (zone_has_index(zone, zone_native_meta_from_addr(addr)->zm_index)))) {
+ return;
+ }
+#if CONFIG_GZALLOC
+ if (__probable(gzalloc_enabled())) {
+ return;
+ }
+#endif
+ zone_require_panic(zone, addr);
+}
-struct zone_page_table_entry {
- volatile uint16_t alloc_count;
- volatile uint16_t collect_count;
-};
+void
+zone_id_require(zone_id_t zid, vm_size_t esize, void *addr)
+{
+ if (__probable(from_general_submap(addr, esize) &&
+ (zid == zone_native_meta_from_addr(addr)->zm_index))) {
+ return;
+ }
+#if CONFIG_GZALLOC
+ if (__probable(gzalloc_enabled())) {
+ return;
+ }
+#endif
+ zone_id_require_panic(zid, addr);
+}
-#define ZONE_PAGE_USED 0
-#define ZONE_PAGE_UNUSED 0xffff
+bool
+zone_owns(zone_t zone, void *addr)
+{
+ if (__probable(from_general_submap(addr, zone_elem_size(zone)) &&
+ (zone_has_index(zone, zone_native_meta_from_addr(addr)->zm_index)))) {
+ return true;
+ }
+#if CONFIG_GZALLOC
+ if (__probable(gzalloc_enabled())) {
+ return true;
+ }
+#endif
+ return false;
+}
-/* Forwards */
-void zone_page_init(
- vm_offset_t addr,
- vm_size_t size);
+#pragma mark ZTAGS
+#if VM_MAX_TAG_ZONES
-void zone_page_alloc(
- vm_offset_t addr,
- vm_size_t size);
+// for zones with tagging enabled:
-void zone_page_free_element(
- zone_page_index_t *free_page_head,
- zone_page_index_t *free_page_tail,
- vm_offset_t addr,
- vm_size_t size);
+// calculate a pointer to the tag base entry,
+// holding either a uint32_t the first tag offset for a page in the zone map,
+// or two uint16_t tags if the page can only hold one or two elements
-void zone_page_collect(
- vm_offset_t addr,
- vm_size_t size);
+#define ZTAGBASE(zone, element) \
+ (&((uint32_t *)zone_tagbase_min)[atop((element) - zone_info.zi_map_range.min_address)])
-boolean_t zone_page_collectable(
- vm_offset_t addr,
- vm_size_t size);
+// pointer to the tag for an element
+#define ZTAG(zone, element) \
+ ({ \
+ vm_tag_t * result; \
+ if ((zone)->tags_inline) { \
+ result = (vm_tag_t *) ZTAGBASE((zone), (element)); \
+ if ((page_mask & element) >= zone_elem_size(zone)) result++; \
+ } else { \
+ result = &((vm_tag_t *)zone_tags_min)[ZTAGBASE((zone), (element))[0] + ((element) & page_mask) / zone_elem_size((zone))]; \
+ } \
+ result; \
+ })
-void zone_page_keep(
- vm_offset_t addr,
- vm_size_t size);
-void zone_display_zprint(void);
+static vm_offset_t zone_tagbase_min;
+static vm_offset_t zone_tagbase_max;
+static vm_offset_t zone_tagbase_map_size;
+static vm_map_t zone_tagbase_map;
-zone_t zone_find_largest(void);
+static vm_offset_t zone_tags_min;
+static vm_offset_t zone_tags_max;
+static vm_offset_t zone_tags_map_size;
+static vm_map_t zone_tags_map;
-/*
- * Async allocation of zones
- * This mechanism allows for bootstrapping an empty zone which is setup with
- * non-blocking flags. The first call to zalloc_noblock() will kick off a thread_call
- * to zalloc_async. We perform a zalloc() (which may block) and then an immediate free.
- * This will prime the zone for the next use.
- *
- * Currently the thread_callout function (zalloc_async) will loop through all zones
- * looking for any zone with async_pending set and do the work for it.
- *
- * NOTE: If the calling thread for zalloc_noblock is lower priority than thread_call,
- * then zalloc_noblock to an empty zone may succeed.
- */
-void zalloc_async(
- thread_call_param_t p0,
- thread_call_param_t p1);
+// simple heap allocator for allocating the tags for new memory
-static thread_call_data_t call_async_alloc;
+LCK_MTX_EARLY_DECLARE(ztLock, &zone_locks_grp); /* heap lock */
-vm_map_t zone_map = VM_MAP_NULL;
+enum{
+ ztFreeIndexCount = 8,
+ ztFreeIndexMax = (ztFreeIndexCount - 1),
+ ztTagsPerBlock = 4
+};
-zone_t zone_zone = ZONE_NULL; /* the zone containing other zones */
+struct ztBlock {
+#if __LITTLE_ENDIAN__
+ uint64_t free:1,
+ next:21,
+ prev:21,
+ size:21;
+#else
+// ztBlock needs free bit least significant
+#error !__LITTLE_ENDIAN__
+#endif
+};
+typedef struct ztBlock ztBlock;
-zone_t zinfo_zone = ZONE_NULL; /* zone of per-task zone info */
+static ztBlock * ztBlocks;
+static uint32_t ztBlocksCount;
+static uint32_t ztBlocksFree;
-/*
- * The VM system gives us an initial chunk of memory.
- * It has to be big enough to allocate the zone_zone
- * all the way through the pmap zone.
- */
+static uint32_t
+ztLog2up(uint32_t size)
+{
+ if (1 == size) {
+ size = 0;
+ } else {
+ size = 32 - __builtin_clz(size - 1);
+ }
+ return size;
+}
-vm_offset_t zdata;
-vm_size_t zdata_size;
-/*
- * Align elements that use the zone page list to 32 byte boundaries.
- */
-#define ZONE_ELEMENT_ALIGNMENT 32
+static uint32_t
+ztLog2down(uint32_t size)
+{
+ size = 31 - __builtin_clz(size);
+ return size;
+}
-#define zone_wakeup(zone) thread_wakeup((event_t)(zone))
-#define zone_sleep(zone) \
- (void) lck_mtx_sleep(&(zone)->lock, LCK_SLEEP_SPIN, (event_t)(zone), THREAD_UNINT);
+static void
+ztFault(vm_map_t map, const void * address, size_t size, uint32_t flags)
+{
+ vm_map_offset_t addr = (vm_map_offset_t) address;
+ vm_map_offset_t page, end;
+
+ page = trunc_page(addr);
+ end = round_page(addr + size);
+
+ for (; page < end; page += page_size) {
+ if (!pmap_find_phys(kernel_pmap, page)) {
+ kern_return_t __unused
+ ret = kernel_memory_populate(map, page, PAGE_SIZE,
+ KMA_KOBJECT | flags, VM_KERN_MEMORY_DIAG);
+ assert(ret == KERN_SUCCESS);
+ }
+ }
+}
-/*
- * The zone_locks_grp allows for collecting lock statistics.
- * All locks are associated to this group in zinit.
- * Look at tools/lockstat for debugging lock contention.
- */
+static boolean_t
+ztPresent(const void * address, size_t size)
+{
+ vm_map_offset_t addr = (vm_map_offset_t) address;
+ vm_map_offset_t page, end;
+ boolean_t result;
+
+ page = trunc_page(addr);
+ end = round_page(addr + size);
+ for (result = TRUE; (page < end); page += page_size) {
+ result = pmap_find_phys(kernel_pmap, page);
+ if (!result) {
+ break;
+ }
+ }
+ return result;
+}
-lck_grp_t zone_locks_grp;
-lck_grp_attr_t zone_locks_grp_attr;
-#define lock_zone_init(zone) \
-MACRO_BEGIN \
- lck_attr_setdefault(&(zone)->lock_attr); \
- lck_mtx_init_ext(&(zone)->lock, &(zone)->lock_ext, \
- &zone_locks_grp, &(zone)->lock_attr); \
-MACRO_END
+void __unused
+ztDump(boolean_t sanity);
+void __unused
+ztDump(boolean_t sanity)
+{
+ uint32_t q, cq, p;
+
+ for (q = 0; q <= ztFreeIndexMax; q++) {
+ p = q;
+ do{
+ if (sanity) {
+ cq = ztLog2down(ztBlocks[p].size);
+ if (cq > ztFreeIndexMax) {
+ cq = ztFreeIndexMax;
+ }
+ if (!ztBlocks[p].free
+ || ((p != q) && (q != cq))
+ || (ztBlocks[ztBlocks[p].next].prev != p)
+ || (ztBlocks[ztBlocks[p].prev].next != p)) {
+ kprintf("zterror at %d", p);
+ ztDump(FALSE);
+ kprintf("zterror at %d", p);
+ assert(FALSE);
+ }
+ continue;
+ }
+ kprintf("zt[%03d]%c %d, %d, %d\n",
+ p, ztBlocks[p].free ? 'F' : 'A',
+ ztBlocks[p].next, ztBlocks[p].prev,
+ ztBlocks[p].size);
+ p = ztBlocks[p].next;
+ if (p == q) {
+ break;
+ }
+ }while (p != q);
+ if (!sanity) {
+ printf("\n");
+ }
+ }
+ if (!sanity) {
+ printf("-----------------------\n");
+ }
+}
-#define lock_try_zone(zone) lck_mtx_try_lock_spin(&zone->lock)
-/*
- * Garbage collection map information
+
+#define ZTBDEQ(idx) \
+ ztBlocks[ztBlocks[(idx)].prev].next = ztBlocks[(idx)].next; \
+ ztBlocks[ztBlocks[(idx)].next].prev = ztBlocks[(idx)].prev;
+
+static void
+ztFree(zone_t zone __unused, uint32_t index, uint32_t count)
+{
+ uint32_t q, w, p, size, merge;
+
+ assert(count);
+ ztBlocksFree += count;
+
+ // merge with preceding
+ merge = (index + count);
+ if ((merge < ztBlocksCount)
+ && ztPresent(&ztBlocks[merge], sizeof(ztBlocks[merge]))
+ && ztBlocks[merge].free) {
+ ZTBDEQ(merge);
+ count += ztBlocks[merge].size;
+ }
+
+ // merge with following
+ merge = (index - 1);
+ if ((merge > ztFreeIndexMax)
+ && ztPresent(&ztBlocks[merge], sizeof(ztBlocks[merge]))
+ && ztBlocks[merge].free) {
+ size = ztBlocks[merge].size;
+ count += size;
+ index -= size;
+ ZTBDEQ(index);
+ }
+
+ q = ztLog2down(count);
+ if (q > ztFreeIndexMax) {
+ q = ztFreeIndexMax;
+ }
+ w = q;
+ // queue in order of size
+ while (TRUE) {
+ p = ztBlocks[w].next;
+ if (p == q) {
+ break;
+ }
+ if (ztBlocks[p].size >= count) {
+ break;
+ }
+ w = p;
+ }
+ ztBlocks[p].prev = index;
+ ztBlocks[w].next = index;
+
+ // fault in first
+ ztFault(zone_tags_map, &ztBlocks[index], sizeof(ztBlocks[index]), 0);
+
+ // mark first & last with free flag and size
+ ztBlocks[index].free = TRUE;
+ ztBlocks[index].size = count;
+ ztBlocks[index].prev = w;
+ ztBlocks[index].next = p;
+ if (count > 1) {
+ index += (count - 1);
+ // fault in last
+ ztFault(zone_tags_map, &ztBlocks[index], sizeof(ztBlocks[index]), 0);
+ ztBlocks[index].free = TRUE;
+ ztBlocks[index].size = count;
+ }
+}
+
+static uint32_t
+ztAlloc(zone_t zone, uint32_t count)
+{
+ uint32_t q, w, p, leftover;
+
+ assert(count);
+
+ q = ztLog2up(count);
+ if (q > ztFreeIndexMax) {
+ q = ztFreeIndexMax;
+ }
+ do{
+ w = q;
+ while (TRUE) {
+ p = ztBlocks[w].next;
+ if (p == q) {
+ break;
+ }
+ if (ztBlocks[p].size >= count) {
+ // dequeue, mark both ends allocated
+ ztBlocks[w].next = ztBlocks[p].next;
+ ztBlocks[ztBlocks[p].next].prev = w;
+ ztBlocks[p].free = FALSE;
+ ztBlocksFree -= ztBlocks[p].size;
+ if (ztBlocks[p].size > 1) {
+ ztBlocks[p + ztBlocks[p].size - 1].free = FALSE;
+ }
+
+ // fault all the allocation
+ ztFault(zone_tags_map, &ztBlocks[p], count * sizeof(ztBlocks[p]), 0);
+ // mark last as allocated
+ if (count > 1) {
+ ztBlocks[p + count - 1].free = FALSE;
+ }
+ // free remainder
+ leftover = ztBlocks[p].size - count;
+ if (leftover) {
+ ztFree(zone, p + ztBlocks[p].size - leftover, leftover);
+ }
+
+ return p;
+ }
+ w = p;
+ }
+ q++;
+ }while (q <= ztFreeIndexMax);
+
+ return -1U;
+}
+
+__startup_func
+static void
+zone_tagging_init(vm_size_t max_zonemap_size)
+{
+ kern_return_t ret;
+ vm_map_kernel_flags_t vmk_flags;
+ uint32_t idx;
+
+ // allocate submaps VM_KERN_MEMORY_DIAG
+
+ zone_tagbase_map_size = atop(max_zonemap_size) * sizeof(uint32_t);
+ vmk_flags = VM_MAP_KERNEL_FLAGS_NONE;
+ vmk_flags.vmkf_permanent = TRUE;
+ ret = kmem_suballoc(kernel_map, &zone_tagbase_min, zone_tagbase_map_size,
+ FALSE, VM_FLAGS_ANYWHERE, vmk_flags, VM_KERN_MEMORY_DIAG,
+ &zone_tagbase_map);
+
+ if (ret != KERN_SUCCESS) {
+ panic("zone_init: kmem_suballoc failed");
+ }
+ zone_tagbase_max = zone_tagbase_min + round_page(zone_tagbase_map_size);
+
+ zone_tags_map_size = 2048 * 1024 * sizeof(vm_tag_t);
+ vmk_flags = VM_MAP_KERNEL_FLAGS_NONE;
+ vmk_flags.vmkf_permanent = TRUE;
+ ret = kmem_suballoc(kernel_map, &zone_tags_min, zone_tags_map_size,
+ FALSE, VM_FLAGS_ANYWHERE, vmk_flags, VM_KERN_MEMORY_DIAG,
+ &zone_tags_map);
+
+ if (ret != KERN_SUCCESS) {
+ panic("zone_init: kmem_suballoc failed");
+ }
+ zone_tags_max = zone_tags_min + round_page(zone_tags_map_size);
+
+ ztBlocks = (ztBlock *) zone_tags_min;
+ ztBlocksCount = (uint32_t)(zone_tags_map_size / sizeof(ztBlock));
+
+ // initialize the qheads
+ lck_mtx_lock(&ztLock);
+
+ ztFault(zone_tags_map, &ztBlocks[0], sizeof(ztBlocks[0]), 0);
+ for (idx = 0; idx < ztFreeIndexCount; idx++) {
+ ztBlocks[idx].free = TRUE;
+ ztBlocks[idx].next = idx;
+ ztBlocks[idx].prev = idx;
+ ztBlocks[idx].size = 0;
+ }
+ // free remaining space
+ ztFree(NULL, ztFreeIndexCount, ztBlocksCount - ztFreeIndexCount);
+
+ lck_mtx_unlock(&ztLock);
+}
+
+static void
+ztMemoryAdd(zone_t zone, vm_offset_t mem, vm_size_t size)
+{
+ uint32_t * tagbase;
+ uint32_t count, block, blocks, idx;
+ size_t pages;
+
+ pages = atop(size);
+ tagbase = ZTAGBASE(zone, mem);
+
+ lck_mtx_lock(&ztLock);
+
+ // fault tagbase
+ ztFault(zone_tagbase_map, tagbase, pages * sizeof(uint32_t), 0);
+
+ if (!zone->tags_inline) {
+ // allocate tags
+ count = (uint32_t)(size / zone_elem_size(zone));
+ blocks = ((count + ztTagsPerBlock - 1) / ztTagsPerBlock);
+ block = ztAlloc(zone, blocks);
+ if (-1U == block) {
+ ztDump(false);
+ }
+ assert(-1U != block);
+ }
+
+ lck_mtx_unlock(&ztLock);
+
+ if (!zone->tags_inline) {
+ // set tag base for each page
+ block *= ztTagsPerBlock;
+ for (idx = 0; idx < pages; idx++) {
+ vm_offset_t esize = zone_elem_size(zone);
+ tagbase[idx] = block + (uint32_t)((ptoa(idx) + esize - 1) / esize);
+ }
+ }
+}
+
+static void
+ztMemoryRemove(zone_t zone, vm_offset_t mem, vm_size_t size)
+{
+ uint32_t * tagbase;
+ uint32_t count, block, blocks, idx;
+ size_t pages;
+
+ // set tag base for each page
+ pages = atop(size);
+ tagbase = ZTAGBASE(zone, mem);
+ block = tagbase[0];
+ for (idx = 0; idx < pages; idx++) {
+ tagbase[idx] = 0xFFFFFFFF;
+ }
+
+ lck_mtx_lock(&ztLock);
+ if (!zone->tags_inline) {
+ count = (uint32_t)(size / zone_elem_size(zone));
+ blocks = ((count + ztTagsPerBlock - 1) / ztTagsPerBlock);
+ assert(block != 0xFFFFFFFF);
+ block /= ztTagsPerBlock;
+ ztFree(NULL /* zone is unlocked */, block, blocks);
+ }
+
+ lck_mtx_unlock(&ztLock);
+}
+
+uint32_t
+zone_index_from_tag_index(uint32_t tag_zone_index, vm_size_t * elem_size)
+{
+ simple_lock(&all_zones_lock, &zone_locks_grp);
+
+ zone_index_foreach(idx) {
+ zone_t z = &zone_array[idx];
+ if (!z->tags) {
+ continue;
+ }
+ if (tag_zone_index != z->tag_zone_index) {
+ continue;
+ }
+
+ *elem_size = zone_elem_size(z);
+ simple_unlock(&all_zones_lock);
+ return idx;
+ }
+
+ simple_unlock(&all_zones_lock);
+
+ return -1U;
+}
+
+#endif /* VM_MAX_TAG_ZONES */
+#pragma mark zalloc helpers
+
+const char *
+zone_name(zone_t z)
+{
+ return z->z_name;
+}
+
+const char *
+zone_heap_name(zone_t z)
+{
+ if (__probable(z->kalloc_heap < KHEAP_ID_COUNT)) {
+ return kalloc_heap_names[z->kalloc_heap];
+ }
+ return "invalid";
+}
+
+static inline vm_size_t
+zone_submaps_approx_size(void)
+{
+ vm_size_t size = 0;
+
+ for (unsigned idx = 0; idx <= zone_last_submap_idx; idx++) {
+ size += zone_submaps[idx]->size;
+ }
+
+ return size;
+}
+
+bool
+zone_maps_owned(vm_address_t addr, vm_size_t size)
+{
+ return from_zone_map(addr, size);
+}
+
+void
+zone_map_sizes(
+ vm_map_size_t *psize,
+ vm_map_size_t *pfree,
+ vm_map_size_t *plargest_free)
+{
+ vm_map_sizes(zone_submaps[Z_SUBMAP_IDX_GENERAL_MAP], psize, pfree, plargest_free);
+}
+
+vm_map_t
+zone_submap(zone_t zone)
+{
+ return submap_for_zone(zone);
+}
+
+unsigned
+zpercpu_count(void)
+{
+ return zpercpu_early_count;
+}
+
+int
+track_this_zone(const char *zonename, const char *logname)
+{
+ unsigned int len;
+ const char *zc = zonename;
+ const char *lc = logname;
+
+ /*
+ * Compare the strings. We bound the compare by MAX_ZONE_NAME.
+ */
+
+ for (len = 1; len <= MAX_ZONE_NAME; zc++, lc++, len++) {
+ /*
+ * If the current characters don't match, check for a space in
+ * in the zone name and a corresponding period in the log name.
+ * If that's not there, then the strings don't match.
+ */
+
+ if (*zc != *lc && !(*zc == ' ' && *lc == '.')) {
+ break;
+ }
+
+ /*
+ * The strings are equal so far. If we're at the end, then it's a match.
+ */
+
+ if (*zc == '\0') {
+ return TRUE;
+ }
+ }
+
+ return FALSE;
+}
+
+#if DEBUG || DEVELOPMENT
+
+vm_size_t
+zone_element_info(void *addr, vm_tag_t * ptag)
+{
+ vm_size_t size = 0;
+ vm_tag_t tag = VM_KERN_MEMORY_NONE;
+ struct zone_page_metadata *meta;
+ struct zone *src_zone;
+
+ if (from_zone_map(addr, sizeof(void *))) {
+ meta = zone_native_meta_from_addr(addr);
+ src_zone = &zone_array[meta->zm_index];
+#if VM_MAX_TAG_ZONES
+ if (__improbable(src_zone->tags)) {
+ tag = (ZTAG(src_zone, (vm_offset_t) addr)[0] >> 1);
+ }
+#endif /* VM_MAX_TAG_ZONES */
+ size = zone_elem_size(src_zone);
+ } else {
+#if CONFIG_GZALLOC
+ gzalloc_element_size(addr, NULL, &size);
+#endif /* CONFIG_GZALLOC */
+ }
+ *ptag = tag;
+ return size;
+}
+
+#endif /* DEBUG || DEVELOPMENT */
+
+/* Someone wrote to freed memory. */
+__abortlike
+static void
+zone_element_was_modified_panic(
+ zone_t zone,
+ vm_offset_t element,
+ vm_offset_t found,
+ vm_offset_t expected,
+ vm_offset_t offset)
+{
+ panic("a freed zone element has been modified in zone %s%s: "
+ "expected %p but found %p, bits changed %p, "
+ "at offset %d of %d in element %p, cookies %p %p",
+ zone_heap_name(zone),
+ zone->z_name,
+ (void *) expected,
+ (void *) found,
+ (void *) (expected ^ found),
+ (uint32_t) offset,
+ (uint32_t) zone_elem_size(zone),
+ (void *) element,
+ (void *) zp_nopoison_cookie,
+ (void *) zp_poisoned_cookie);
+}
+
+/* The backup pointer is stored in the last pointer-sized location in an element. */
+__header_always_inline vm_offset_t *
+get_backup_ptr(vm_size_t elem_size, vm_offset_t *element)
+{
+ return (vm_offset_t *)((vm_offset_t)element + elem_size - sizeof(vm_offset_t));
+}
+
+/*
+ * The primary and backup pointers don't match.
+ * Determine which one was likely the corrupted pointer, find out what it
+ * probably should have been, and panic.
*/
-#define ZONE_PAGE_TABLE_FIRST_LEVEL_SIZE (32)
-struct zone_page_table_entry * volatile zone_page_table[ZONE_PAGE_TABLE_FIRST_LEVEL_SIZE];
-vm_size_t zone_page_table_used_size;
-unsigned int zone_pages;
-unsigned int zone_page_table_second_level_size; /* power of 2 */
-unsigned int zone_page_table_second_level_shift_amount;
+__abortlike
+static void
+backup_ptr_mismatch_panic(
+ zone_t zone,
+ struct zone_page_metadata *page_meta,
+ vm_offset_t page,
+ vm_offset_t element)
+{
+ vm_offset_t primary = *(vm_offset_t *)element;
+ vm_offset_t backup = *get_backup_ptr(zone_elem_size(zone), &element);
+ vm_offset_t likely_backup;
+ vm_offset_t likely_primary;
+ zone_addr_kind_t kind = zone_addr_kind(page, zone_elem_size(zone));
+
+ likely_primary = primary ^ zp_nopoison_cookie;
+ boolean_t sane_backup;
+ boolean_t sane_primary = zone_page_meta_is_sane_element(zone, page_meta,
+ page, likely_primary, kind);
+ boolean_t element_was_poisoned = (backup & 0x1);
+
+#if defined(__LP64__)
+ /* We can inspect the tag in the upper bits for additional confirmation */
+ if ((backup & 0xFFFFFF0000000000) == 0xFACADE0000000000) {
+ element_was_poisoned = TRUE;
+ } else if ((backup & 0xFFFFFF0000000000) == 0xC0FFEE0000000000) {
+ element_was_poisoned = FALSE;
+ }
+#endif
-#define zone_page_table_first_level_slot(x) ((x) >> zone_page_table_second_level_shift_amount)
-#define zone_page_table_second_level_slot(x) ((x) & (zone_page_table_second_level_size - 1))
+ if (element_was_poisoned) {
+ likely_backup = backup ^ zp_poisoned_cookie;
+ } else {
+ likely_backup = backup ^ zp_nopoison_cookie;
+ }
+ sane_backup = zone_page_meta_is_sane_element(zone, page_meta,
+ page, likely_backup, kind);
-void zone_page_table_expand(zone_page_index_t pindex);
-struct zone_page_table_entry *zone_page_table_lookup(zone_page_index_t pindex);
+ /* The primary is definitely the corrupted one */
+ if (!sane_primary && sane_backup) {
+ zone_element_was_modified_panic(zone, element, primary, (likely_backup ^ zp_nopoison_cookie), 0);
+ }
+
+ /* The backup is definitely the corrupted one */
+ if (sane_primary && !sane_backup) {
+ zone_element_was_modified_panic(zone, element, backup,
+ (likely_primary ^ (element_was_poisoned ? zp_poisoned_cookie : zp_nopoison_cookie)),
+ zone_elem_size(zone) - sizeof(vm_offset_t));
+ }
+
+ /*
+ * Not sure which is the corrupted one.
+ * It's less likely that the backup pointer was overwritten with
+ * ( (sane address) ^ (valid cookie) ), so we'll guess that the
+ * primary pointer has been overwritten with a sane but incorrect address.
+ */
+ if (sane_primary && sane_backup) {
+ zone_element_was_modified_panic(zone, element, primary, (likely_backup ^ zp_nopoison_cookie), 0);
+ }
+
+ /* Neither are sane, so just guess. */
+ zone_element_was_modified_panic(zone, element, primary, (likely_backup ^ zp_nopoison_cookie), 0);
+}
/*
- * Exclude more than one concurrent garbage collection
+ * zone_sequestered_page_get
+ * z is locked
*/
-decl_lck_mtx_data(, zone_gc_lock)
+static struct zone_page_metadata *
+zone_sequestered_page_get(zone_t z, vm_offset_t *page)
+{
+ const zone_addr_kind_t kind = ZONE_ADDR_NATIVE;
-lck_attr_t zone_gc_lck_attr;
-lck_grp_t zone_gc_lck_grp;
-lck_grp_attr_t zone_gc_lck_grp_attr;
-lck_mtx_ext_t zone_gc_lck_ext;
+ if (!zone_pva_is_null(z->pages_sequester)) {
+ if (os_sub_overflow(z->sequester_page_count, z->alloc_pages,
+ &z->sequester_page_count)) {
+ zone_accounting_panic(z, "sequester_page_count wrap-around");
+ }
+ return zone_meta_queue_pop(z, &z->pages_sequester, kind, page);
+ }
+
+ return NULL;
+}
/*
- * Protects first_zone, last_zone, num_zones,
- * and the next_zone field of zones.
+ * zone_sequestered_page_populate
+ * z is unlocked
+ * page_meta is invalid on failure
*/
-decl_simple_lock_data(, all_zones_lock)
-zone_t first_zone;
-zone_t *last_zone;
-unsigned int num_zones;
+static kern_return_t
+zone_sequestered_page_populate(zone_t z, struct zone_page_metadata *page_meta,
+ vm_offset_t space, vm_size_t alloc_size, int zflags)
+{
+ kern_return_t retval;
-boolean_t zone_gc_allowed = TRUE;
-boolean_t zone_gc_forced = FALSE;
-boolean_t panic_include_zprint = FALSE;
-boolean_t zone_gc_allowed_by_time_throttle = TRUE;
+ assert(alloc_size == ptoa(z->alloc_pages));
+ retval = kernel_memory_populate(submap_for_zone(z), space, alloc_size,
+ zflags, VM_KERN_MEMORY_ZONE);
+ if (retval != KERN_SUCCESS) {
+ lock_zone(z);
+ zone_meta_queue_push(z, &z->pages_sequester, page_meta, ZONE_ADDR_NATIVE);
+ z->sequester_page_count += z->alloc_pages;
+ unlock_zone(z);
+ }
+ return retval;
+}
-vm_offset_t panic_kext_memory_info = 0;
-vm_size_t panic_kext_memory_size = 0;
+#pragma mark Zone poisoning/zeroing
-#define ZALLOC_DEBUG_ZONEGC 0x00000001
-#define ZALLOC_DEBUG_ZCRAM 0x00000002
-uint32_t zalloc_debug = 0;
+/*
+ * Initialize zone poisoning
+ * called from zone_bootstrap before any allocations are made from zalloc
+ */
+__startup_func
+static void
+zp_bootstrap(void)
+{
+ char temp_buf[16];
+
+ /*
+ * Initialize backup pointer random cookie for poisoned elements
+ * Try not to call early_random() back to back, it may return
+ * the same value if mach_absolute_time doesn't have sufficient time
+ * to tick over between calls. <rdar://problem/11597395>
+ * (This is only a problem on embedded devices)
+ */
+ zp_poisoned_cookie = (uintptr_t) early_random();
+
+ /* -zp: enable poisoning for every alloc and free */
+ if (PE_parse_boot_argn("-zp", temp_buf, sizeof(temp_buf))) {
+ zp_factor = 1;
+ }
+
+ /* -no-zp: disable poisoning */
+ if (PE_parse_boot_argn("-no-zp", temp_buf, sizeof(temp_buf))) {
+ zp_factor = 0;
+ printf("Zone poisoning disabled\n");
+ }
+
+ /* Initialize backup pointer random cookie for unpoisoned elements */
+ zp_nopoison_cookie = (uintptr_t) early_random();
+
+#if MACH_ASSERT
+ if (zp_poisoned_cookie == zp_nopoison_cookie) {
+ panic("early_random() is broken: %p and %p are not random\n",
+ (void *) zp_poisoned_cookie, (void *) zp_nopoison_cookie);
+ }
+#endif
+
+ /*
+ * Use the last bit in the backup pointer to hint poisoning state
+ * to backup_ptr_mismatch_panic. Valid zone pointers are aligned, so
+ * the low bits are zero.
+ */
+ zp_poisoned_cookie |= (uintptr_t)0x1ULL;
+ zp_nopoison_cookie &= ~((uintptr_t)0x1ULL);
+
+#if defined(__LP64__)
+ /*
+ * Make backup pointers more obvious in GDB for 64 bit
+ * by making OxFFFFFF... ^ cookie = 0xFACADE...
+ * (0xFACADE = 0xFFFFFF ^ 0x053521)
+ * (0xC0FFEE = 0xFFFFFF ^ 0x3f0011)
+ * The high 3 bytes of a zone pointer are always 0xFFFFFF, and are checked
+ * by the sanity check, so it's OK for that part of the cookie to be predictable.
+ *
+ * TODO: Use #defines, xors, and shifts
+ */
+
+ zp_poisoned_cookie &= 0x000000FFFFFFFFFF;
+ zp_poisoned_cookie |= 0x0535210000000000; /* 0xFACADE */
+
+ zp_nopoison_cookie &= 0x000000FFFFFFFFFF;
+ zp_nopoison_cookie |= 0x3f00110000000000; /* 0xC0FFEE */
+#endif
+
+ /*
+ * Initialize zp_min_size to two cachelines. Elements smaller than this will
+ * be zero-ed.
+ */
+ ml_cpu_info_t cpu_info;
+ ml_cpu_get_info(&cpu_info);
+ zp_min_size = 2 * cpu_info.cache_line_size;
+}
+
+inline uint32_t
+zone_poison_count_init(zone_t zone)
+{
+ return zp_factor + (((uint32_t)zone_elem_size(zone)) >> zp_scale) ^
+ (mach_absolute_time() & 0x7);
+}
+
+#if ZALLOC_ENABLE_POISONING
+static bool
+zfree_poison_element(zone_t zone, uint32_t *zp_count, vm_offset_t elem)
+{
+ bool poison = false;
+ uint32_t zp_count_local;
+
+ assert(!zone->percpu);
+ if (zp_factor != 0) {
+ /*
+ * Poison the memory of every zp_count-th element before it ends up
+ * on the freelist to catch use-after-free and use of uninitialized
+ * memory.
+ *
+ * Every element is poisoned when zp_factor is set to 1.
+ *
+ */
+ zp_count_local = os_atomic_load(zp_count, relaxed);
+ if (__improbable(zp_count_local == 0 || zp_factor == 1)) {
+ poison = true;
+
+ os_atomic_store(zp_count, zone_poison_count_init(zone), relaxed);
+
+ /* memset_pattern{4|8} could help make this faster: <rdar://problem/4662004> */
+ vm_offset_t *element_cursor = ((vm_offset_t *) elem);
+ vm_offset_t *end_cursor = (vm_offset_t *)(elem + zone_elem_size(zone));
+
+ for (; element_cursor < end_cursor; element_cursor++) {
+ *element_cursor = ZONE_POISON;
+ }
+ } else {
+ os_atomic_store(zp_count, zp_count_local - 1, relaxed);
+ /*
+ * Zero first zp_min_size bytes of elements that aren't being poisoned.
+ * Element size is larger than zp_min_size in this path as elements
+ * that are smaller will always be zero-ed.
+ */
+ bzero((void *) elem, zp_min_size);
+ }
+ }
+ return poison;
+}
+#else
+static bool
+zfree_poison_element(zone_t zone, uint32_t *zp_count, vm_offset_t elem)
+{
+#pragma unused(zone, zp_count, elem)
+ assert(!zone->percpu);
+ return false;
+}
+#endif
+
+__attribute__((always_inline))
+static bool
+zfree_clear(zone_t zone, vm_offset_t addr, vm_size_t elem_size)
+{
+ assert(zone->zfree_clear_mem);
+ if (zone->percpu) {
+ zpercpu_foreach_cpu(i) {
+ bzero((void *)(addr + ptoa(i)), elem_size);
+ }
+ } else {
+ bzero((void *)addr, elem_size);
+ }
+
+ return true;
+}
+
+/*
+ * Zero the element if zone has zfree_clear_mem flag set else poison
+ * the element if zp_count hits 0.
+ */
+__attribute__((always_inline))
+bool
+zfree_clear_or_poison(zone_t zone, uint32_t *zp_count, vm_offset_t addr)
+{
+ vm_size_t elem_size = zone_elem_size(zone);
+
+ if (zone->zfree_clear_mem) {
+ return zfree_clear(zone, addr, elem_size);
+ }
+
+ return zfree_poison_element(zone, zp_count, (vm_offset_t)addr);
+}
+
+/*
+ * Clear out the old next pointer and backup to avoid leaking the zone
+ * poisoning cookie and so that only values on the freelist have a valid
+ * cookie.
+ */
+void
+zone_clear_freelist_pointers(zone_t zone, vm_offset_t addr)
+{
+ vm_offset_t perm_value = 0;
+
+ if (!zone->zfree_clear_mem) {
+ perm_value = ZONE_POISON;
+ }
+
+ vm_offset_t *primary = (vm_offset_t *) addr;
+ vm_offset_t *backup = get_backup_ptr(zone_elem_size(zone), primary);
+
+ *primary = perm_value;
+ *backup = perm_value;
+}
+
+#if ZALLOC_ENABLE_POISONING
+__abortlike
+static void
+zone_element_not_clear_panic(zone_t zone, void *addr)
+{
+ panic("Zone element %p was modified after free for zone %s%s: "
+ "Expected element to be cleared", addr, zone_heap_name(zone),
+ zone->z_name);
+}
+
+/*
+ * Validate that the element was not tampered with while it was in the
+ * freelist.
+ */
+void
+zalloc_validate_element(zone_t zone, vm_offset_t addr, vm_size_t size, bool validate)
+{
+ if (zone->percpu) {
+ assert(zone->zfree_clear_mem);
+ zpercpu_foreach_cpu(i) {
+ if (memcmp_zero_ptr_aligned((void *)(addr + ptoa(i)), size)) {
+ zone_element_not_clear_panic(zone, (void *)(addr + ptoa(i)));
+ }
+ }
+ } else if (zone->zfree_clear_mem) {
+ if (memcmp_zero_ptr_aligned((void *)addr, size)) {
+ zone_element_not_clear_panic(zone, (void *)addr);
+ }
+ } else if (__improbable(validate)) {
+ const vm_offset_t *p = (vm_offset_t *)addr;
+ const vm_offset_t *end = (vm_offset_t *)(addr + size);
+
+ for (; p < end; p++) {
+ if (*p != ZONE_POISON) {
+ zone_element_was_modified_panic(zone, addr,
+ *p, ZONE_POISON, (vm_offset_t)p - addr);
+ }
+ }
+ } else {
+ /*
+ * If element wasn't poisoned or entirely cleared, validate that the
+ * minimum bytes that were cleared on free haven't been corrupted.
+ * addr is advanced by ptr size as we have already validated and cleared
+ * the freelist pointer/zcache canary.
+ */
+ if (memcmp_zero_ptr_aligned((void *) (addr + sizeof(vm_offset_t)),
+ zp_min_size - sizeof(vm_offset_t))) {
+ zone_element_not_clear_panic(zone, (void *)addr);
+ }
+ }
+}
+#endif /* ZALLOC_ENABLE_POISONING */
+
+#pragma mark Zone Leak Detection
/*
* Zone leak debugging code
* off by default.
*
* Enable the logging via the boot-args. Add the parameter "zlog=<zone>" to boot-args where <zone>
- * is the name of the zone you wish to log.
+ * is the name of the zone you wish to log.
*
* This code only tracks one zone, so you need to identify which one is leaking first.
* Generally, you'll know you have a leak when you get a "zalloc retry failed 3" panic from the zone
* corrupted to examine its history. This should lead to the source of the corruption.
*/
-static int log_records; /* size of the log, expressed in number of records */
-
-#define MAX_ZONE_NAME 32 /* max length of a zone name we can take from the boot-args */
+/* Returns TRUE if we rolled over the counter at factor */
+__header_always_inline bool
+sample_counter(volatile uint32_t *count_p, uint32_t factor)
+{
+ uint32_t old_count, new_count = 0;
+ if (count_p != NULL) {
+ os_atomic_rmw_loop(count_p, old_count, new_count, relaxed, {
+ new_count = old_count + 1;
+ if (new_count >= factor) {
+ new_count = 0;
+ }
+ });
+ }
-static char zone_name_to_log[MAX_ZONE_NAME] = ""; /* the zone name we're logging, if any */
+ return new_count == 0;
+}
+#if ZONE_ENABLE_LOGGING
/* Log allocations and frees to help debug a zone element corruption */
-boolean_t corruption_debug_flag = FALSE; /* enabled by "-zc" boot-arg */
+TUNABLE(bool, corruption_debug_flag, "-zc", false);
+
+#define MAX_NUM_ZONES_ALLOWED_LOGGING 10 /* Maximum 10 zones can be logged at once */
+
+static int max_num_zones_to_log = MAX_NUM_ZONES_ALLOWED_LOGGING;
+static int num_zones_logged = 0;
/*
- * The number of records in the log is configurable via the zrecs parameter in boot-args. Set this to
- * the number of records you want in the log. For example, "zrecs=1000" sets it to 1000 records. Note
- * that the larger the size of the log, the slower the system will run due to linear searching in the log,
- * but one doesn't generally care about performance when tracking down a leak. The log is capped at 8000
- * records since going much larger than this tends to make the system unresponsive and unbootable on small
- * memory configurations. The default value is 4000 records.
+ * The number of records in the log is configurable via the zrecs parameter in boot-args. Set this to
+ * the number of records you want in the log. For example, "zrecs=10" sets it to 10 records. Since this
+ * is the number of stacks suspected of leaking, we don't need many records.
*/
-#if defined(__LP64__)
-#define ZRECORDS_MAX 128000 /* Max records allowed in the log */
+#if defined(__LP64__)
+#define ZRECORDS_MAX 2560 /* Max records allowed in the log */
#else
-#define ZRECORDS_MAX 8000 /* Max records allowed in the log */
+#define ZRECORDS_MAX 1536 /* Max records allowed in the log */
#endif
-#define ZRECORDS_DEFAULT 4000 /* default records in log if zrecs is not specificed in boot-args */
+#define ZRECORDS_DEFAULT 1024 /* default records in log if zrecs is not specificed in boot-args */
+
+static TUNABLE(uint32_t, log_records, "zrecs", ZRECORDS_DEFAULT);
+
+static void
+zone_enable_logging(zone_t z)
+{
+ z->zlog_btlog = btlog_create(log_records, MAX_ZTRACE_DEPTH,
+ (corruption_debug_flag == FALSE) /* caller_will_remove_entries_for_element? */);
+
+ if (z->zlog_btlog) {
+ printf("zone: logging started for zone %s%s\n",
+ zone_heap_name(z), z->z_name);
+ } else {
+ printf("zone: couldn't allocate memory for zrecords, turning off zleak logging\n");
+ z->zone_logging = false;
+ }
+}
+
+/**
+ * @function zone_setup_logging
+ *
+ * @abstract
+ * Optionally sets up a zone for logging.
+ *
+ * @discussion
+ * We recognized two boot-args:
+ *
+ * zlog=<zone_to_log>
+ * zrecs=<num_records_in_log>
+ *
+ * The zlog arg is used to specify the zone name that should be logged,
+ * and zrecs is used to control the size of the log.
+ *
+ * If zrecs is not specified, a default value is used.
+ */
+static void
+zone_setup_logging(zone_t z)
+{
+ char zone_name[MAX_ZONE_NAME]; /* Temp. buffer for the zone name */
+ char zlog_name[MAX_ZONE_NAME]; /* Temp. buffer to create the strings zlog1, zlog2 etc... */
+ char zlog_val[MAX_ZONE_NAME]; /* the zone name we're logging, if any */
+
+ /*
+ * Don't allow more than ZRECORDS_MAX records even if the user asked for more.
+ *
+ * This prevents accidentally hogging too much kernel memory
+ * and making the system unusable.
+ */
+ if (log_records > ZRECORDS_MAX) {
+ log_records = ZRECORDS_MAX;
+ }
+
+ /*
+ * Append kalloc heap name to zone name (if zone is used by kalloc)
+ */
+ snprintf(zone_name, MAX_ZONE_NAME, "%s%s", zone_heap_name(z), z->z_name);
+
+ /* zlog0 isn't allowed. */
+ for (int i = 1; i <= max_num_zones_to_log; i++) {
+ snprintf(zlog_name, MAX_ZONE_NAME, "zlog%d", i);
+
+ if (PE_parse_boot_argn(zlog_name, zlog_val, sizeof(zlog_val)) &&
+ track_this_zone(zone_name, zlog_val)) {
+ z->zone_logging = true;
+ num_zones_logged++;
+ break;
+ }
+ }
+
+ /*
+ * Backwards compat. with the old boot-arg used to specify single zone
+ * logging i.e. zlog Needs to happen after the newer zlogn checks
+ * because the prefix will match all the zlogn
+ * boot-args.
+ */
+ if (!z->zone_logging &&
+ PE_parse_boot_argn("zlog", zlog_val, sizeof(zlog_val)) &&
+ track_this_zone(zone_name, zlog_val)) {
+ z->zone_logging = true;
+ num_zones_logged++;
+ }
+
+
+ /*
+ * If we want to log a zone, see if we need to allocate buffer space for
+ * the log.
+ *
+ * Some vm related zones are zinit'ed before we can do a kmem_alloc, so
+ * we have to defer allocation in that case.
+ *
+ * zone_init() will finish the job.
+ *
+ * If we want to log one of the VM related zones that's set up early on,
+ * we will skip allocation of the log until zinit is called again later
+ * on some other zone.
+ */
+ if (z->zone_logging && startup_phase >= STARTUP_SUB_KMEM_ALLOC) {
+ zone_enable_logging(z);
+ }
+}
/*
* Each record in the log contains a pointer to the zone element it refers to,
*/
-/*
- * Opcodes for the btlog operation field:
- */
-
-#define ZOP_ALLOC 1
-#define ZOP_FREE 0
-
-/*
- * The allocation log and all the related variables are protected by the zone lock for the zone_of_interest
- */
-static btlog_t *zlog_btlog; /* the log itself, dynamically allocated when logging is enabled */
-static zone_t zone_of_interest = NULL; /* the zone being watched; corresponds to zone_name_to_log */
-
/*
* Decide if we want to log this zone by doing a string compare between a zone name and the name
* of the zone to log. Return true if the strings are equal, false otherwise. Because it's not
* match a space in the zone name.
*/
-static int
-log_this_zone(const char *zonename, const char *logname)
-{
- int len;
- const char *zc = zonename;
- const char *lc = logname;
-
- /*
- * Compare the strings. We bound the compare by MAX_ZONE_NAME.
- */
-
- for (len = 1; len <= MAX_ZONE_NAME; zc++, lc++, len++) {
-
- /*
- * If the current characters don't match, check for a space in
- * in the zone name and a corresponding period in the log name.
- * If that's not there, then the strings don't match.
- */
-
- if (*zc != *lc && !(*zc == ' ' && *lc == '.'))
- break;
-
- /*
- * The strings are equal so far. If we're at the end, then it's a match.
- */
-
- if (*zc == '\0')
- return TRUE;
- }
-
- return FALSE;
-}
-
-
/*
* Test if we want to log this zalloc/zfree event. We log if this is the zone we're interested in and
* the buffer for the records has been allocated.
*/
-#define DO_LOGGING(z) (zlog_btlog && (z) == zone_of_interest)
-
-extern boolean_t kmem_alloc_ready;
+#define DO_LOGGING(z) (z->zlog_btlog != NULL)
+#else /* !ZONE_ENABLE_LOGGING */
+#define DO_LOGGING(z) 0
+#endif /* !ZONE_ENABLE_LOGGING */
#if CONFIG_ZLEAKS
-#pragma mark -
-#pragma mark Zone Leak Detection
-/*
+/*
* The zone leak detector, abbreviated 'zleak', keeps track of a subset of the currently outstanding
* allocations made by the zone allocator. Every zleak_sample_factor allocations in each zone, we capture a
- * backtrace. Every free, we examine the table and determine if the allocation was being tracked,
+ * backtrace. Every free, we examine the table and determine if the allocation was being tracked,
* and stop tracking it if it was being tracked.
*
- * We track the allocations in the zallocations hash table, which stores the address that was returned from
+ * We track the allocations in the zallocations hash table, which stores the address that was returned from
* the zone allocator. Each stored entry in the zallocations table points to an entry in the ztraces table, which
* stores the backtrace associated with that allocation. This provides uniquing for the relatively large
* backtraces - we don't store them more than once.
* Data collection begins when the zone map is 50% full, and only occurs for zones that are taking up
* a large amount of virtual space.
*/
-#define ZLEAK_STATE_ENABLED 0x01 /* Zone leak monitoring should be turned on if zone_map fills up. */
-#define ZLEAK_STATE_ACTIVE 0x02 /* We are actively collecting traces. */
-#define ZLEAK_STATE_ACTIVATING 0x04 /* Some thread is doing setup; others should move along. */
-#define ZLEAK_STATE_FAILED 0x08 /* Attempt to allocate tables failed. We will not try again. */
-uint32_t zleak_state = 0; /* State of collection, as above */
+#define ZLEAK_STATE_ENABLED 0x01 /* Zone leak monitoring should be turned on if zone_map fills up. */
+#define ZLEAK_STATE_ACTIVE 0x02 /* We are actively collecting traces. */
+#define ZLEAK_STATE_ACTIVATING 0x04 /* Some thread is doing setup; others should move along. */
+#define ZLEAK_STATE_FAILED 0x08 /* Attempt to allocate tables failed. We will not try again. */
+uint32_t zleak_state = 0; /* State of collection, as above */
-boolean_t panic_include_ztrace = FALSE; /* Enable zleak logging on panic */
-vm_size_t zleak_global_tracking_threshold; /* Size of zone map at which to start collecting data */
-vm_size_t zleak_per_zone_tracking_threshold; /* Size a zone will have before we will collect data on it */
-unsigned int zleak_sample_factor = 1000; /* Allocations per sample attempt */
+boolean_t panic_include_ztrace = FALSE; /* Enable zleak logging on panic */
+vm_size_t zleak_global_tracking_threshold; /* Size of zone map at which to start collecting data */
+vm_size_t zleak_per_zone_tracking_threshold; /* Size a zone will have before we will collect data on it */
+unsigned int zleak_sample_factor = 1000; /* Allocations per sample attempt */
/*
* Counters for allocation statistics.
- */
+ */
/* Times two active records want to occupy the same spot */
unsigned int z_alloc_collisions = 0;
unsigned int z_trace_overwrites = 0;
/* Times a new alloc or trace is put into the hash table */
-unsigned int z_alloc_recorded = 0;
-unsigned int z_trace_recorded = 0;
+unsigned int z_alloc_recorded = 0;
+unsigned int z_trace_recorded = 0;
/* Times zleak_log returned false due to not being able to acquire the lock */
-unsigned int z_total_conflicts = 0;
+unsigned int z_total_conflicts = 0;
-
-#pragma mark struct zallocation
/*
* Structure for keeping track of an allocation
* An allocation bucket is in use if its element is not NULL
*/
struct zallocation {
- uintptr_t za_element; /* the element that was zalloc'ed or zfree'ed, NULL if bucket unused */
- vm_size_t za_size; /* how much memory did this allocation take up? */
- uint32_t za_trace_index; /* index into ztraces for backtrace associated with allocation */
+ uintptr_t za_element; /* the element that was zalloc'ed or zfree'ed, NULL if bucket unused */
+ vm_size_t za_size; /* how much memory did this allocation take up? */
+ uint32_t za_trace_index; /* index into ztraces for backtrace associated with allocation */
/* TODO: #if this out */
- uint32_t za_hit_count; /* for determining effectiveness of hash function */
+ uint32_t za_hit_count; /* for determining effectiveness of hash function */
};
/* Size must be a power of two for the zhash to be able to just mask off bits instead of mod */
vm_size_t zleak_max_zonemap_size;
/* Hashmaps of allocations and their corresponding traces */
-static struct zallocation* zallocations;
-static struct ztrace* ztraces;
+static struct zallocation* zallocations;
+static struct ztrace* ztraces;
/* not static so that panic can see this, see kern/debug.c */
-struct ztrace* top_ztrace;
+struct ztrace* top_ztrace;
/* Lock to protect zallocations, ztraces, and top_ztrace from concurrent modification. */
-static lck_spin_t zleak_lock;
-static lck_attr_t zleak_lock_attr;
-static lck_grp_t zleak_lock_grp;
-static lck_grp_attr_t zleak_lock_grp_attr;
+LCK_GRP_DECLARE(zleak_lock_grp, "zleak_lock");
+LCK_SPIN_DECLARE(zleak_lock, &zleak_lock_grp);
/*
* Initializes the zone leak monitor. Called from zone_init()
*/
-static void
-zleak_init(vm_size_t max_zonemap_size)
+__startup_func
+static void
+zleak_init(vm_size_t max_zonemap_size)
{
- char scratch_buf[16];
- boolean_t zleak_enable_flag = FALSE;
+ char scratch_buf[16];
+ boolean_t zleak_enable_flag = FALSE;
zleak_max_zonemap_size = max_zonemap_size;
- zleak_global_tracking_threshold = max_zonemap_size / 2;
+ zleak_global_tracking_threshold = max_zonemap_size / 2;
zleak_per_zone_tracking_threshold = zleak_global_tracking_threshold / 8;
+#if CONFIG_EMBEDDED
+ if (PE_parse_boot_argn("-zleakon", scratch_buf, sizeof(scratch_buf))) {
+ zleak_enable_flag = TRUE;
+ printf("zone leak detection enabled\n");
+ } else {
+ zleak_enable_flag = FALSE;
+ printf("zone leak detection disabled\n");
+ }
+#else /* CONFIG_EMBEDDED */
/* -zleakoff (flag to disable zone leak monitor) */
if (PE_parse_boot_argn("-zleakoff", scratch_buf, sizeof(scratch_buf))) {
zleak_enable_flag = FALSE;
zleak_enable_flag = TRUE;
printf("zone leak detection enabled\n");
}
-
+#endif /* CONFIG_EMBEDDED */
+
/* zfactor=XXXX (override how often to sample the zone allocator) */
if (PE_parse_boot_argn("zfactor", &zleak_sample_factor, sizeof(zleak_sample_factor))) {
printf("Zone leak factor override: %u\n", zleak_sample_factor);
if (PE_parse_boot_argn("zleak-allocs", &zleak_alloc_buckets, sizeof(zleak_alloc_buckets))) {
printf("Zone leak alloc buckets override: %u\n", zleak_alloc_buckets);
/* uses 'is power of 2' trick: (0x01000 & 0x00FFF == 0) */
- if (zleak_alloc_buckets == 0 || (zleak_alloc_buckets & (zleak_alloc_buckets-1))) {
+ if (zleak_alloc_buckets == 0 || (zleak_alloc_buckets & (zleak_alloc_buckets - 1))) {
printf("Override isn't a power of two, bad things might happen!\n");
}
}
-
+
/* zleak-traces=XXXX (override number of buckets in ztraces) */
if (PE_parse_boot_argn("zleak-traces", &zleak_trace_buckets, sizeof(zleak_trace_buckets))) {
printf("Zone leak trace buckets override: %u\n", zleak_trace_buckets);
/* uses 'is power of 2' trick: (0x01000 & 0x00FFF == 0) */
- if (zleak_trace_buckets == 0 || (zleak_trace_buckets & (zleak_trace_buckets-1))) {
+ if (zleak_trace_buckets == 0 || (zleak_trace_buckets & (zleak_trace_buckets - 1))) {
printf("Override isn't a power of two, bad things might happen!\n");
}
}
-
- /* allocate the zleak_lock */
- lck_grp_attr_setdefault(&zleak_lock_grp_attr);
- lck_grp_init(&zleak_lock_grp, "zleak_lock", &zleak_lock_grp_attr);
- lck_attr_setdefault(&zleak_lock_attr);
- lck_spin_init(&zleak_lock, &zleak_lock_grp, &zleak_lock_attr);
-
+
if (zleak_enable_flag) {
zleak_state = ZLEAK_STATE_ENABLED;
}
}
-#if CONFIG_ZLEAKS
-
/*
* Support for kern.zleak.active sysctl - a simplified
* version of the zleak_state variable.
int
get_zleak_state(void)
{
- if (zleak_state & ZLEAK_STATE_FAILED)
- return (-1);
- if (zleak_state & ZLEAK_STATE_ACTIVE)
- return (1);
- return (0);
+ if (zleak_state & ZLEAK_STATE_FAILED) {
+ return -1;
+ }
+ if (zleak_state & ZLEAK_STATE_ACTIVE) {
+ return 1;
+ }
+ return 0;
}
-#endif
-
-
kern_return_t
zleak_activate(void)
{
ztraces = traces_ptr;
/*
- * Initialize the top_ztrace to the first entry in ztraces,
+ * Initialize the top_ztrace to the first entry in ztraces,
* so we don't have to check for null in zleak_log
*/
top_ztrace = &ztraces[0];
zleak_state |= ZLEAK_STATE_ACTIVE;
zleak_state &= ~ZLEAK_STATE_ACTIVATING;
lck_spin_unlock(&zleak_lock);
-
+
return 0;
-fail:
+fail:
/*
* If we fail to allocate memory, don't further tax
* the system by trying again.
}
/*
- * TODO: What about allocations that never get deallocated,
+ * TODO: What about allocations that never get deallocated,
* especially ones with unique backtraces? Should we wait to record
- * until after boot has completed?
+ * until after boot has completed?
* (How many persistent zallocs are there?)
*/
/*
- * This function records the allocation in the allocations table,
- * and stores the associated backtrace in the traces table
+ * This function records the allocation in the allocations table,
+ * and stores the associated backtrace in the traces table
* (or just increments the refcount if the trace is already recorded)
* If the allocation slot is in use, the old allocation is replaced with the new allocation, and
* the associated trace's refcount is decremented.
*/
static boolean_t
zleak_log(uintptr_t* bt,
- uintptr_t addr,
- uint32_t depth,
- vm_size_t allocation_size)
+ uintptr_t addr,
+ uint32_t depth,
+ vm_size_t allocation_size)
{
/* Quit if there's someone else modifying the hash tables */
if (!lck_spin_try_lock(&zleak_lock)) {
z_total_conflicts++;
return FALSE;
}
-
- struct zallocation* allocation = &zallocations[hashaddr(addr, zleak_alloc_buckets)];
-
+
+ struct zallocation* allocation = &zallocations[hashaddr(addr, zleak_alloc_buckets)];
+
uint32_t trace_index = hashbacktrace(bt, depth, zleak_trace_buckets);
struct ztrace* trace = &ztraces[trace_index];
-
+
allocation->za_hit_count++;
trace->zt_hit_count++;
-
- /*
+
+ /*
* If the allocation bucket we want to be in is occupied, and if the occupier
- * has the same trace as us, just bail.
+ * has the same trace as us, just bail.
*/
if (allocation->za_element != (uintptr_t) 0 && trace_index == allocation->za_trace_index) {
z_alloc_collisions++;
-
+
lck_spin_unlock(&zleak_lock);
return TRUE;
}
-
+
/* STEP 1: Store the backtrace in the traces array. */
/* A size of zero indicates that the trace bucket is free. */
-
- if (trace->zt_size > 0 && bcmp(trace->zt_stack, bt, (depth * sizeof(uintptr_t))) != 0 ) {
- /*
+
+ if (trace->zt_size > 0 && bcmp(trace->zt_stack, bt, (depth * sizeof(uintptr_t))) != 0) {
+ /*
* Different unique trace with same hash!
* Just bail - if we're trying to record the leaker, hopefully the other trace will be deallocated
* and get out of the way for later chances
*/
trace->zt_collisions++;
z_trace_collisions++;
-
+
lck_spin_unlock(&zleak_lock);
return TRUE;
} else if (trace->zt_size > 0) {
trace->zt_size += allocation_size;
} else {
/* Found an unused trace bucket, record the trace here! */
- if (trace->zt_depth != 0) /* if this slot was previously used but not currently in use */
+ if (trace->zt_depth != 0) { /* if this slot was previously used but not currently in use */
z_trace_overwrites++;
-
+ }
+
z_trace_recorded++;
- trace->zt_size = allocation_size;
- memcpy(trace->zt_stack, bt, (depth * sizeof(uintptr_t)) );
-
- trace->zt_depth = depth;
- trace->zt_collisions = 0;
+ trace->zt_size = allocation_size;
+ memcpy(trace->zt_stack, bt, (depth * sizeof(uintptr_t)));
+
+ trace->zt_depth = depth;
+ trace->zt_collisions = 0;
}
-
+
/* STEP 2: Store the allocation record in the allocations array. */
-
+
if (allocation->za_element != (uintptr_t) 0) {
- /*
+ /*
* Straight up replace any allocation record that was there. We don't want to do the work
- * to preserve the allocation entries that were there, because we only record a subset of the
+ * to preserve the allocation entries that were there, because we only record a subset of the
* allocations anyways.
*/
-
+
z_alloc_collisions++;
-
+
struct ztrace* associated_trace = &ztraces[allocation->za_trace_index];
/* Knock off old allocation's size, not the new allocation */
associated_trace->zt_size -= allocation->za_size;
z_alloc_overwrites++;
}
- allocation->za_element = addr;
- allocation->za_trace_index = trace_index;
- allocation->za_size = allocation_size;
-
+ allocation->za_element = addr;
+ allocation->za_trace_index = trace_index;
+ allocation->za_size = allocation_size;
+
z_alloc_recorded++;
-
- if (top_ztrace->zt_size < trace->zt_size)
+
+ if (top_ztrace->zt_size < trace->zt_size) {
top_ztrace = trace;
-
+ }
+
lck_spin_unlock(&zleak_lock);
return TRUE;
}
* Free the allocation record and release the stacktrace.
* This should be as fast as possible because it will be called for every free.
*/
+__attribute__((noinline))
static void
zleak_free(uintptr_t addr,
- vm_size_t allocation_size)
+ vm_size_t allocation_size)
{
- if (addr == (uintptr_t) 0)
+ if (addr == (uintptr_t) 0) {
return;
-
+ }
+
struct zallocation* allocation = &zallocations[hashaddr(addr, zleak_alloc_buckets)];
-
+
/* Double-checked locking: check to find out if we're interested, lock, check to make
* sure it hasn't changed, then modify it, and release the lock.
*/
-
+
if (allocation->za_element == addr && allocation->za_trace_index < zleak_trace_buckets) {
/* if the allocation was the one, grab the lock, check again, then delete it */
lck_spin_lock(&zleak_lock);
-
+
if (allocation->za_element == addr && allocation->za_trace_index < zleak_trace_buckets) {
struct ztrace *trace;
/* allocation_size had better match what was passed into zleak_log - otherwise someone is freeing into the wrong zone! */
if (allocation->za_size != allocation_size) {
- panic("Freeing as size %lu memory that was allocated with size %lu\n",
- (uintptr_t)allocation_size, (uintptr_t)allocation->za_size);
+ panic("Freeing as size %lu memory that was allocated with size %lu\n",
+ (uintptr_t)allocation_size, (uintptr_t)allocation->za_size);
}
-
+
trace = &ztraces[allocation->za_trace_index];
-
+
/* size of 0 indicates trace bucket is unused */
if (trace->zt_size > 0) {
trace->zt_size -= allocation_size;
}
-
- /* A NULL element means the allocation bucket is unused */
- allocation->za_element = 0;
+
+ /* A NULL element means the allocation bucket is unused */
+ allocation->za_element = 0;
+ }
+ lck_spin_unlock(&zleak_lock);
+ }
+}
+
+#endif /* CONFIG_ZLEAKS */
+
+/* These functions outside of CONFIG_ZLEAKS because they are also used in
+ * mbuf.c for mbuf leak-detection. This is why they lack the z_ prefix.
+ */
+
+/* "Thomas Wang's 32/64 bit mix functions." http://www.concentric.net/~Ttwang/tech/inthash.htm */
+uintptr_t
+hash_mix(uintptr_t x)
+{
+#ifndef __LP64__
+ x += ~(x << 15);
+ x ^= (x >> 10);
+ x += (x << 3);
+ x ^= (x >> 6);
+ x += ~(x << 11);
+ x ^= (x >> 16);
+#else
+ x += ~(x << 32);
+ x ^= (x >> 22);
+ x += ~(x << 13);
+ x ^= (x >> 8);
+ x += (x << 3);
+ x ^= (x >> 15);
+ x += ~(x << 27);
+ x ^= (x >> 31);
+#endif
+ return x;
+}
+
+uint32_t
+hashbacktrace(uintptr_t* bt, uint32_t depth, uint32_t max_size)
+{
+ uintptr_t hash = 0;
+ uintptr_t mask = max_size - 1;
+
+ while (depth) {
+ hash += bt[--depth];
+ }
+
+ hash = hash_mix(hash) & mask;
+
+ assert(hash < max_size);
+
+ return (uint32_t) hash;
+}
+
+/*
+ * TODO: Determine how well distributed this is
+ * max_size must be a power of 2. i.e 0x10000 because 0x10000-1 is 0x0FFFF which is a great bitmask
+ */
+uint32_t
+hashaddr(uintptr_t pt, uint32_t max_size)
+{
+ uintptr_t hash = 0;
+ uintptr_t mask = max_size - 1;
+
+ hash = hash_mix(pt) & mask;
+
+ assert(hash < max_size);
+
+ return (uint32_t) hash;
+}
+
+/* End of all leak-detection code */
+#pragma mark zone creation, configuration, destruction
+
+static zone_t
+zone_init_defaults(zone_id_t zid)
+{
+ zone_t z = &zone_array[zid];
+
+ z->page_count_max = ~0u;
+ z->collectable = true;
+ z->expandable = true;
+ z->submap_idx = Z_SUBMAP_IDX_GENERAL_MAP;
+
+ simple_lock_init(&z->lock, 0);
+
+ return z;
+}
+
+static bool
+zone_is_initializing(zone_t z)
+{
+ return !z->z_self && !z->destroyed;
+}
+
+static void
+zone_set_max(zone_t z, vm_size_t max)
+{
+#if KASAN_ZALLOC
+ if (z->kasan_redzone) {
+ /*
+ * Adjust the max memory for the kasan redzones
+ */
+ max += (max / z->pcpu_elem_size) * z->kasan_redzone * 2;
+ }
+#endif
+ if (max < z->percpu ? 1 : z->alloc_pages) {
+ max = z->percpu ? 1 : z->alloc_pages;
+ } else {
+ max = atop(round_page(max));
+ }
+ z->page_count_max = max;
+}
+
+void
+zone_set_submap_idx(zone_t zone, unsigned int sub_map_idx)
+{
+ if (!zone_is_initializing(zone)) {
+ panic("%s: called after zone_create()", __func__);
+ }
+ if (sub_map_idx > zone_last_submap_idx) {
+ panic("zone_set_submap_idx(%d) > %d", sub_map_idx, zone_last_submap_idx);
+ }
+ zone->submap_idx = sub_map_idx;
+}
+
+void
+zone_set_noexpand(
+ zone_t zone,
+ vm_size_t max)
+{
+ if (!zone_is_initializing(zone)) {
+ panic("%s: called after zone_create()", __func__);
+ }
+ zone->expandable = false;
+ zone_set_max(zone, max);
+}
+
+void
+zone_set_exhaustible(
+ zone_t zone,
+ vm_size_t max)
+{
+ if (!zone_is_initializing(zone)) {
+ panic("%s: called after zone_create()", __func__);
+ }
+ zone->expandable = false;
+ zone->exhaustible = true;
+ zone_set_max(zone, max);
+}
+
+/**
+ * @function zone_create_find
+ *
+ * @abstract
+ * Finds an unused zone for the given name and element size.
+ *
+ * @param name the zone name
+ * @param size the element size (including redzones, ...)
+ * @param flags the flags passed to @c zone_create*
+ * @param zid the desired zone ID or ZONE_ID_ANY
+ *
+ * @returns a zone to initialize further.
+ */
+static zone_t
+zone_create_find(
+ const char *name,
+ vm_size_t size,
+ zone_create_flags_t flags,
+ zone_id_t zid)
+{
+ zone_id_t nzones;
+ zone_t z;
+
+ simple_lock(&all_zones_lock, &zone_locks_grp);
+
+ nzones = (zone_id_t)os_atomic_load(&num_zones, relaxed);
+ assert(num_zones_in_use <= nzones && nzones < MAX_ZONES);
+
+ if (__improbable(nzones < ZONE_ID__FIRST_DYNAMIC)) {
+ /*
+ * The first time around, make sure the reserved zone IDs
+ * have an initialized lock as zone_index_foreach() will
+ * enumerate them.
+ */
+ while (nzones < ZONE_ID__FIRST_DYNAMIC) {
+ zone_init_defaults(nzones++);
+ }
+
+ os_atomic_store(&num_zones, nzones, release);
+ }
+
+ if (zid != ZONE_ID_ANY) {
+ if (zid >= ZONE_ID__FIRST_DYNAMIC) {
+ panic("zone_create: invalid desired zone ID %d for %s",
+ zid, name);
+ }
+ if (flags & ZC_DESTRUCTIBLE) {
+ panic("zone_create: ID %d (%s) must be permanent", zid, name);
+ }
+ if (zone_array[zid].z_self) {
+ panic("zone_create: creating zone ID %d (%s) twice", zid, name);
+ }
+ z = &zone_array[zid];
+ } else {
+ if (flags & ZC_DESTRUCTIBLE) {
+ /*
+ * If possible, find a previously zdestroy'ed zone in the
+ * zone_array that we can reuse.
+ */
+ for (int i = bitmap_first(zone_destroyed_bitmap, MAX_ZONES);
+ i >= 0; i = bitmap_next(zone_destroyed_bitmap, i)) {
+ z = &zone_array[i];
+
+ /*
+ * If the zone name and the element size are the
+ * same, we can just reuse the old zone struct.
+ */
+ if (strcmp(z->z_name, name) || zone_elem_size(z) != size) {
+ continue;
+ }
+ bitmap_clear(zone_destroyed_bitmap, i);
+ z->destroyed = false;
+ z->z_self = z;
+ zid = (zone_id_t)i;
+ goto out;
+ }
+ }
+
+ zid = nzones++;
+ z = zone_init_defaults(zid);
+
+ /*
+ * The release barrier pairs with the acquire in
+ * zone_index_foreach() and makes sure that enumeration loops
+ * always see an initialized zone lock.
+ */
+ os_atomic_store(&num_zones, nzones, release);
+ }
+
+out:
+ num_zones_in_use++;
+ simple_unlock(&all_zones_lock);
+
+ return z;
+}
+
+__abortlike
+static void
+zone_create_panic(const char *name, const char *f1, const char *f2)
+{
+ panic("zone_create: creating zone %s: flag %s and %s are incompatible",
+ name, f1, f2);
+}
+#define zone_create_assert_not_both(name, flags, current_flag, forbidden_flag) \
+ if ((flags) & forbidden_flag) { \
+ zone_create_panic(name, #current_flag, #forbidden_flag); \
+ }
+
+/*
+ * Adjusts the size of the element based on minimum size, alignment
+ * and kasan redzones
+ */
+static vm_size_t
+zone_elem_adjust_size(
+ const char *name __unused,
+ vm_size_t elem_size,
+ zone_create_flags_t flags,
+ vm_size_t *redzone __unused)
+{
+ vm_size_t size;
+ /*
+ * Adjust element size for minimum size and pointer alignment
+ */
+ size = (elem_size + sizeof(vm_offset_t) - 1) & -sizeof(vm_offset_t);
+ if (((flags & ZC_PERCPU) == 0) && size < ZONE_MIN_ELEM_SIZE) {
+ size = ZONE_MIN_ELEM_SIZE;
+ }
+
+#if KASAN_ZALLOC
+ /*
+ * Expand the zone allocation size to include the redzones.
+ *
+ * For page-multiple zones add a full guard page because they
+ * likely require alignment.
+ */
+ vm_size_t redzone_tmp;
+ if (flags & (ZC_KASAN_NOREDZONE | ZC_PERCPU)) {
+ redzone_tmp = 0;
+ } else if ((size & PAGE_MASK) == 0) {
+ if (size != PAGE_SIZE && (flags & ZC_ALIGNMENT_REQUIRED)) {
+ panic("zone_create: zone %s can't provide more than PAGE_SIZE"
+ "alignment", name);
+ }
+ redzone_tmp = PAGE_SIZE;
+ } else if (flags & ZC_ALIGNMENT_REQUIRED) {
+ redzone_tmp = 0;
+ } else {
+ redzone_tmp = KASAN_GUARD_SIZE;
+ }
+ size += redzone_tmp * 2;
+ if (redzone) {
+ *redzone = redzone_tmp;
+ }
+#endif
+ return size;
+}
+
+/*
+ * Returns the allocation chunk size that has least framentation
+ */
+static vm_size_t
+zone_get_min_alloc_granule(
+ vm_size_t elem_size,
+ zone_create_flags_t flags)
+{
+ vm_size_t alloc_granule = PAGE_SIZE;
+ if (flags & ZC_PERCPU) {
+ alloc_granule = PAGE_SIZE * zpercpu_count();
+ if (PAGE_SIZE % elem_size > 256) {
+ panic("zone_create: per-cpu zone has too much fragmentation");
+ }
+ } else if ((elem_size & PAGE_MASK) == 0) {
+ /* zero fragmentation by definition */
+ alloc_granule = elem_size;
+ } else if (alloc_granule % elem_size == 0) {
+ /* zero fragmentation by definition */
+ } else {
+ vm_size_t frag = (alloc_granule % elem_size) * 100 / alloc_granule;
+ vm_size_t alloc_tmp = PAGE_SIZE;
+ while ((alloc_tmp += PAGE_SIZE) <= ZONE_MAX_ALLOC_SIZE) {
+ vm_size_t frag_tmp = (alloc_tmp % elem_size) * 100 / alloc_tmp;
+ if (frag_tmp < frag) {
+ frag = frag_tmp;
+ alloc_granule = alloc_tmp;
+ }
}
- lck_spin_unlock(&zleak_lock);
}
+ return alloc_granule;
}
-#endif /* CONFIG_ZLEAKS */
+vm_size_t
+zone_get_foreign_alloc_size(
+ const char *name __unused,
+ vm_size_t elem_size,
+ zone_create_flags_t flags,
+ uint16_t min_pages)
+{
+ vm_size_t adjusted_size = zone_elem_adjust_size(name, elem_size, flags,
+ NULL);
+ vm_size_t alloc_granule = zone_get_min_alloc_granule(adjusted_size,
+ flags);
+ vm_size_t min_size = min_pages * PAGE_SIZE;
+ /*
+ * Round up min_size to a multiple of alloc_granule
+ */
+ return ((min_size + alloc_granule - 1) / alloc_granule)
+ * alloc_granule;
+}
-/* These functions outside of CONFIG_ZLEAKS because they are also used in
- * mbuf.c for mbuf leak-detection. This is why they lack the z_ prefix.
- */
+zone_t
+zone_create_ext(
+ const char *name,
+ vm_size_t size,
+ zone_create_flags_t flags,
+ zone_id_t desired_zid,
+ void (^extra_setup)(zone_t))
+{
+ vm_size_t alloc;
+ vm_size_t redzone;
+ zone_t z;
-/*
- * This function captures a backtrace from the current stack and
- * returns the number of frames captured, limited by max_frames.
- * It's fast because it does no checking to make sure there isn't bad data.
- * Since it's only called from threads that we're going to keep executing,
- * if there's bad data we were going to die eventually.
- * If this function is inlined, it doesn't record the frame of the function it's inside.
- * (because there's no stack frame!)
- */
+ if (size > ZONE_MAX_ALLOC_SIZE) {
+ panic("zone_create: element size too large: %zd", (size_t)size);
+ }
-uint32_t
-fastbacktrace(uintptr_t* bt, uint32_t max_frames)
-{
- uintptr_t* frameptr = NULL, *frameptr_next = NULL;
- uintptr_t retaddr = 0;
- uint32_t frame_index = 0, frames = 0;
- uintptr_t kstackb, kstackt;
- thread_t cthread = current_thread();
+ size = zone_elem_adjust_size(name, size, flags, &redzone);
+ /*
+ * Allocate the zone slot, return early if we found an older match.
+ */
+ z = zone_create_find(name, size, flags, desired_zid);
+ if (__improbable(z->z_self)) {
+ /* We found a zone to reuse */
+ return z;
+ }
- if (__improbable(cthread == NULL))
- return 0;
+ /*
+ * Initialize the zone properly.
+ */
- kstackb = cthread->kernel_stack;
- kstackt = kstackb + kernel_stack_size;
- /* Load stack frame pointer (EBP on x86) into frameptr */
- frameptr = __builtin_frame_address(0);
- if (((uintptr_t)frameptr > kstackt) || ((uintptr_t)frameptr < kstackb))
- frameptr = NULL;
+ /*
+ * If the kernel is post lockdown, copy the zone name passed in.
+ * Else simply maintain a pointer to the name string as it can only
+ * be a core XNU zone (no unloadable kext exists before lockdown).
+ */
+ if (startup_phase >= STARTUP_SUB_LOCKDOWN) {
+ size_t nsz = MIN(strlen(name) + 1, MACH_ZONE_NAME_MAX_LEN);
+ char *buf = zalloc_permanent(nsz, ZALIGN_NONE);
+ strlcpy(buf, name, nsz);
+ z->z_name = buf;
+ } else {
+ z->z_name = name;
+ }
+ /*
+ * If zone_init() hasn't run yet, the permanent zones do not exist.
+ * We can limp along without properly initialized stats for a while,
+ * zone_init() will rebuild the missing stats when it runs.
+ */
+ if (__probable(zone_array[ZONE_ID_PERCPU_PERMANENT].z_self)) {
+ z->z_stats = zalloc_percpu_permanent_type(struct zone_stats);
+ }
- while (frameptr != NULL && frame_index < max_frames ) {
- /* Next frame pointer is pointed to by the previous one */
- frameptr_next = (uintptr_t*) *frameptr;
+ alloc = zone_get_min_alloc_granule(size, flags);
- /* Bail if we see a zero in the stack frame, that means we've reached the top of the stack */
- /* That also means the return address is worthless, so don't record it */
- if (frameptr_next == NULL)
- break;
- /* Verify thread stack bounds */
- if (((uintptr_t)frameptr_next > kstackt) || ((uintptr_t)frameptr_next < kstackb))
- break;
- /* Pull return address from one spot above the frame pointer */
- retaddr = *(frameptr + 1);
+ if (flags & ZC_KALLOC_HEAP) {
+ size_t rem = (alloc % size) / (alloc / size);
- /* Store it in the backtrace array */
- bt[frame_index++] = retaddr;
+ /*
+ * Try to grow the elements size and spread them more if the remaining
+ * space is large enough.
+ */
+ size += rem & ~(KALLOC_MINALIGN - 1);
+ }
- frameptr = frameptr_next;
+ z->pcpu_elem_size = z->z_elem_size = (uint16_t)size;
+ z->alloc_pages = (uint16_t)atop(alloc);
+#if KASAN_ZALLOC
+ z->kasan_redzone = redzone;
+ if (strncmp(name, "fakestack.", sizeof("fakestack.") - 1) == 0) {
+ z->kasan_fakestacks = true;
}
+#endif
- /* Save the number of frames captured for return value */
- frames = frame_index;
+ /*
+ * Handle KPI flags
+ */
+#if __LP64__
+ if (flags & ZC_SEQUESTER) {
+ z->va_sequester = true;
+ }
+#endif
+ /* ZC_CACHING applied after all configuration is done */
- /* Fill in the rest of the backtrace with zeros */
- while (frame_index < max_frames)
- bt[frame_index++] = 0;
+ if (flags & ZC_PERCPU) {
+ /*
+ * ZC_CACHING is disallowed because it uses per-cpu zones for its
+ * implementation and it would be circular. These allocations are
+ * also quite expensive, so caching feels dangerous memory wise too.
+ *
+ * ZC_ZFREE_CLEARMEM is forced because per-cpu zones allow for
+ * pointer-sized allocations which poisoning doesn't support.
+ */
+ zone_create_assert_not_both(name, flags, ZC_PERCPU, ZC_CACHING);
+ zone_create_assert_not_both(name, flags, ZC_PERCPU, ZC_ALLOW_FOREIGN);
+ z->percpu = true;
+ z->gzalloc_exempt = true;
+ z->zfree_clear_mem = true;
+ z->pcpu_elem_size *= zpercpu_count();
+ }
+ if (flags & ZC_ZFREE_CLEARMEM) {
+ z->zfree_clear_mem = true;
+ }
+ if (flags & ZC_NOGC) {
+ z->collectable = false;
+ }
+ if (flags & ZC_NOENCRYPT) {
+ z->noencrypt = true;
+ }
+ if (flags & ZC_ALIGNMENT_REQUIRED) {
+ z->alignment_required = true;
+ }
+ if (flags & ZC_NOGZALLOC) {
+ z->gzalloc_exempt = true;
+ }
+ if (flags & ZC_NOCALLOUT) {
+ z->no_callout = true;
+ }
+ if (flags & ZC_DESTRUCTIBLE) {
+ zone_create_assert_not_both(name, flags, ZC_DESTRUCTIBLE, ZC_CACHING);
+ zone_create_assert_not_both(name, flags, ZC_DESTRUCTIBLE, ZC_ALLOW_FOREIGN);
+ z->destructible = true;
+ }
- return frames;
-}
+ /*
+ * Handle Internal flags
+ */
+ if (flags & ZC_ALLOW_FOREIGN) {
+ z->allows_foreign = true;
+ }
+ if ((ZSECURITY_OPTIONS_SUBMAP_USER_DATA & zsecurity_options) &&
+ (flags & ZC_DATA_BUFFERS)) {
+ z->submap_idx = Z_SUBMAP_IDX_BAG_OF_BYTES_MAP;
+ }
+ if (flags & ZC_KASAN_NOQUARANTINE) {
+ z->kasan_noquarantine = true;
+ }
+ /* ZC_KASAN_NOREDZONE already handled */
-/* "Thomas Wang's 32/64 bit mix functions." http://www.concentric.net/~Ttwang/tech/inthash.htm */
-uintptr_t
-hash_mix(uintptr_t x)
-{
-#ifndef __LP64__
- x += ~(x << 15);
- x ^= (x >> 10);
- x += (x << 3 );
- x ^= (x >> 6 );
- x += ~(x << 11);
- x ^= (x >> 16);
-#else
- x += ~(x << 32);
- x ^= (x >> 22);
- x += ~(x << 13);
- x ^= (x >> 8 );
- x += (x << 3 );
- x ^= (x >> 15);
- x += ~(x << 27);
- x ^= (x >> 31);
-#endif
- return x;
-}
+ /*
+ * Then if there's extra tuning, do it
+ */
+ if (extra_setup) {
+ extra_setup(z);
+ }
-uint32_t
-hashbacktrace(uintptr_t* bt, uint32_t depth, uint32_t max_size)
-{
+ /*
+ * Configure debugging features
+ */
+#if CONFIG_GZALLOC
+ gzalloc_zone_init(z); /* might set z->gzalloc_tracked */
+#endif
+#if ZONE_ENABLE_LOGGING
+ if (!z->gzalloc_tracked && num_zones_logged < max_num_zones_to_log) {
+ /*
+ * Check for and set up zone leak detection if requested via boot-args.
+ * might set z->zone_logging
+ */
+ zone_setup_logging(z);
+ }
+#endif /* ZONE_ENABLE_LOGGING */
+#if VM_MAX_TAG_ZONES
+ if (!z->gzalloc_tracked && z->kalloc_heap && zone_tagging_on) {
+ static int tag_zone_index;
+ vm_offset_t esize = zone_elem_size(z);
+ z->tags = true;
+ z->tags_inline = (((page_size + esize - 1) / esize) <=
+ (sizeof(uint32_t) / sizeof(uint16_t)));
+ z->tag_zone_index = os_atomic_inc_orig(&tag_zone_index, relaxed);
+ assert(z->tag_zone_index < VM_MAX_TAG_ZONES);
+ }
+#endif
- uintptr_t hash = 0;
- uintptr_t mask = max_size - 1;
+ /*
+ * Finally, fixup properties based on security policies, boot-args, ...
+ */
+ if ((ZSECURITY_OPTIONS_SUBMAP_USER_DATA & zsecurity_options) &&
+ z->kalloc_heap == KHEAP_ID_DATA_BUFFERS) {
+ z->submap_idx = Z_SUBMAP_IDX_BAG_OF_BYTES_MAP;
+ }
+#if __LP64__
+ if ((ZSECURITY_OPTIONS_SEQUESTER & zsecurity_options) &&
+ (flags & ZC_NOSEQUESTER) == 0 &&
+ z->submap_idx == Z_SUBMAP_IDX_GENERAL_MAP) {
+ z->va_sequester = true;
+ }
+#endif
+ /*
+ * Always clear zone elements smaller than a cacheline,
+ * because it's pretty close to free.
+ */
+ if (size <= zp_min_size) {
+ z->zfree_clear_mem = true;
+ }
+ if (zp_factor != 0 && !z->zfree_clear_mem) {
+ z->zp_count = zone_poison_count_init(z);
+ }
- while (depth) {
- hash += bt[--depth];
+#if CONFIG_ZCACHE
+ if ((flags & ZC_NOCACHING) == 0) {
+ /*
+ * Append kalloc heap name to zone name (if zone is used by kalloc)
+ */
+ char temp_zone_name[MAX_ZONE_NAME] = "";
+ snprintf(temp_zone_name, MAX_ZONE_NAME, "%s%s", zone_heap_name(z), z->z_name);
+
+ /* Check if boot-arg specified it should have a cache */
+ if (track_this_zone(temp_zone_name, cache_zone_name)) {
+ flags |= ZC_CACHING;
+ } else if (zcc_kalloc && z->kalloc_heap) {
+ flags |= ZC_CACHING;
+ }
+ }
+ if ((flags & ZC_CACHING) &&
+ !z->tags && !z->zone_logging && !z->gzalloc_tracked) {
+ zcache_init(z);
}
+#endif /* CONFIG_ZCACHE */
- hash = hash_mix(hash) & mask;
+ lock_zone(z);
+ z->z_self = z;
+ unlock_zone(z);
- assert(hash < max_size);
+ return z;
+}
- return (uint32_t) hash;
+__startup_func
+void
+zone_create_startup(struct zone_create_startup_spec *spec)
+{
+ *spec->z_var = zone_create_ext(spec->z_name, spec->z_size,
+ spec->z_flags, spec->z_zid, spec->z_setup);
}
/*
- * TODO: Determine how well distributed this is
- * max_size must be a power of 2. i.e 0x10000 because 0x10000-1 is 0x0FFFF which is a great bitmask
+ * The 4 first field of a zone_view and a zone alias, so that the zone_or_view_t
+ * union works. trust but verify.
*/
-uint32_t
-hashaddr(uintptr_t pt, uint32_t max_size)
+#define zalloc_check_zov_alias(f1, f2) \
+ static_assert(offsetof(struct zone, f1) == offsetof(struct zone_view, f2))
+zalloc_check_zov_alias(z_self, zv_zone);
+zalloc_check_zov_alias(z_stats, zv_stats);
+zalloc_check_zov_alias(z_name, zv_name);
+zalloc_check_zov_alias(z_views, zv_next);
+#undef zalloc_check_zov_alias
+
+__startup_func
+void
+zone_view_startup_init(struct zone_view_startup_spec *spec)
{
- uintptr_t hash = 0;
- uintptr_t mask = max_size - 1;
-
- hash = hash_mix(pt) & mask;
+ struct kalloc_heap *heap = NULL;
+ zone_view_t zv = spec->zv_view;
+ zone_t z;
+
+ switch (spec->zv_heapid) {
+ case KHEAP_ID_DEFAULT:
+ heap = KHEAP_DEFAULT;
+ break;
+ case KHEAP_ID_DATA_BUFFERS:
+ heap = KHEAP_DATA_BUFFERS;
+ break;
+ case KHEAP_ID_KEXT:
+ heap = KHEAP_KEXT;
+ break;
+ default:
+ heap = NULL;
+ }
- assert(hash < max_size);
+ if (heap) {
+ z = kalloc_heap_zone_for_size(heap, spec->zv_size);
+ assert(z);
+ } else {
+ z = spec->zv_zone;
+ assert(spec->zv_size <= zone_elem_size(z));
+ }
- return (uint32_t) hash;
+ zv->zv_zone = z;
+ zv->zv_stats = zalloc_percpu_permanent_type(struct zone_stats);
+ zv->zv_next = z->z_views;
+ if (z->z_views == NULL && z->kalloc_heap == KHEAP_ID_NONE) {
+ /*
+ * count the raw view for zones not in a heap,
+ * kalloc_heap_init() already counts it for its members.
+ */
+ zone_view_count += 2;
+ } else {
+ zone_view_count += 1;
+ }
+ z->z_views = zv;
}
-/* End of all leak-detection code */
-#pragma mark -
+zone_t
+zone_create(
+ const char *name,
+ vm_size_t size,
+ zone_create_flags_t flags)
+{
+ return zone_create_ext(name, size, flags, ZONE_ID_ANY, NULL);
+}
-/*
- * zinit initializes a new zone. The zone data structures themselves
- * are stored in a zone, which is initially a static structure that
- * is initialized by zone_init.
- */
zone_t
zinit(
- vm_size_t size, /* the size of an element */
- vm_size_t max, /* maximum memory to use */
- vm_size_t alloc, /* allocation size */
- const char *name) /* a name for the zone */
+ vm_size_t size, /* the size of an element */
+ vm_size_t max, /* maximum memory to use */
+ vm_size_t alloc __unused, /* allocation size */
+ const char *name) /* a name for the zone */
{
- zone_t z;
- boolean_t use_page_list = FALSE;
+ zone_t z = zone_create(name, size, ZC_DESTRUCTIBLE);
+ zone_set_max(z, max);
+ return z;
+}
- if (zone_zone == ZONE_NULL) {
+void
+zdestroy(zone_t z)
+{
+ unsigned int zindex = zone_index(z);
- z = (struct zone *)zdata;
- /* special handling in zcram() because the first element is being used */
- } else
- z = (zone_t) zalloc(zone_zone);
+ lock_zone(z);
- if (z == ZONE_NULL)
- return(ZONE_NULL);
+ if (!z->destructible || zone_caching_enabled(z) || z->allows_foreign) {
+ panic("zdestroy: Zone %s%s isn't destructible",
+ zone_heap_name(z), z->z_name);
+ }
- /* Zone elements must fit both a next pointer and a backup pointer */
- vm_size_t minimum_element_size = sizeof(vm_offset_t) * 2;
- if (size < minimum_element_size)
- size = minimum_element_size;
+ if (!z->z_self || z->expanding_no_vm_priv || z->expanding_vm_priv ||
+ z->async_pending || z->waiting) {
+ panic("zdestroy: Zone %s%s in an invalid state for destruction",
+ zone_heap_name(z), z->z_name);
+ }
+#if !KASAN_ZALLOC
/*
- * Round element size to a multiple of sizeof(pointer)
- * This also enforces that allocations will be aligned on pointer boundaries
+ * Unset the valid bit. We'll hit an assert failure on further operations
+ * on this zone, until zinit() is called again.
+ *
+ * Leave the zone valid for KASan as we will see zfree's on quarantined free
+ * elements even after the zone is destroyed.
*/
- size = ((size-1) + sizeof(vm_offset_t)) -
- ((size-1) % sizeof(vm_offset_t));
-
- if (alloc == 0)
- alloc = PAGE_SIZE;
+ z->z_self = NULL;
+#endif
+ z->destroyed = true;
+ unlock_zone(z);
- alloc = round_page(alloc);
- max = round_page(max);
+ /* Dump all the free elements */
+ zone_drop_free_elements(z);
- /*
- * we look for an allocation size with less than 1% waste
- * up to 5 pages in size...
- * otherwise, we look for an allocation size with least fragmentation
- * in the range of 1 - 5 pages
- * This size will be used unless
- * the user suggestion is larger AND has less fragmentation
- */
-#if ZONE_ALIAS_ADDR
- /* Favor PAGE_SIZE allocations unless we waste >10% space */
- if ((size < PAGE_SIZE) && (PAGE_SIZE % size <= PAGE_SIZE / 10))
- alloc = PAGE_SIZE;
- else
+#if CONFIG_GZALLOC
+ if (__improbable(z->gzalloc_tracked)) {
+ /* If the zone is gzalloc managed dump all the elements in the free cache */
+ gzalloc_empty_free_cache(z);
+ }
#endif
-#if defined(__LP64__)
- if (((alloc % size) != 0) || (alloc > PAGE_SIZE * 8))
+
+ lock_zone(z);
+
+ while (!zone_pva_is_null(z->pages_sequester)) {
+ struct zone_page_metadata *page_meta;
+ vm_offset_t free_addr;
+
+ page_meta = zone_sequestered_page_get(z, &free_addr);
+ unlock_zone(z);
+ kmem_free(submap_for_zone(z), free_addr, ptoa(z->alloc_pages));
+ lock_zone(z);
+ }
+
+#if !KASAN_ZALLOC
+ /* Assert that all counts are zero */
+ if (z->countavail || z->countfree || zone_size_wired(z) ||
+ z->allfree_page_count || z->sequester_page_count) {
+ panic("zdestroy: Zone %s%s isn't empty at zdestroy() time",
+ zone_heap_name(z), z->z_name);
+ }
+
+ /* consistency check: make sure everything is indeed empty */
+ assert(zone_pva_is_null(z->pages_any_free_foreign));
+ assert(zone_pva_is_null(z->pages_all_used_foreign));
+ assert(zone_pva_is_null(z->pages_all_free));
+ assert(zone_pva_is_null(z->pages_intermediate));
+ assert(zone_pva_is_null(z->pages_all_used));
+ assert(zone_pva_is_null(z->pages_sequester));
#endif
- {
- vm_size_t best, waste; unsigned int i;
- best = PAGE_SIZE;
- waste = best % size;
- for (i = 1; i <= 5; i++) {
- vm_size_t tsize, twaste;
+ unlock_zone(z);
- tsize = i * PAGE_SIZE;
+ simple_lock(&all_zones_lock, &zone_locks_grp);
- if ((tsize % size) < (tsize / 100)) {
- alloc = tsize;
- goto use_this_allocation;
- }
- twaste = tsize % size;
- if (twaste < waste)
- best = tsize, waste = twaste;
- }
- if (alloc <= best || (alloc % size >= waste))
- alloc = best;
+ assert(!bitmap_test(zone_destroyed_bitmap, zindex));
+ /* Mark the zone as empty in the bitmap */
+ bitmap_set(zone_destroyed_bitmap, zindex);
+ num_zones_in_use--;
+ assert(num_zones_in_use > 0);
+
+ simple_unlock(&all_zones_lock);
+}
+
+#pragma mark zone (re)fill, jetsam
+
+/*
+ * Dealing with zone allocations from the mach VM code.
+ *
+ * The implementation of the mach VM itself uses the zone allocator
+ * for things like the vm_map_entry data structure. In order to prevent
+ * an infinite recursion problem when adding more pages to a zone, zalloc
+ * uses a replenish thread to refill the VM layer's zones before they have
+ * too few remaining free entries. The reserved remaining free entries
+ * guarantee that the VM routines can get entries from already mapped pages.
+ *
+ * In order for that to work, the amount of allocations in the nested
+ * case have to be bounded. There are currently 2 replenish zones, and
+ * if each needs 1 element of each zone to add a new page to itself, that
+ * gives us a minumum reserve of 2 elements.
+ *
+ * There is also a deadlock issue with the zone garbage collection thread,
+ * or any thread that is trying to free zone pages. While holding
+ * the kernel's map lock they may need to allocate new VM map entries, hence
+ * we need enough reserve to allow them to get past the point of holding the
+ * map lock. After freeing that page, the GC thread will wait in drop_free_elements()
+ * until the replenish threads can finish. Since there's only 1 GC thread at a time,
+ * that adds a minimum of 1 to the reserve size.
+ *
+ * Since the minumum amount you can add to a zone is 1 page, we'll use 16K (from ARM)
+ * as the refill size on all platforms.
+ *
+ * When a refill zone drops to half that available, i.e. REFILL_SIZE / 2,
+ * zalloc_ext() will wake the replenish thread. The replenish thread runs
+ * until at least REFILL_SIZE worth of free elements exist, before sleeping again.
+ * In the meantime threads may continue to use the reserve until there are only REFILL_SIZE / 4
+ * elements left. Below that point only the replenish threads themselves and the GC
+ * thread may continue to use from the reserve.
+ */
+static unsigned zone_replenish_loops;
+static unsigned zone_replenish_wakeups;
+static unsigned zone_replenish_wakeups_initiated;
+static unsigned zone_replenish_throttle_count;
+
+#define ZONE_REPLENISH_TARGET (16 * 1024)
+static unsigned zone_replenish_active = 0; /* count of zones currently replenishing */
+static unsigned zone_replenish_max_threads = 0;
+
+LCK_GRP_DECLARE(zone_replenish_lock_grp, "zone_replenish_lock");
+LCK_SPIN_DECLARE(zone_replenish_lock, &zone_replenish_lock_grp);
+
+__abortlike
+static void
+zone_replenish_panic(zone_t zone, kern_return_t kr)
+{
+ panic_include_zprint = TRUE;
+#if CONFIG_ZLEAKS
+ if ((zleak_state & ZLEAK_STATE_ACTIVE)) {
+ panic_include_ztrace = TRUE;
+ }
+#endif /* CONFIG_ZLEAKS */
+ if (kr == KERN_NO_SPACE) {
+ zone_t zone_largest = zone_find_largest();
+ panic("zalloc: zone map exhausted while allocating from zone %s%s, "
+ "likely due to memory leak in zone %s%s "
+ "(%lu total bytes, %d elements allocated)",
+ zone_heap_name(zone), zone->z_name,
+ zone_heap_name(zone_largest), zone_largest->z_name,
+ (unsigned long)zone_size_wired(zone_largest),
+ zone_count_allocated(zone_largest));
}
-use_this_allocation:
- if (max && (max < alloc))
- max = alloc;
+ panic("zalloc: %s%s (%d elements) retry fail %d",
+ zone_heap_name(zone), zone->z_name,
+ zone_count_allocated(zone), kr);
+}
- /*
- * Opt into page list tracking if we can reliably map an allocation
- * to its page_metadata, and if the wastage in the tail of
- * the allocation is not too large
- */
+static void
+zone_replenish_locked(zone_t z, zalloc_flags_t flags, bool asynchronously)
+{
+ int kmaflags = KMA_KOBJECT | KMA_ZERO;
+ vm_offset_t space, alloc_size;
+ uint32_t retry = 0;
+ kern_return_t kr;
+
+ if (z->noencrypt) {
+ kmaflags |= KMA_NOENCRYPT;
+ }
+ if (flags & Z_NOPAGEWAIT) {
+ kmaflags |= KMA_NOPAGEWAIT;
+ }
+ if (z->permanent) {
+ kmaflags |= KMA_PERMANENT;
+ }
- /* zone_zone can't use page metadata since the page metadata will overwrite zone metadata */
- if (alloc == PAGE_SIZE && zone_zone != ZONE_NULL) {
- vm_offset_t first_element_offset;
- size_t zone_page_metadata_size = sizeof(struct zone_page_metadata);
+ for (;;) {
+ struct zone_page_metadata *page_meta = NULL;
- if (zone_page_metadata_size % ZONE_ELEMENT_ALIGNMENT == 0) {
- first_element_offset = zone_page_metadata_size;
+ /*
+ * Try to allocate our regular chunk of pages,
+ * unless the system is under massive pressure
+ * and we're looking for more than 2 pages.
+ */
+ if (!z->percpu && z->alloc_pages > 2 && (vm_pool_low() || retry > 0)) {
+ alloc_size = round_page(zone_elem_size(z));
} else {
- first_element_offset = zone_page_metadata_size + (ZONE_ELEMENT_ALIGNMENT - (zone_page_metadata_size % ZONE_ELEMENT_ALIGNMENT));
- }
-
- if (((PAGE_SIZE - first_element_offset) % size) <= PAGE_SIZE / 100) {
- use_page_list = TRUE;
- }
- }
-
- z->free_elements = NULL;
- queue_init(&z->pages.any_free_foreign);
- queue_init(&z->pages.all_free);
- queue_init(&z->pages.intermediate);
- queue_init(&z->pages.all_used);
- z->cur_size = 0;
- z->page_count = 0;
- z->max_size = max;
- z->elem_size = size;
- z->alloc_size = alloc;
- z->zone_name = name;
- z->count = 0;
- z->countfree = 0;
- z->sum_count = 0LL;
- z->doing_alloc_without_vm_priv = FALSE;
- z->doing_alloc_with_vm_priv = FALSE;
- z->doing_gc = FALSE;
- z->exhaustible = FALSE;
- z->collectable = TRUE;
- z->allows_foreign = FALSE;
- z->expandable = TRUE;
- z->waiting = FALSE;
- z->async_pending = FALSE;
- z->caller_acct = TRUE;
- z->noencrypt = FALSE;
- z->no_callout = FALSE;
- z->async_prio_refill = FALSE;
- z->gzalloc_exempt = FALSE;
- z->alignment_required = FALSE;
- z->use_page_list = use_page_list;
- z->prio_refill_watermark = 0;
- z->zone_replenish_thread = NULL;
- z->zp_count = 0;
+ alloc_size = ptoa(z->alloc_pages);
+ page_meta = zone_sequestered_page_get(z, &space);
+ }
+
+ unlock_zone(z);
+
#if CONFIG_ZLEAKS
- z->zleak_capture = 0;
- z->zleak_on = FALSE;
+ /*
+ * Do the zone leak activation here because zleak_activate()
+ * may block, and can't be done on the way out.
+ */
+ if (__improbable(zleak_state & ZLEAK_STATE_ENABLED)) {
+ if (!(zleak_state & ZLEAK_STATE_ACTIVE) &&
+ zone_submaps_approx_size() >= zleak_global_tracking_threshold) {
+ kr = zleak_activate();
+ if (kr != KERN_SUCCESS) {
+ printf("Failed to activate live zone leak debugging (%d).\n", kr);
+ }
+ }
+ }
#endif /* CONFIG_ZLEAKS */
-#if ZONE_DEBUG
- z->active_zones.next = z->active_zones.prev = NULL;
- zone_debug_enable(z);
-#endif /* ZONE_DEBUG */
- lock_zone_init(z);
+ /*
+ * Trigger jetsams via the vm_pageout_garbage_collect thread if
+ * we're running out of zone memory
+ */
+ if (is_zone_map_nearing_exhaustion()) {
+ thread_wakeup((event_t) &vm_pageout_garbage_collect);
+ }
- /*
- * Add the zone to the all-zones list.
- * If we are tracking zone info per task, and we have
- * already used all the available stat slots, then keep
- * using the overflow zone slot.
- */
- z->next_zone = ZONE_NULL;
- simple_lock(&all_zones_lock);
- *last_zone = z;
- last_zone = &z->next_zone;
- z->index = num_zones;
- if (zinfo_per_task) {
- if (num_zones > ZONES_MAX)
- z->index = ZONES_MAX;
- }
- num_zones++;
- simple_unlock(&all_zones_lock);
+ if (page_meta) {
+ kr = zone_sequestered_page_populate(z, page_meta, space,
+ alloc_size, kmaflags);
+ } else {
+ if (z->submap_idx == Z_SUBMAP_IDX_GENERAL_MAP && z->kalloc_heap != KHEAP_ID_NONE) {
+ kmaflags |= KMA_KHEAP;
+ }
+ kr = kernel_memory_allocate(submap_for_zone(z),
+ &space, alloc_size, 0, kmaflags, VM_KERN_MEMORY_ZONE);
+ }
- /*
- * Check if we should be logging this zone. If so, remember the zone pointer.
- */
- if (log_this_zone(z->zone_name, zone_name_to_log)) {
- zone_of_interest = z;
+#if !__LP64__
+ if (kr == KERN_NO_SPACE && z->allows_foreign) {
+ /*
+ * For zones allowing foreign pages, fallback to the kernel map
+ */
+ kr = kernel_memory_allocate(kernel_map, &space,
+ alloc_size, 0, kmaflags, VM_KERN_MEMORY_ZONE);
+ }
+#endif
+
+ if (kr == KERN_SUCCESS) {
+ break;
+ }
+
+ if (flags & Z_NOPAGEWAIT) {
+ lock_zone(z);
+ return;
+ }
+
+ if (asynchronously) {
+ assert_wait_timeout(&z->prio_refill_count,
+ THREAD_UNINT, 1, 100 * NSEC_PER_USEC);
+ thread_block(THREAD_CONTINUE_NULL);
+ } else if (++retry == 3) {
+ zone_replenish_panic(z, kr);
+ }
+
+ lock_zone(z);
}
- /*
- * If we want to log a zone, see if we need to allocate buffer space for the log. Some vm related zones are
- * zinit'ed before we can do a kmem_alloc, so we have to defer allocation in that case. kmem_alloc_ready is set to
- * TRUE once enough of the VM system is up and running to allow a kmem_alloc to work. If we want to log one
- * of the VM related zones that's set up early on, we will skip allocation of the log until zinit is called again
- * later on some other zone. So note we may be allocating a buffer to log a zone other than the one being initialized
- * right now.
- */
- if (zone_of_interest != NULL && zlog_btlog == NULL && kmem_alloc_ready) {
- zlog_btlog = btlog_create(log_records, MAX_ZTRACE_DEPTH, NULL, NULL, NULL);
- if (zlog_btlog) {
- printf("zone: logging started for zone %s\n", zone_of_interest->zone_name);
- } else {
- printf("zone: couldn't allocate memory for zrecords, turning off zleak logging\n");
- zone_of_interest = NULL;
+ zcram_and_lock(z, space, alloc_size);
+
+#if CONFIG_ZLEAKS
+ if (__improbable(zleak_state & ZLEAK_STATE_ACTIVE)) {
+ if (!z->zleak_on &&
+ zone_size_wired(z) >= zleak_per_zone_tracking_threshold) {
+ z->zleak_on = true;
}
}
-#if CONFIG_GZALLOC
- gzalloc_zone_init(z);
-#endif
- return(z);
+#endif /* CONFIG_ZLEAKS */
}
-unsigned zone_replenish_loops, zone_replenish_wakeups, zone_replenish_wakeups_initiated, zone_replenish_throttle_count;
-
-static void zone_replenish_thread(zone_t);
-/* High priority VM privileged thread used to asynchronously refill a designated
- * zone, such as the reserved VM map entry zone.
+/*
+ * High priority VM privileged thread used to asynchronously refill a given zone.
+ * These are needed for data structures used by the lower level VM itself. The
+ * replenish thread maintains a reserve of elements, so that the VM will never
+ * block in the zone allocator.
*/
-static void zone_replenish_thread(zone_t z) {
- vm_size_t free_size;
- current_thread()->options |= TH_OPT_VMPRIV;
+__dead2
+static void
+zone_replenish_thread(void *_z, wait_result_t __unused wr)
+{
+ zone_t z = _z;
+
+ current_thread()->options |= (TH_OPT_VMPRIV | TH_OPT_ZONE_PRIV);
for (;;) {
lock_zone(z);
- assert(z->prio_refill_watermark != 0);
- while ((free_size = (z->cur_size - (z->count * z->elem_size))) < (z->prio_refill_watermark * z->elem_size)) {
- assert(z->doing_alloc_without_vm_priv == FALSE);
- assert(z->doing_alloc_with_vm_priv == FALSE);
- assert(z->async_prio_refill == TRUE);
+ assert(z->z_self == z);
+ assert(z->zone_replenishing);
+ assert(z->prio_refill_count != 0);
- unlock_zone(z);
- int zflags = KMA_KOBJECT|KMA_NOPAGEWAIT;
- vm_offset_t space, alloc_size;
- kern_return_t kr;
-
- if (vm_pool_low())
- alloc_size = round_page(z->elem_size);
- else
- alloc_size = z->alloc_size;
-
- if (z->noencrypt)
- zflags |= KMA_NOENCRYPT;
-
- kr = kernel_memory_allocate(zone_map, &space, alloc_size, 0, zflags, VM_KERN_MEMORY_ZONE);
-
- if (kr == KERN_SUCCESS) {
-#if ZONE_ALIAS_ADDR
- if (alloc_size == PAGE_SIZE)
- space = zone_alias_addr(space);
-#endif
- zcram(z, space, alloc_size);
- } else if (kr == KERN_RESOURCE_SHORTAGE) {
- VM_PAGE_WAIT();
- } else if (kr == KERN_NO_SPACE) {
- kr = kernel_memory_allocate(kernel_map, &space, alloc_size, 0, zflags, VM_KERN_MEMORY_ZONE);
- if (kr == KERN_SUCCESS) {
-#if ZONE_ALIAS_ADDR
- if (alloc_size == PAGE_SIZE)
- space = zone_alias_addr(space);
-#endif
- zcram(z, space, alloc_size);
- } else {
- assert_wait_timeout(&z->zone_replenish_thread, THREAD_UNINT, 1, 100 * NSEC_PER_USEC);
- thread_block(THREAD_CONTINUE_NULL);
- }
- }
+ while (z->countfree < z->prio_refill_count) {
+ assert(!z->expanding_no_vm_priv);
+ assert(!z->expanding_vm_priv);
+
+ zone_replenish_locked(z, Z_WAITOK, true);
+
+ assert(z->z_self == z);
+ zone_replenish_loops++;
+ }
+
+ /* Wakeup any potentially throttled allocations. */
+ thread_wakeup(z);
- lock_zone(z);
- zone_replenish_loops++;
+ assert_wait(&z->prio_refill_count, THREAD_UNINT);
+
+ /*
+ * We finished refilling the zone, so decrement the active count
+ * and wake up any waiting GC threads.
+ */
+ lck_spin_lock(&zone_replenish_lock);
+ assert(zone_replenish_active > 0);
+ if (--zone_replenish_active == 0) {
+ thread_wakeup((event_t)&zone_replenish_active);
}
+ lck_spin_unlock(&zone_replenish_lock);
+ z->zone_replenishing = false;
unlock_zone(z);
- /* Signal any potential throttled consumers, terminating
- * their timer-bounded waits.
- */
- thread_wakeup(z);
- assert_wait(&z->zone_replenish_thread, THREAD_UNINT);
thread_block(THREAD_CONTINUE_NULL);
zone_replenish_wakeups++;
}
}
void
-zone_prio_refill_configure(zone_t z, vm_size_t low_water_mark) {
- z->prio_refill_watermark = low_water_mark;
-
- z->async_prio_refill = TRUE;
+zone_prio_refill_configure(zone_t z)
+{
+ thread_t th;
+ kern_return_t tres;
+
+ lock_zone(z);
+ assert(!z->prio_refill_count && !z->destructible);
+ z->prio_refill_count = (uint16_t)(ZONE_REPLENISH_TARGET / zone_elem_size(z));
+ z->zone_replenishing = true;
+ unlock_zone(z);
+
+ lck_spin_lock(&zone_replenish_lock);
+ ++zone_replenish_max_threads;
+ ++zone_replenish_active;
+ lck_spin_unlock(&zone_replenish_lock);
OSMemoryBarrier();
- kern_return_t tres = kernel_thread_start_priority((thread_continue_t)zone_replenish_thread, z, MAXPRI_KERNEL, &z->zone_replenish_thread);
+ tres = kernel_thread_start_priority(zone_replenish_thread, z,
+ MAXPRI_KERNEL, &th);
if (tres != KERN_SUCCESS) {
panic("zone_prio_refill_configure, thread create: 0x%x", tres);
}
- thread_deallocate(z->zone_replenish_thread);
+ thread_deallocate(th);
+}
+
+static void
+zone_randomize_freelist(zone_t zone, struct zone_page_metadata *meta,
+ vm_offset_t size, zone_addr_kind_t kind, unsigned int *entropy_buffer)
+{
+ const vm_size_t elem_size = zone_elem_size(zone);
+ vm_offset_t left, right, head, base;
+ vm_offset_t element;
+
+ left = ZONE_PAGE_FIRST_OFFSET(kind);
+ right = size - ((size - left) % elem_size);
+ head = 0;
+ base = zone_meta_to_addr(meta, kind);
+
+ while (left < right) {
+ if (zone_leaks_scan_enable || __improbable(zone->tags) ||
+ random_bool_gen_bits(&zone_bool_gen, entropy_buffer, MAX_ENTROPY_PER_ZCRAM, 1)) {
+ element = base + left;
+ left += elem_size;
+ } else {
+ right -= elem_size;
+ element = base + right;
+ }
+
+ vm_offset_t *primary = (vm_offset_t *)element;
+ vm_offset_t *backup = get_backup_ptr(elem_size, primary);
+
+ *primary = *backup = head ^ zp_nopoison_cookie;
+ head = element;
+ }
+
+ meta->zm_freelist_offs = (uint16_t)(head - base);
}
/*
* Cram the given memory into the specified zone. Update the zone page count accordingly.
*/
-void
-zcram(
- zone_t zone,
- vm_offset_t newmem,
- vm_size_t size)
+static void
+zcram_and_lock(zone_t zone, vm_offset_t newmem, vm_size_t size)
{
- vm_size_t elem_size;
- boolean_t from_zm = FALSE;
+ unsigned int entropy_buffer[MAX_ENTROPY_PER_ZCRAM] = { 0 };
+ struct zone_page_metadata *meta;
+ zone_addr_kind_t kind;
+ uint32_t pg_count = (uint32_t)atop(size);
+ uint32_t zindex = zone_index(zone);
+ uint32_t free_count;
+ uint16_t empty_freelist_offs = PAGE_METADATA_EMPTY_FREELIST;
/* Basic sanity checks */
assert(zone != ZONE_NULL && newmem != (vm_offset_t)0);
- assert(!zone->collectable || zone->allows_foreign
- || (from_zone_map(newmem, size)));
+ assert((newmem & PAGE_MASK) == 0);
+ assert((size & PAGE_MASK) == 0);
+
+ KDBG(MACHDBG_CODE(DBG_MACH_ZALLOC, ZALLOC_ZCRAM) | DBG_FUNC_START,
+ zindex, size);
+
+ kind = zone_addr_kind(newmem, size);
+#if DEBUG || DEVELOPMENT
+ if (zalloc_debug & ZALLOC_DEBUG_ZCRAM) {
+ kprintf("zcram(%p[%s%s], 0x%lx%s, 0x%lx)\n", zone,
+ zone_heap_name(zone), zone->z_name, (uintptr_t)newmem,
+ kind == ZONE_ADDR_FOREIGN ? "[F]" : "", (uintptr_t)size);
+ }
+#endif /* DEBUG || DEVELOPMENT */
+
+ /*
+ * Initialize the metadata for all pages. We dont need the zone lock
+ * here because we are not manipulating any zone related state yet.
+ *
+ * This includes randomizing the freelists as the metadata isn't
+ * published yet.
+ */
+
+ if (kind == ZONE_ADDR_NATIVE) {
+ /*
+ * We're being called by zfill,
+ * zone_replenish_thread or vm_page_more_fictitious,
+ *
+ * which will only either allocate a single page, or `alloc_pages`
+ * worth.
+ */
+ assert(pg_count <= zone->alloc_pages);
+
+ /*
+ * Make sure the range of metadata entries we're about to init
+ * have proper physical backing, then initialize them.
+ */
+ meta = zone_meta_from_addr(newmem, kind);
+ zone_meta_populate(meta, meta + pg_count);
+
+ if (zone->permanent) {
+ empty_freelist_offs = 0;
+ }
- elem_size = zone->elem_size;
+ meta[0] = (struct zone_page_metadata){
+ .zm_index = zindex,
+ .zm_page_count = pg_count,
+ .zm_percpu = zone->percpu,
+ .zm_freelist_offs = empty_freelist_offs,
+ };
+
+ for (uint32_t i = 1; i < pg_count; i++) {
+ meta[i] = (struct zone_page_metadata){
+ .zm_index = zindex,
+ .zm_page_count = i,
+ .zm_percpu = zone->percpu,
+ .zm_secondary_page = true,
+ .zm_freelist_offs = empty_freelist_offs,
+ };
+ }
- if (from_zone_map(newmem, size))
- from_zm = TRUE;
+ if (!zone->permanent) {
+ zone_randomize_freelist(zone, meta,
+ zone->percpu ? PAGE_SIZE : size, kind, entropy_buffer);
+ }
+ } else {
+ if (!zone->allows_foreign || !from_foreign_range(newmem, size)) {
+ panic("zcram_and_lock: foreign memory [%lx] being crammed is "
+ "outside of foreign range", (uintptr_t)newmem);
+ }
- if (zalloc_debug & ZALLOC_DEBUG_ZCRAM)
- kprintf("zcram(%p[%s], 0x%lx%s, 0x%lx)\n", zone, zone->zone_name,
- (unsigned long)newmem, from_zm ? "" : "[F]", (unsigned long)size);
+ /*
+ * We cannot support elements larger than page size for foreign
+ * memory because we put metadata on the page itself for each
+ * page of foreign memory.
+ *
+ * We need to do this in order to be able to reach the metadata
+ * when any element is freed.
+ */
+ assert(!zone->percpu && !zone->permanent);
+ assert(zone_elem_size(zone) <= PAGE_SIZE - sizeof(struct zone_page_metadata));
+
+ bzero((void *)newmem, size);
+
+ for (vm_offset_t offs = 0; offs < size; offs += PAGE_SIZE) {
+ meta = (struct zone_page_metadata *)(newmem + offs);
+ *meta = (struct zone_page_metadata){
+ .zm_index = zindex,
+ .zm_page_count = 1,
+ .zm_freelist_offs = empty_freelist_offs,
+ };
+ meta->zm_foreign_cookie[0] = ZONE_FOREIGN_COOKIE;
+ zone_randomize_freelist(zone, meta, PAGE_SIZE, kind,
+ entropy_buffer);
+ }
+ }
- if (from_zm && !zone->use_page_list)
- zone_page_init(newmem, size);
+#if VM_MAX_TAG_ZONES
+ if (__improbable(zone->tags)) {
+ assert(kind == ZONE_ADDR_NATIVE && !zone->percpu);
+ ztMemoryAdd(zone, newmem, size);
+ }
+#endif /* VM_MAX_TAG_ZONES */
- ZONE_PAGE_COUNT_INCR(zone, (size / PAGE_SIZE));
+ /*
+ * Insert the initialized pages / metadatas into the right lists.
+ */
lock_zone(zone);
+ assert(zone->z_self == zone);
- if (zone->use_page_list) {
- struct zone_page_metadata *page_metadata;
- size_t zone_page_metadata_size = sizeof(struct zone_page_metadata);
-
- assert((newmem & PAGE_MASK) == 0);
- assert((size & PAGE_MASK) == 0);
- for (; size > 0; newmem += PAGE_SIZE, size -= PAGE_SIZE) {
-
- vm_size_t pos_in_page;
- page_metadata = (struct zone_page_metadata *)(newmem);
-
- page_metadata->pages.next = NULL;
- page_metadata->pages.prev = NULL;
- page_metadata->elements = NULL;
- page_metadata->zone = zone;
- page_metadata->alloc_count = 0;
- page_metadata->free_count = 0;
-
- enqueue_tail(&zone->pages.all_used, (queue_entry_t)page_metadata);
-
- vm_offset_t first_element_offset;
- if (zone_page_metadata_size % ZONE_ELEMENT_ALIGNMENT == 0){
- first_element_offset = zone_page_metadata_size;
- } else {
- first_element_offset = zone_page_metadata_size + (ZONE_ELEMENT_ALIGNMENT - (zone_page_metadata_size % ZONE_ELEMENT_ALIGNMENT));
- }
+ zone->page_count += pg_count;
+ if (zone->page_count_hwm < zone->page_count) {
+ zone->page_count_hwm = zone->page_count;
+ }
+ os_atomic_add(&zones_phys_page_count, pg_count, relaxed);
- for (pos_in_page = first_element_offset; (newmem + pos_in_page + elem_size) < (vm_offset_t)(newmem + PAGE_SIZE); pos_in_page += elem_size) {
- page_metadata->alloc_count++;
- zone->count++; /* compensate for free_to_zone */
- free_to_zone(zone, newmem + pos_in_page, FALSE);
- zone->cur_size += elem_size;
- }
+ if (kind == ZONE_ADDR_NATIVE) {
+ os_atomic_add(&zones_phys_page_mapped_count, pg_count, relaxed);
+ if (zone->permanent) {
+ zone_meta_queue_push(zone, &zone->pages_intermediate, meta, kind);
+ } else {
+ zone_meta_queue_push(zone, &zone->pages_all_free, meta, kind);
+ zone->allfree_page_count += meta->zm_page_count;
}
+ free_count = zone_elem_count(zone, size, kind);
+ zone->countfree += free_count;
+ zone->countavail += free_count;
} else {
- while (size >= elem_size) {
- zone->count++; /* compensate for free_to_zone */
- if (newmem == (vm_offset_t)zone) {
- /* Don't free zone_zone zone */
- } else {
- free_to_zone(zone, newmem, FALSE);
- }
- if (from_zm)
- zone_page_alloc(newmem, elem_size);
- size -= elem_size;
- newmem += elem_size;
- zone->cur_size += elem_size;
+ free_count = zone_elem_count(zone, PAGE_SIZE, kind);
+ for (vm_offset_t offs = 0; offs < size; offs += PAGE_SIZE) {
+ meta = (struct zone_page_metadata *)(newmem + offs);
+ zone_meta_queue_push(zone, &zone->pages_any_free_foreign, meta, kind);
+ zone->countfree += free_count;
+ zone->countavail += free_count;
}
}
- unlock_zone(zone);
-}
+ KDBG(MACHDBG_CODE(DBG_MACH_ZALLOC, ZALLOC_ZCRAM) | DBG_FUNC_END, zindex);
+}
-/*
- * Steal memory for the zone package. Called from
- * vm_page_bootstrap().
- */
void
-zone_steal_memory(void)
+zcram(zone_t zone, vm_offset_t newmem, vm_size_t size)
{
-#if CONFIG_GZALLOC
- gzalloc_configure();
-#endif
- /* Request enough early memory to get to the pmap zone */
- zdata_size = 12 * sizeof(struct zone);
- zdata_size = round_page(zdata_size);
- zdata = (vm_offset_t)pmap_steal_memory(zdata_size);
+ zcram_and_lock(zone, newmem, size);
+ unlock_zone(zone);
}
-
/*
* Fill a zone with enough memory to contain at least nelem elements.
- * Memory is obtained with kmem_alloc_kobject from the kernel_map.
* Return the number of elements actually put into the zone, which may
* be more than the caller asked for since the memory allocation is
- * rounded up to a full page.
+ * rounded up to the next zone allocation size.
*/
int
zfill(
- zone_t zone,
- int nelem)
+ zone_t zone,
+ int nelem)
{
- kern_return_t kr;
- vm_size_t size;
- vm_offset_t memory;
- int nalloc;
+ kern_return_t kr;
+ vm_offset_t memory;
- assert(nelem > 0);
- if (nelem <= 0)
- return 0;
- size = nelem * zone->elem_size;
- size = round_page(size);
- kr = kmem_alloc_kobject(kernel_map, &memory, size, VM_KERN_MEMORY_ZONE);
- if (kr != KERN_SUCCESS)
- return 0;
+ vm_size_t alloc_size = ptoa(zone->alloc_pages);
+ vm_size_t nalloc_inc = zone_elem_count(zone, alloc_size, ZONE_ADDR_NATIVE);
+ vm_size_t nalloc = 0, goal = MAX(0, nelem);
+ int kmaflags = KMA_KOBJECT | KMA_ZERO;
+
+ if (zone->noencrypt) {
+ kmaflags |= KMA_NOENCRYPT;
+ }
+
+ assert(!zone->allows_foreign && !zone->permanent);
+
+ /*
+ * Trigger jetsams via the vm_pageout_garbage_collect thread if we're
+ * running out of zone memory
+ */
+ if (is_zone_map_nearing_exhaustion()) {
+ thread_wakeup((event_t) &vm_pageout_garbage_collect);
+ }
+
+ if (zone->va_sequester) {
+ lock_zone(zone);
+
+ do {
+ struct zone_page_metadata *page_meta;
+ page_meta = zone_sequestered_page_get(zone, &memory);
+ if (NULL == page_meta) {
+ break;
+ }
+ unlock_zone(zone);
+
+ kr = zone_sequestered_page_populate(zone, page_meta,
+ memory, alloc_size, kmaflags);
+ if (KERN_SUCCESS != kr) {
+ goto out_nolock;
+ }
+
+ zcram_and_lock(zone, memory, alloc_size);
+ nalloc += nalloc_inc;
+ } while (nalloc < goal);
+
+ unlock_zone(zone);
+ }
+
+out_nolock:
+ while (nalloc < goal) {
+ kr = kernel_memory_allocate(submap_for_zone(zone), &memory,
+ alloc_size, 0, kmaflags, VM_KERN_MEMORY_ZONE);
+ if (kr != KERN_SUCCESS) {
+ printf("%s: kernel_memory_allocate() of %lu bytes failed\n",
+ __func__, (unsigned long)(nalloc * alloc_size));
+ break;
+ }
+
+ zcram(zone, memory, alloc_size);
+ nalloc += nalloc_inc;
+ }
+
+ return (int)nalloc;
+}
+
+/*
+ * We're being very conservative here and picking a value of 95%. We might need to lower this if
+ * we find that we're not catching the problem and are still hitting zone map exhaustion panics.
+ */
+#define ZONE_MAP_JETSAM_LIMIT_DEFAULT 95
+
+/*
+ * Trigger zone-map-exhaustion jetsams if the zone map is X% full, where X=zone_map_jetsam_limit.
+ * Can be set via boot-arg "zone_map_jetsam_limit". Set to 95% by default.
+ */
+TUNABLE_WRITEABLE(unsigned int, zone_map_jetsam_limit, "zone_map_jetsam_limit",
+ ZONE_MAP_JETSAM_LIMIT_DEFAULT);
+
+void
+get_zone_map_size(uint64_t *current_size, uint64_t *capacity)
+{
+ vm_offset_t phys_pages = os_atomic_load(&zones_phys_page_mapped_count, relaxed);
+ *current_size = ptoa_64(phys_pages);
+ *capacity = zone_phys_mapped_max;
+}
+
+void
+get_largest_zone_info(char *zone_name, size_t zone_name_len, uint64_t *zone_size)
+{
+ zone_t largest_zone = zone_find_largest();
+
+ /*
+ * Append kalloc heap name to zone name (if zone is used by kalloc)
+ */
+ snprintf(zone_name, zone_name_len, "%s%s",
+ zone_heap_name(largest_zone), largest_zone->z_name);
+
+ *zone_size = zone_size_wired(largest_zone);
+}
+
+boolean_t
+is_zone_map_nearing_exhaustion(void)
+{
+ vm_offset_t phys_pages = os_atomic_load(&zones_phys_page_mapped_count, relaxed);
+ return ptoa_64(phys_pages) > (zone_phys_mapped_max * zone_map_jetsam_limit) / 100;
+}
+
+
+#define VMENTRY_TO_VMOBJECT_COMPARISON_RATIO 98
+
+/*
+ * Tries to kill a single process if it can attribute one to the largest zone. If not, wakes up the memorystatus thread
+ * to walk through the jetsam priority bands and kill processes.
+ */
+static void
+kill_process_in_largest_zone(void)
+{
+ pid_t pid = -1;
+ zone_t largest_zone = zone_find_largest();
+
+ printf("zone_map_exhaustion: Zone mapped %lld of %lld, used %lld, map size %lld, capacity %lld [jetsam limit %d%%]\n",
+ ptoa_64(os_atomic_load(&zones_phys_page_mapped_count, relaxed)), ptoa_64(zone_phys_mapped_max),
+ ptoa_64(os_atomic_load(&zones_phys_page_count, relaxed)),
+ (uint64_t)zone_submaps_approx_size(),
+ (uint64_t)zone_range_size(&zone_info.zi_map_range),
+ zone_map_jetsam_limit);
+ printf("zone_map_exhaustion: Largest zone %s%s, size %lu\n", zone_heap_name(largest_zone),
+ largest_zone->z_name, (uintptr_t)zone_size_wired(largest_zone));
+
+ /*
+ * We want to make sure we don't call this function from userspace.
+ * Or we could end up trying to synchronously kill the process
+ * whose context we're in, causing the system to hang.
+ */
+ assert(current_task() == kernel_task);
- zone_change(zone, Z_FOREIGN, TRUE);
- zcram(zone, memory, size);
- nalloc = (int)(size / zone->elem_size);
- assert(nalloc >= nelem);
+ /*
+ * If vm_object_zone is the largest, check to see if the number of
+ * elements in vm_map_entry_zone is comparable.
+ *
+ * If so, consider vm_map_entry_zone as the largest. This lets us target
+ * a specific process to jetsam to quickly recover from the zone map
+ * bloat.
+ */
+ if (largest_zone == vm_object_zone) {
+ unsigned int vm_object_zone_count = zone_count_allocated(vm_object_zone);
+ unsigned int vm_map_entry_zone_count = zone_count_allocated(vm_map_entry_zone);
+ /* Is the VM map entries zone count >= 98% of the VM objects zone count? */
+ if (vm_map_entry_zone_count >= ((vm_object_zone_count * VMENTRY_TO_VMOBJECT_COMPARISON_RATIO) / 100)) {
+ largest_zone = vm_map_entry_zone;
+ printf("zone_map_exhaustion: Picking VM map entries as the zone to target, size %lu\n",
+ (uintptr_t)zone_size_wired(largest_zone));
+ }
+ }
- return nalloc;
+ /* TODO: Extend this to check for the largest process in other zones as well. */
+ if (largest_zone == vm_map_entry_zone) {
+ pid = find_largest_process_vm_map_entries();
+ } else {
+ printf("zone_map_exhaustion: Nothing to do for the largest zone [%s%s]. "
+ "Waking up memorystatus thread.\n", zone_heap_name(largest_zone),
+ largest_zone->z_name);
+ }
+ if (!memorystatus_kill_on_zone_map_exhaustion(pid)) {
+ printf("zone_map_exhaustion: Call to memorystatus failed, victim pid: %d\n", pid);
+ }
}
+#pragma mark zalloc module init
+
/*
* Initialize the "zone of zones" which uses fixed memory allocated
* earlier in memory initialization. zone_bootstrap is called
* before zone_init.
*/
+__startup_func
void
zone_bootstrap(void)
{
- char temp_buf[16];
+ /* Validate struct zone_page_metadata expectations */
+ if ((1U << ZONE_PAGECOUNT_BITS) <
+ atop(ZONE_MAX_ALLOC_SIZE) * sizeof(struct zone_page_metadata)) {
+ panic("ZONE_PAGECOUNT_BITS is not large enough to hold page counts");
+ }
- if (PE_parse_boot_argn("-zinfop", temp_buf, sizeof(temp_buf))) {
- zinfo_per_task = TRUE;
+ /* Validate struct zone_packed_virtual_address expectations */
+ static_assert((intptr_t)VM_MIN_KERNEL_ADDRESS < 0, "the top bit must be 1");
+ if (VM_KERNEL_POINTER_SIGNIFICANT_BITS - PAGE_SHIFT > 31) {
+ panic("zone_pva_t can't pack a kernel page address in 31 bits");
}
- if (!PE_parse_boot_argn("zalloc_debug", &zalloc_debug, sizeof(zalloc_debug)))
- zalloc_debug = 0;
+ zpercpu_early_count = ml_early_cpu_max_number() + 1;
/* Set up zone element poisoning */
- zp_init();
+ zp_bootstrap();
- /* should zlog log to debug zone corruption instead of leaks? */
- if (PE_parse_boot_argn("-zc", temp_buf, sizeof(temp_buf))) {
- corruption_debug_flag = TRUE;
- }
+ random_bool_init(&zone_bool_gen);
/*
- * Check for and set up zone leak detection if requested via boot-args. We recognized two
- * boot-args:
- *
- * zlog=<zone_to_log>
- * zrecs=<num_records_in_log>
+ * the KASAN quarantine for kalloc doesn't understand heaps
+ * and trips the heap confusion panics. At the end of the day,
+ * all these security measures are double duty with KASAN.
*
- * The zlog arg is used to specify the zone name that should be logged, and zrecs is used to
- * control the size of the log. If zrecs is not specified, a default value is used.
+ * On 32bit kernels, these protections are just too expensive.
*/
+#if !defined(__LP64__) || KASAN_ZALLOC
+ zsecurity_options &= ~ZSECURITY_OPTIONS_SEQUESTER;
+ zsecurity_options &= ~ZSECURITY_OPTIONS_SUBMAP_USER_DATA;
+ zsecurity_options &= ~ZSECURITY_OPTIONS_SEQUESTER_KEXT_KALLOC;
+#endif
- if (PE_parse_boot_argn("zlog", zone_name_to_log, sizeof(zone_name_to_log)) == TRUE) {
- if (PE_parse_boot_argn("zrecs", &log_records, sizeof(log_records)) == TRUE) {
-
- /*
- * Don't allow more than ZRECORDS_MAX records even if the user asked for more.
- * This prevents accidentally hogging too much kernel memory and making the system
- * unusable.
- */
-
- log_records = MIN(ZRECORDS_MAX, log_records);
+ thread_call_setup(&call_async_alloc, zalloc_async, NULL);
- } else {
- log_records = ZRECORDS_DEFAULT;
- }
+#if CONFIG_ZCACHE
+ /* zcc_enable_for_zone_name=<zone>: enable per-cpu zone caching for <zone>. */
+ if (PE_parse_boot_arg_str("zcc_enable_for_zone_name", cache_zone_name, sizeof(cache_zone_name))) {
+ printf("zcache: caching enabled for zone %s\n", cache_zone_name);
}
+#endif /* CONFIG_ZCACHE */
+}
- simple_lock_init(&all_zones_lock, 0);
+#if __LP64__
+#if CONFIG_EMBEDDED
+#define ZONE_MAP_VIRTUAL_SIZE_LP64 (32ULL * 1024ULL * 1024 * 1024)
+#else
+#define ZONE_MAP_VIRTUAL_SIZE_LP64 (128ULL * 1024ULL * 1024 * 1024)
+#endif
+#endif /* __LP64__ */
- first_zone = ZONE_NULL;
- last_zone = &first_zone;
- num_zones = 0;
- thread_call_setup(&call_async_alloc, zalloc_async, NULL);
+#define SINGLE_GUARD 16384
+#define MULTI_GUARD (3 * SINGLE_GUARD)
- /* assertion: nobody else called zinit before us */
- assert(zone_zone == ZONE_NULL);
-
- /* initializing global lock group for zones */
- lck_grp_attr_setdefault(&zone_locks_grp_attr);
- lck_grp_init(&zone_locks_grp, "zone_locks", &zone_locks_grp_attr);
+#if __LP64__
+static inline vm_offset_t
+zone_restricted_va_max(void)
+{
+ vm_offset_t compressor_max = VM_PACKING_MAX_PACKABLE(C_SLOT_PACKED_PTR);
+ vm_offset_t vm_page_max = VM_PACKING_MAX_PACKABLE(VM_PAGE_PACKED_PTR);
+
+ return trunc_page(MIN(compressor_max, vm_page_max));
+}
+#endif
- zone_zone = zinit(sizeof(struct zone), 128 * sizeof(struct zone),
- sizeof(struct zone), "zones");
- zone_change(zone_zone, Z_COLLECT, FALSE);
- zone_change(zone_zone, Z_CALLERACCT, FALSE);
- zone_change(zone_zone, Z_NOENCRYPT, TRUE);
+__startup_func
+static void
+zone_tunables_fixup(void)
+{
+ if (zone_map_jetsam_limit == 0 || zone_map_jetsam_limit > 100) {
+ zone_map_jetsam_limit = ZONE_MAP_JETSAM_LIMIT_DEFAULT;
+ }
+}
+STARTUP(TUNABLES, STARTUP_RANK_MIDDLE, zone_tunables_fixup);
- zcram(zone_zone, zdata, zdata_size);
- VM_PAGE_MOVE_STOLEN(atop_64(zdata_size));
+__startup_func
+static vm_size_t
+zone_phys_size_max(void)
+{
+ mach_vm_size_t zsize;
+ vm_size_t zsizearg;
- /* initialize fake zones and zone info if tracking by task */
- if (zinfo_per_task) {
- vm_size_t zisize = sizeof(zinfo_usage_store_t) * ZINFO_SLOTS;
- unsigned int i;
+ if (PE_parse_boot_argn("zsize", &zsizearg, sizeof(zsizearg))) {
+ zsize = zsizearg * (1024ULL * 1024);
+ } else {
+ zsize = sane_size >> 2; /* Set target zone size as 1/4 of physical memory */
+#if defined(__LP64__)
+ zsize += zsize >> 1;
+#endif /* __LP64__ */
+ }
- for (i = 0; i < num_fake_zones; i++)
- fake_zones[i].init(ZINFO_SLOTS - num_fake_zones + i);
- zinfo_zone = zinit(zisize, zisize * CONFIG_TASK_MAX,
- zisize, "per task zinfo");
- zone_change(zinfo_zone, Z_CALLERACCT, FALSE);
+ if (zsize < CONFIG_ZONE_MAP_MIN) {
+ zsize = CONFIG_ZONE_MAP_MIN; /* Clamp to min */
+ }
+ if (zsize > sane_size >> 1) {
+ zsize = sane_size >> 1; /* Clamp to half of RAM max */
}
+ if (zsizearg == 0 && zsize > ZONE_MAP_MAX) {
+ /* if zsize boot-arg not present and zsize exceeds platform maximum, clip zsize */
+ vm_size_t orig_zsize = zsize;
+ zsize = ZONE_MAP_MAX;
+ printf("NOTE: zonemap size reduced from 0x%lx to 0x%lx\n",
+ (uintptr_t)orig_zsize, (uintptr_t)zsize);
+ }
+
+ assert((vm_size_t) zsize == zsize);
+ return (vm_size_t)trunc_page(zsize);
}
-void
-zinfo_task_init(task_t task)
+__startup_func
+static struct zone_map_range
+zone_init_allocate_va(vm_offset_t *submap_min, vm_size_t size, bool guard)
{
- if (zinfo_per_task) {
- task->tkm_zinfo = zalloc(zinfo_zone);
- memset(task->tkm_zinfo, 0, sizeof(zinfo_usage_store_t) * ZINFO_SLOTS);
+ struct zone_map_range r;
+ kern_return_t kr;
+
+ if (guard) {
+ vm_map_offset_t addr = *submap_min;
+ vm_map_kernel_flags_t vmk_flags = VM_MAP_KERNEL_FLAGS_NONE;
+
+ vmk_flags.vmkf_permanent = TRUE;
+ kr = vm_map_enter(kernel_map, &addr, size, 0,
+ VM_FLAGS_FIXED, vmk_flags, VM_KERN_MEMORY_ZONE, kernel_object,
+ 0, FALSE, VM_PROT_NONE, VM_PROT_NONE, VM_INHERIT_DEFAULT);
+ *submap_min = (vm_offset_t)addr;
} else {
- task->tkm_zinfo = NULL;
+ kr = kernel_memory_allocate(kernel_map, submap_min, size,
+ 0, KMA_KOBJECT | KMA_PAGEABLE | KMA_VAONLY, VM_KERN_MEMORY_ZONE);
+ }
+ if (kr != KERN_SUCCESS) {
+ panic("zone_init_allocate_va(0x%lx:0x%zx) failed: %d",
+ (uintptr_t)*submap_min, (size_t)size, kr);
}
+
+ r.min_address = *submap_min;
+ *submap_min += size;
+ r.max_address = *submap_min;
+
+ return r;
}
-void
-zinfo_task_free(task_t task)
+__startup_func
+static void
+zone_submap_init(
+ vm_offset_t *submap_min,
+ unsigned idx,
+ uint64_t zone_sub_map_numer,
+ uint64_t *remaining_denom,
+ vm_offset_t *remaining_size,
+ vm_size_t guard_size)
{
- assert(task != kernel_task);
- if (task->tkm_zinfo != NULL) {
- zfree(zinfo_zone, task->tkm_zinfo);
- task->tkm_zinfo = NULL;
+ vm_offset_t submap_start, submap_end;
+ vm_size_t submap_size;
+ vm_map_t submap;
+ kern_return_t kr;
+
+ submap_size = trunc_page(zone_sub_map_numer * *remaining_size /
+ *remaining_denom);
+ submap_start = *submap_min;
+ submap_end = submap_start + submap_size;
+
+#if defined(__LP64__)
+ if (idx == Z_SUBMAP_IDX_VA_RESTRICTED_MAP) {
+ vm_offset_t restricted_va_max = zone_restricted_va_max();
+ if (submap_end > restricted_va_max) {
+#if DEBUG || DEVELOPMENT
+ printf("zone_init: submap[%d] clipped to %zdM of %zdM\n", idx,
+ (size_t)(restricted_va_max - submap_start) >> 20,
+ (size_t)submap_size >> 20);
+#endif /* DEBUG || DEVELOPMENT */
+ guard_size += submap_end - restricted_va_max;
+ *remaining_size -= submap_end - restricted_va_max;
+ submap_end = restricted_va_max;
+ submap_size = restricted_va_max - submap_start;
+ }
+
+ vm_packing_verify_range("vm_compressor",
+ submap_start, submap_end, VM_PACKING_PARAMS(C_SLOT_PACKED_PTR));
+ vm_packing_verify_range("vm_page",
+ submap_start, submap_end, VM_PACKING_PARAMS(VM_PAGE_PACKED_PTR));
}
+#endif /* defined(__LP64__) */
+
+ vm_map_kernel_flags_t vmk_flags = VM_MAP_KERNEL_FLAGS_NONE;
+ vmk_flags.vmkf_permanent = TRUE;
+ kr = kmem_suballoc(kernel_map, submap_min, submap_size,
+ FALSE, VM_FLAGS_FIXED, vmk_flags,
+ VM_KERN_MEMORY_ZONE, &submap);
+ if (kr != KERN_SUCCESS) {
+ panic("kmem_suballoc(kernel_map[%d] %p:%p) failed: %d",
+ idx, (void *)submap_start, (void *)submap_end, kr);
+ }
+
+#if DEBUG || DEVELOPMENT
+ printf("zone_init: submap[%d] %p:%p (%zuM)\n",
+ idx, (void *)submap_start, (void *)submap_end,
+ (size_t)submap_size >> 20);
+#endif /* DEBUG || DEVELOPMENT */
+
+ zone_submaps[idx] = submap;
+ *submap_min = submap_end;
+ *remaining_size -= submap_size;
+ *remaining_denom -= zone_sub_map_numer;
+
+ zone_init_allocate_va(submap_min, guard_size, true);
}
-
+
/* Global initialization of Zone Allocator.
* Runs after zone_bootstrap.
*/
-void
-zone_init(
- vm_size_t max_zonemap_size)
+__startup_func
+static void
+zone_init(void)
{
- kern_return_t retval;
- vm_offset_t zone_min;
- vm_offset_t zone_max;
+ vm_size_t zone_meta_size;
+ vm_size_t zone_map_size;
+ vm_size_t remaining_size;
+ vm_offset_t submap_min = 0;
- retval = kmem_suballoc(kernel_map, &zone_min, max_zonemap_size,
- FALSE, VM_FLAGS_ANYWHERE | VM_FLAGS_PERMANENT | VM_MAKE_TAG(VM_KERN_MEMORY_ZONE),
- &zone_map);
+ if (ZSECURITY_OPTIONS_SUBMAP_USER_DATA & zsecurity_options) {
+ zone_last_submap_idx = Z_SUBMAP_IDX_BAG_OF_BYTES_MAP;
+ } else {
+ zone_last_submap_idx = Z_SUBMAP_IDX_GENERAL_MAP;
+ }
+ zone_phys_mapped_max = zone_phys_size_max();
- if (retval != KERN_SUCCESS)
- panic("zone_init: kmem_suballoc failed");
- zone_max = zone_min + round_page(max_zonemap_size);
-#if CONFIG_GZALLOC
- gzalloc_init(max_zonemap_size);
+#if __LP64__
+ zone_map_size = ZONE_MAP_VIRTUAL_SIZE_LP64;
+#else
+ zone_map_size = zone_phys_mapped_max;
#endif
+ zone_meta_size = round_page(atop(zone_map_size) *
+ sizeof(struct zone_page_metadata));
+
/*
- * Setup garbage collection information:
+ * Zone "map" setup:
+ *
+ * [ VA_RESTRICTED ] <-- LP64 only
+ * [ SINGLE_GUARD ] <-- LP64 only
+ * [ meta ]
+ * [ SINGLE_GUARD ]
+ * [ map<i> ] \ for each extra map
+ * [ MULTI_GUARD ] /
*/
- zone_map_min_address = zone_min;
- zone_map_max_address = zone_max;
-
+ remaining_size = zone_map_size;
#if defined(__LP64__)
+ remaining_size -= SINGLE_GUARD;
+#endif
+ remaining_size -= zone_meta_size + SINGLE_GUARD;
+ remaining_size -= MULTI_GUARD * (zone_last_submap_idx -
+ Z_SUBMAP_IDX_GENERAL_MAP + 1);
+
+#if VM_MAX_TAG_ZONES
+ if (zone_tagging_on) {
+ zone_tagging_init(zone_map_size);
+ }
+#endif
+
+ uint64_t remaining_denom = 0;
+ uint64_t zone_sub_map_numer[Z_SUBMAP_IDX_COUNT] = {
+#ifdef __LP64__
+ [Z_SUBMAP_IDX_VA_RESTRICTED_MAP] = 20,
+#endif /* defined(__LP64__) */
+ [Z_SUBMAP_IDX_GENERAL_MAP] = 40,
+ [Z_SUBMAP_IDX_BAG_OF_BYTES_MAP] = 40,
+ };
+
+ for (unsigned idx = 0; idx <= zone_last_submap_idx; idx++) {
+#if DEBUG || DEVELOPMENT
+ char submap_name[MAX_SUBMAP_NAME];
+ snprintf(submap_name, MAX_SUBMAP_NAME, "submap%d", idx);
+ PE_parse_boot_argn(submap_name, &zone_sub_map_numer[idx], sizeof(uint64_t));
+#endif
+ remaining_denom += zone_sub_map_numer[idx];
+ }
+
/*
- * ensure that any vm_page_t that gets created from
- * the vm_page zone can be packed properly (see vm_page.h
- * for the packing requirements
+ * And now allocate the various pieces of VA and submaps.
+ *
+ * Make a first allocation of contiguous VA, that we'll deallocate,
+ * and we'll carve-out memory in that range again linearly.
+ * The kernel is stil single threaded at this stage.
*/
- if (VM_PAGE_UNPACK_PTR(VM_PAGE_PACK_PTR(zone_map_min_address)) != (vm_page_t)zone_map_min_address)
- panic("VM_PAGE_PACK_PTR failed on zone_map_min_address - %p", (void *)zone_map_min_address);
- if (VM_PAGE_UNPACK_PTR(VM_PAGE_PACK_PTR(zone_map_max_address)) != (vm_page_t)zone_map_max_address)
- panic("VM_PAGE_PACK_PTR failed on zone_map_max_address - %p", (void *)zone_map_max_address);
-#endif
+ struct zone_map_range *map_range = &zone_info.zi_map_range;
- zone_pages = (unsigned int)atop_kernel(zone_max - zone_min);
- zone_page_table_used_size = sizeof(zone_page_table);
+ *map_range = zone_init_allocate_va(&submap_min, zone_map_size, false);
+ submap_min = map_range->min_address;
+ kmem_free(kernel_map, submap_min, zone_map_size);
- zone_page_table_second_level_size = 1;
- zone_page_table_second_level_shift_amount = 0;
-
+#if defined(__LP64__)
/*
- * Find the power of 2 for the second level that allows
- * the first level to fit in ZONE_PAGE_TABLE_FIRST_LEVEL_SIZE
- * slots.
+ * Allocate `Z_SUBMAP_IDX_VA_RESTRICTED_MAP` first because its VA range
+ * can't go beyond RESTRICTED_VA_MAX for the vm_page_t packing to work.
*/
- while ((zone_page_table_first_level_slot(zone_pages-1)) >= ZONE_PAGE_TABLE_FIRST_LEVEL_SIZE) {
- zone_page_table_second_level_size <<= 1;
- zone_page_table_second_level_shift_amount++;
- }
-
- lck_grp_attr_setdefault(&zone_gc_lck_grp_attr);
- lck_grp_init(&zone_gc_lck_grp, "zone_gc", &zone_gc_lck_grp_attr);
- lck_attr_setdefault(&zone_gc_lck_attr);
- lck_mtx_init_ext(&zone_gc_lock, &zone_gc_lck_ext, &zone_gc_lck_grp, &zone_gc_lck_attr);
-
-#if CONFIG_ZLEAKS
+ zone_submap_init(&submap_min, Z_SUBMAP_IDX_VA_RESTRICTED_MAP,
+ zone_sub_map_numer[Z_SUBMAP_IDX_VA_RESTRICTED_MAP], &remaining_denom,
+ &remaining_size, SINGLE_GUARD);
+#endif /* defined(__LP64__) */
+
/*
- * Initialize the zone leak monitor
+ * Allocate metadata array
*/
- zleak_init(max_zonemap_size);
-#endif /* CONFIG_ZLEAKS */
-}
+ zone_info.zi_meta_range =
+ zone_init_allocate_va(&submap_min, zone_meta_size, true);
+ zone_init_allocate_va(&submap_min, SINGLE_GUARD, true);
-void
-zone_page_table_expand(zone_page_index_t pindex)
-{
- unsigned int first_index;
- struct zone_page_table_entry * volatile * first_level_ptr;
+ zone_info.zi_array_base =
+ (struct zone_page_metadata *)zone_info.zi_meta_range.min_address -
+ zone_pva_from_addr(map_range->min_address).packed_address;
- assert(pindex < zone_pages);
+ /*
+ * Allocate other submaps
+ */
+ for (unsigned idx = Z_SUBMAP_IDX_GENERAL_MAP; idx <= zone_last_submap_idx; idx++) {
+ zone_submap_init(&submap_min, idx, zone_sub_map_numer[idx],
+ &remaining_denom, &remaining_size, MULTI_GUARD);
+ }
- first_index = zone_page_table_first_level_slot(pindex);
- first_level_ptr = &zone_page_table[first_index];
+ vm_map_t general_map = zone_submaps[Z_SUBMAP_IDX_GENERAL_MAP];
+ zone_info.zi_general_range.min_address = vm_map_min(general_map);
+ zone_info.zi_general_range.max_address = vm_map_max(general_map);
- if (*first_level_ptr == NULL) {
- /*
- * We were able to verify the old first-level slot
- * had NULL, so attempt to populate it.
- */
+ assert(submap_min == map_range->max_address);
- vm_offset_t second_level_array = 0;
- vm_size_t second_level_size = round_page(zone_page_table_second_level_size * sizeof(struct zone_page_table_entry));
- zone_page_index_t i;
- struct zone_page_table_entry *entry_array;
+#if CONFIG_GZALLOC
+ gzalloc_init(zone_map_size);
+#endif
- if (kmem_alloc_kobject(zone_map, &second_level_array,
- second_level_size, VM_KERN_MEMORY_OSFMK) != KERN_SUCCESS) {
- panic("zone_page_table_expand");
- }
- zone_map_table_page_count += (second_level_size / PAGE_SIZE);
+ zone_create_flags_t kma_flags = ZC_NOCACHING |
+ ZC_NOGC | ZC_NOENCRYPT | ZC_NOGZALLOC | ZC_NOCALLOUT |
+ ZC_KASAN_NOQUARANTINE | ZC_KASAN_NOREDZONE;
- /*
- * zone_gc() may scan the "zone_page_table" directly,
- * so make sure any slots have a valid unused state.
- */
- entry_array = (struct zone_page_table_entry *)second_level_array;
- for (i=0; i < zone_page_table_second_level_size; i++) {
- entry_array[i].alloc_count = ZONE_PAGE_UNUSED;
- entry_array[i].collect_count = 0;
- }
+ (void)zone_create_ext("vm.permanent", 1, kma_flags,
+ ZONE_ID_PERMANENT, ^(zone_t z){
+ z->permanent = true;
+ z->z_elem_size = 1;
+ z->pcpu_elem_size = 1;
+#if defined(__LP64__)
+ z->submap_idx = Z_SUBMAP_IDX_VA_RESTRICTED_MAP;
+#endif
+ });
+ (void)zone_create_ext("vm.permanent.percpu", 1, kma_flags | ZC_PERCPU,
+ ZONE_ID_PERCPU_PERMANENT, ^(zone_t z){
+ z->permanent = true;
+ z->z_elem_size = 1;
+ z->pcpu_elem_size = zpercpu_count();
+#if defined(__LP64__)
+ z->submap_idx = Z_SUBMAP_IDX_VA_RESTRICTED_MAP;
+#endif
+ });
- if (OSCompareAndSwapPtr(NULL, entry_array, first_level_ptr)) {
- /* Old slot was NULL, replaced with expanded level */
- OSAddAtomicLong(second_level_size, &zone_page_table_used_size);
- } else {
- /* Old slot was not NULL, someone else expanded first */
- kmem_free(zone_map, second_level_array, second_level_size);
- zone_map_table_page_count -= (second_level_size / PAGE_SIZE);
+ /*
+ * Now fix the zones that are missing their zone stats
+ * we don't really know if zfree()s happened so our stats
+ * are slightly off for early boot. ¯\_(ツ)_/¯
+ */
+ zone_index_foreach(idx) {
+ zone_t tz = &zone_array[idx];
+
+ if (tz->z_self) {
+ zone_stats_t zs = zalloc_percpu_permanent_type(struct zone_stats);
+
+ zpercpu_get_cpu(zs, 0)->zs_mem_allocated +=
+ (tz->countavail - tz->countfree) *
+ zone_elem_size(tz);
+ assert(tz->z_stats == NULL);
+ tz->z_stats = zs;
+#if ZONE_ENABLE_LOGGING
+ if (tz->zone_logging && !tz->zlog_btlog) {
+ zone_enable_logging(tz);
+ }
+#endif
}
- } else {
- /* Old slot was not NULL, already been expanded */
}
-}
-struct zone_page_table_entry *
-zone_page_table_lookup(zone_page_index_t pindex)
-{
- unsigned int first_index = zone_page_table_first_level_slot(pindex);
- struct zone_page_table_entry *second_level = zone_page_table[first_index];
+#if CONFIG_ZLEAKS
+ /*
+ * Initialize the zone leak monitor
+ */
+ zleak_init(zone_map_size);
+#endif /* CONFIG_ZLEAKS */
- if (second_level) {
- return &second_level[zone_page_table_second_level_slot(pindex)];
+#if VM_MAX_TAG_ZONES
+ if (zone_tagging_on) {
+ vm_allocation_zones_init();
}
+#endif
+}
+STARTUP(ZALLOC, STARTUP_RANK_FIRST, zone_init);
- return NULL;
+__startup_func
+static void
+zone_set_foreign_range(
+ vm_offset_t range_min,
+ vm_offset_t range_max)
+{
+ zone_info.zi_foreign_range.min_address = range_min;
+ zone_info.zi_foreign_range.max_address = range_max;
}
-extern volatile SInt32 kfree_nop_count;
+__startup_func
+vm_offset_t
+zone_foreign_mem_init(vm_size_t size)
+{
+ vm_offset_t mem = (vm_offset_t) pmap_steal_memory(size);
+ zone_set_foreign_range(mem, mem + size);
+ return mem;
+}
-#pragma mark -
-#pragma mark zalloc_canblock
+#pragma mark zalloc
+#if KASAN_ZALLOC
/*
- * zalloc returns an element from the specified zone.
- */
-static void *
-zalloc_internal(
- zone_t zone,
- boolean_t canblock,
- boolean_t nopagewait)
-{
- vm_offset_t addr = 0;
- kern_return_t retval;
- uintptr_t zbt[MAX_ZTRACE_DEPTH]; /* used in zone leak logging and zone leak detection */
- int numsaved = 0;
- boolean_t zone_replenish_wakeup = FALSE, zone_alloc_throttle = FALSE;
-#if CONFIG_GZALLOC || ZONE_DEBUG
- boolean_t did_gzalloc = FALSE;
-#endif
- thread_t thr = current_thread();
- boolean_t check_poison = FALSE;
- boolean_t set_doing_alloc_with_vm_priv = FALSE;
-
-#if CONFIG_ZLEAKS
- uint32_t zleak_tracedepth = 0; /* log this allocation if nonzero */
-#endif /* CONFIG_ZLEAKS */
-
- assert(zone != ZONE_NULL);
-
-#if CONFIG_GZALLOC
- addr = gzalloc_alloc(zone, canblock);
- did_gzalloc = (addr != 0);
-#endif
+ * Called from zfree() to add the element being freed to the KASan quarantine.
+ *
+ * Returns true if the newly-freed element made it into the quarantine without
+ * displacing another, false otherwise. In the latter case, addrp points to the
+ * address of the displaced element, which will be freed by the zone.
+ */
+static bool
+kasan_quarantine_freed_element(
+ zone_t *zonep, /* the zone the element is being freed to */
+ void **addrp) /* address of the element being freed */
+{
+ zone_t zone = *zonep;
+ void *addr = *addrp;
/*
- * If zone logging is turned on and this is the zone we're tracking, grab a backtrace.
+ * Resize back to the real allocation size and hand off to the KASan
+ * quarantine. `addr` may then point to a different allocation, if the
+ * current element replaced another in the quarantine. The zone then
+ * takes ownership of the swapped out free element.
*/
- if (__improbable(DO_LOGGING(zone)))
- numsaved = OSBacktrace((void*) zbt, MAX_ZTRACE_DEPTH);
+ vm_size_t usersz = zone_elem_size(zone) - 2 * zone->kasan_redzone;
+ vm_size_t sz = usersz;
-#if CONFIG_ZLEAKS
- /*
- * Zone leak detection: capture a backtrace every zleak_sample_factor
- * allocations in this zone.
- */
- if (__improbable(zone->zleak_on && sample_counter(&zone->zleak_capture, zleak_sample_factor) == TRUE)) {
- /* Avoid backtracing twice if zone logging is on */
- if (numsaved == 0)
- zleak_tracedepth = fastbacktrace(zbt, MAX_ZTRACE_DEPTH);
- else
- zleak_tracedepth = numsaved;
+ if (addr && zone->kasan_redzone) {
+ kasan_check_free((vm_address_t)addr, usersz, KASAN_HEAP_ZALLOC);
+ addr = (void *)kasan_dealloc((vm_address_t)addr, &sz);
+ assert(sz == zone_elem_size(zone));
}
-#endif /* CONFIG_ZLEAKS */
+ if (addr && !zone->kasan_noquarantine) {
+ kasan_free(&addr, &sz, KASAN_HEAP_ZALLOC, zonep, usersz, true);
+ if (!addr) {
+ return TRUE;
+ }
+ }
+ if (addr && zone->kasan_noquarantine) {
+ kasan_unpoison(addr, zone_elem_size(zone));
+ }
+ *addrp = addr;
+ return FALSE;
+}
- lock_zone(zone);
+#endif /* KASAN_ZALLOC */
+
+static inline bool
+zone_needs_async_refill(zone_t zone)
+{
+ if (zone->countfree != 0 || zone->async_pending || zone->no_callout) {
+ return false;
+ }
+
+ return zone->expandable || zone->page_count < zone->page_count_max;
+}
+
+__attribute__((noinline))
+static void
+zone_refill_synchronously_locked(
+ zone_t zone,
+ zalloc_flags_t flags)
+{
+ thread_t thr = current_thread();
+ bool set_expanding_vm_priv = false;
+ zone_pva_t orig = zone->pages_intermediate;
- if (zone->async_prio_refill && zone->zone_replenish_thread) {
- do {
- vm_size_t zfreec = (zone->cur_size - (zone->count * zone->elem_size));
- vm_size_t zrefillwm = zone->prio_refill_watermark * zone->elem_size;
- zone_replenish_wakeup = (zfreec < zrefillwm);
- zone_alloc_throttle = (zfreec < (zrefillwm / 2)) && ((thr->options & TH_OPT_VMPRIV) == 0);
-
- if (zone_replenish_wakeup) {
- zone_replenish_wakeups_initiated++;
- unlock_zone(zone);
- /* Signal the potentially waiting
- * refill thread.
- */
- thread_wakeup(&zone->zone_replenish_thread);
-
- /* Scheduling latencies etc. may prevent
- * the refill thread from keeping up
- * with demand. Throttle consumers
- * when we fall below half the
- * watermark, unless VM privileged
- */
- if (zone_alloc_throttle) {
- zone_replenish_throttle_count++;
- assert_wait_timeout(zone, THREAD_UNINT, 1, NSEC_PER_MSEC);
- thread_block(THREAD_CONTINUE_NULL);
- }
- lock_zone(zone);
- }
- } while (zone_alloc_throttle == TRUE);
- }
-
- if (__probable(addr == 0))
- addr = try_alloc_from_zone(zone, &check_poison);
-
-
- while ((addr == 0) && canblock) {
+ while ((flags & Z_NOWAIT) == 0 && (zone->permanent
+ ? zone_pva_is_equal(zone->pages_intermediate, orig)
+ : zone->countfree == 0)) {
/*
- * zone is empty, try to expand it
- *
- * Note that we now allow up to 2 threads (1 vm_privliged and 1 non-vm_privliged)
- * to expand the zone concurrently... this is necessary to avoid stalling
- * vm_privileged threads running critical code necessary to continue compressing/swapping
- * pages (i.e. making new free pages) from stalling behind non-vm_privileged threads
- * waiting to acquire free pages when the vm_page_free_count is below the
+ * zone is empty, try to expand it
+ *
+ * Note that we now allow up to 2 threads (1 vm_privliged and
+ * 1 non-vm_privliged) to expand the zone concurrently...
+ *
+ * this is necessary to avoid stalling vm_privileged threads
+ * running critical code necessary to continue
+ * compressing/swapping pages (i.e. making new free pages) from
+ * stalling behind non-vm_privileged threads waiting to acquire
+ * free pages when the vm_page_free_count is below the
* vm_page_free_reserved limit.
*/
- if ((zone->doing_alloc_without_vm_priv || zone->doing_alloc_with_vm_priv) &&
- (((thr->options & TH_OPT_VMPRIV) == 0) || zone->doing_alloc_with_vm_priv)) {
+ if ((zone->expanding_no_vm_priv || zone->expanding_vm_priv) &&
+ (((thr->options & TH_OPT_VMPRIV) == 0) || zone->expanding_vm_priv)) {
/*
* This is a non-vm_privileged thread and a non-vm_privileged or
* a vm_privileged thread is already expanding the zone...
*
* In either case wait for a thread to finish, then try again.
*/
- zone->waiting = TRUE;
- zone_sleep(zone);
- } else if (zone->doing_gc) {
- /*
- * zone_gc() is running. Since we need an element
- * from the free list that is currently being
- * collected, set the waiting bit and
- * wait for the GC process to finish
- * before trying again
- */
- zone->waiting = TRUE;
- zone_sleep(zone);
- } else {
- vm_offset_t space;
- vm_size_t alloc_size;
- int retry = 0;
-
- if ((zone->cur_size + zone->elem_size) >
- zone->max_size) {
- if (zone->exhaustible)
- break;
- if (zone->expandable) {
- /*
- * We're willing to overflow certain
- * zones, but not without complaining.
- *
- * This is best used in conjunction
- * with the collectable flag. What we
- * want is an assurance we can get the
- * memory back, assuming there's no
- * leak.
- */
- zone->max_size += (zone->max_size >> 1);
- } else {
- unlock_zone(zone);
-
- panic_include_zprint = TRUE;
-#if CONFIG_ZLEAKS
- if (zleak_state & ZLEAK_STATE_ACTIVE)
- panic_include_ztrace = TRUE;
-#endif /* CONFIG_ZLEAKS */
- panic("zalloc: zone \"%s\" empty.", zone->zone_name);
- }
+ zone->waiting = true;
+ assert_wait(zone, THREAD_UNINT);
+ unlock_zone(zone);
+ thread_block(THREAD_CONTINUE_NULL);
+ lock_zone(zone);
+ continue;
+ }
+
+ if (zone->page_count >= zone->page_count_max) {
+ if (zone->exhaustible) {
+ break;
}
- if ((thr->options & TH_OPT_VMPRIV)) {
- zone->doing_alloc_with_vm_priv = TRUE;
- set_doing_alloc_with_vm_priv = TRUE;
+ if (zone->expandable) {
+ /*
+ * If we're expandable, just don't go through this again.
+ */
+ zone->page_count_max = ~0u;
} else {
- zone->doing_alloc_without_vm_priv = TRUE;
- }
- unlock_zone(zone);
+ unlock_zone(zone);
- for (;;) {
- int zflags = KMA_KOBJECT|KMA_NOPAGEWAIT;
-
- if (vm_pool_low() || retry >= 1)
- alloc_size =
- round_page(zone->elem_size);
- else
- alloc_size = zone->alloc_size;
-
- if (zone->noencrypt)
- zflags |= KMA_NOENCRYPT;
-
- retval = kernel_memory_allocate(zone_map, &space, alloc_size, 0, zflags, VM_KERN_MEMORY_ZONE);
- if (retval == KERN_SUCCESS) {
-#if ZONE_ALIAS_ADDR
- if (alloc_size == PAGE_SIZE)
- space = zone_alias_addr(space);
-#endif
-
+ panic_include_zprint = true;
#if CONFIG_ZLEAKS
- if ((zleak_state & (ZLEAK_STATE_ENABLED | ZLEAK_STATE_ACTIVE)) == ZLEAK_STATE_ENABLED) {
- if (zone_map->size >= zleak_global_tracking_threshold) {
- kern_return_t kr;
-
- kr = zleak_activate();
- if (kr != KERN_SUCCESS) {
- printf("Failed to activate live zone leak debugging (%d).\n", kr);
- }
- }
- }
-
- if ((zleak_state & ZLEAK_STATE_ACTIVE) && !(zone->zleak_on)) {
- if (zone->cur_size > zleak_per_zone_tracking_threshold) {
- zone->zleak_on = TRUE;
- }
- }
-#endif /* CONFIG_ZLEAKS */
- zcram(zone, space, alloc_size);
-
- break;
- } else if (retval != KERN_RESOURCE_SHORTAGE) {
- retry++;
-
- if (retry == 2) {
- zone_gc(TRUE);
- printf("zalloc did gc\n");
- zone_display_zprint();
- }
- if (retry == 3) {
- panic_include_zprint = TRUE;
-#if CONFIG_ZLEAKS
- if ((zleak_state & ZLEAK_STATE_ACTIVE)) {
- panic_include_ztrace = TRUE;
- }
-#endif /* CONFIG_ZLEAKS */
- if (retval == KERN_NO_SPACE) {
- zone_t zone_largest = zone_find_largest();
- panic("zalloc: zone map exhausted while allocating from zone %s, likely due to memory leak in zone %s (%lu total bytes, %d elements allocated)",
- zone->zone_name, zone_largest->zone_name,
- (unsigned long)zone_largest->cur_size, zone_largest->count);
-
- }
- panic("zalloc: \"%s\" (%d elements) retry fail %d, kfree_nop_count: %d", zone->zone_name, zone->count, retval, (int)kfree_nop_count);
- }
- } else {
- break;
+ if (zleak_state & ZLEAK_STATE_ACTIVE) {
+ panic_include_ztrace = true;
}
+#endif /* CONFIG_ZLEAKS */
+ panic("zalloc: zone \"%s\" empty.", zone->z_name);
}
- lock_zone(zone);
+ }
- if (set_doing_alloc_with_vm_priv == TRUE)
- zone->doing_alloc_with_vm_priv = FALSE;
- else
- zone->doing_alloc_without_vm_priv = FALSE;
-
- if (zone->waiting) {
- zone->waiting = FALSE;
- zone_wakeup(zone);
- }
- addr = try_alloc_from_zone(zone, &check_poison);
- if (addr == 0 &&
- retval == KERN_RESOURCE_SHORTAGE) {
- if (nopagewait == TRUE)
- break; /* out of the main while loop */
- unlock_zone(zone);
+ /*
+ * It is possible that a BG thread is refilling/expanding the zone
+ * and gets pre-empted during that operation. That blocks all other
+ * threads from making progress leading to a watchdog timeout. To
+ * avoid that, boost the thread priority using the rwlock boost
+ */
+ set_thread_rwlock_boost();
- VM_PAGE_WAIT();
- lock_zone(zone);
- }
+ if ((thr->options & TH_OPT_VMPRIV)) {
+ zone->expanding_vm_priv = true;
+ set_expanding_vm_priv = true;
+ } else {
+ zone->expanding_no_vm_priv = true;
+ }
+
+ zone_replenish_locked(zone, flags, false);
+
+ if (set_expanding_vm_priv == true) {
+ zone->expanding_vm_priv = false;
+ } else {
+ zone->expanding_no_vm_priv = false;
+ }
+
+ if (zone->waiting) {
+ zone->waiting = false;
+ thread_wakeup(zone);
+ }
+ clear_thread_rwlock_boost();
+
+ if (zone->countfree == 0) {
+ assert(flags & Z_NOPAGEWAIT);
+ break;
}
- if (addr == 0)
- addr = try_alloc_from_zone(zone, &check_poison);
}
-#if CONFIG_ZLEAKS
- /* Zone leak detection:
- * If we're sampling this allocation, add it to the zleaks hash table.
- */
- if (addr && zleak_tracedepth > 0) {
- /* Sampling can fail if another sample is happening at the same time in a different zone. */
- if (!zleak_log(zbt, addr, zleak_tracedepth, zone->elem_size)) {
- /* If it failed, roll back the counter so we sample the next allocation instead. */
- zone->zleak_capture = zleak_sample_factor;
- }
- }
-#endif /* CONFIG_ZLEAKS */
-
-
- if ((addr == 0) && (!canblock || nopagewait) && (zone->async_pending == FALSE) && (zone->no_callout == FALSE) && (zone->exhaustible == FALSE) && (!vm_pool_low())) {
- zone->async_pending = TRUE;
+ if ((flags & (Z_NOWAIT | Z_NOPAGEWAIT)) &&
+ zone_needs_async_refill(zone) && !vm_pool_low()) {
+ zone->async_pending = true;
unlock_zone(zone);
thread_call_enter(&call_async_alloc);
lock_zone(zone);
- addr = try_alloc_from_zone(zone, &check_poison);
+ assert(zone->z_self == zone);
}
+}
+
+__attribute__((noinline))
+static void
+zone_refill_asynchronously_locked(zone_t zone)
+{
+ uint32_t min_free = zone->prio_refill_count / 2;
+ uint32_t resv_free = zone->prio_refill_count / 4;
+ thread_t thr = current_thread();
/*
- * See if we should be logging allocations in this zone. Logging is rarely done except when a leak is
- * suspected, so this code rarely executes. We need to do this code while still holding the zone lock
- * since it protects the various log related data structures.
+ * Nothing to do if there are plenty of elements.
*/
+ while (zone->countfree <= min_free) {
+ /*
+ * Wakeup the replenish thread if not running.
+ */
+ if (!zone->zone_replenishing) {
+ lck_spin_lock(&zone_replenish_lock);
+ assert(zone_replenish_active < zone_replenish_max_threads);
+ ++zone_replenish_active;
+ lck_spin_unlock(&zone_replenish_lock);
+ zone->zone_replenishing = true;
+ zone_replenish_wakeups_initiated++;
+ thread_wakeup(&zone->prio_refill_count);
+ }
- if (__improbable(DO_LOGGING(zone) && addr)) {
- btlog_add_entry(zlog_btlog, (void *)addr, ZOP_ALLOC, (void **)zbt, numsaved);
- }
-
- vm_offset_t inner_size = zone->elem_size;
-
-#if ZONE_DEBUG
- if (!did_gzalloc && addr && zone_debug_enabled(zone)) {
- enqueue_tail(&zone->active_zones, (queue_entry_t)addr);
- addr += ZONE_DEBUG_OFFSET;
- inner_size -= ZONE_DEBUG_OFFSET;
- }
-#endif
-
- unlock_zone(zone);
-
- if (__improbable(check_poison && addr)) {
- vm_offset_t *element_cursor = ((vm_offset_t *) addr) + 1;
- vm_offset_t *backup = get_backup_ptr(inner_size, (vm_offset_t *) addr);
-
- for ( ; element_cursor < backup ; element_cursor++)
- if (__improbable(*element_cursor != ZP_POISON))
- zone_element_was_modified_panic(zone,
- addr,
- *element_cursor,
- ZP_POISON,
- ((vm_offset_t)element_cursor) - addr);
- }
-
- if (addr) {
/*
- * Clear out the old next pointer and backup to avoid leaking the cookie
- * and so that only values on the freelist have a valid cookie
+ * We'll let VM_PRIV threads to continue to allocate until the
+ * reserve drops to 25%. After that only TH_OPT_ZONE_PRIV threads
+ * may continue.
+ *
+ * TH_OPT_ZONE_PRIV threads are the GC thread and a replenish thread itself.
+ * Replenish threads *need* to use the reserve. GC threads need to
+ * get through the current allocation, but then will wait at a higher
+ * level after they've dropped any locks which would deadlock the
+ * replenish thread.
*/
+ if ((zone->countfree > resv_free && (thr->options & TH_OPT_VMPRIV)) ||
+ (thr->options & TH_OPT_ZONE_PRIV)) {
+ break;
+ }
- vm_offset_t *primary = (vm_offset_t *) addr;
- vm_offset_t *backup = get_backup_ptr(inner_size, primary);
+ /*
+ * Wait for the replenish threads to add more elements for us to allocate from.
+ */
+ zone_replenish_throttle_count++;
+ unlock_zone(zone);
+ assert_wait_timeout(zone, THREAD_UNINT, 1, NSEC_PER_MSEC);
+ thread_block(THREAD_CONTINUE_NULL);
+ lock_zone(zone);
- *primary = ZP_POISON;
- *backup = ZP_POISON;
+ assert(zone->z_self == zone);
}
- TRACE_MACHLEAKS(ZALLOC_CODE, ZALLOC_CODE_2, zone->elem_size, addr);
-
- if (addr) {
- task_t task;
- zinfo_usage_t zinfo;
- vm_size_t sz = zone->elem_size;
-
- if (zone->caller_acct)
- ledger_credit(thr->t_ledger, task_ledgers.tkm_private, sz);
- else
- ledger_credit(thr->t_ledger, task_ledgers.tkm_shared, sz);
-
- if ((task = thr->task) != NULL && (zinfo = task->tkm_zinfo) != NULL)
- OSAddAtomic64(sz, (int64_t *)&zinfo[zone->index].alloc);
+ /*
+ * If we're here because of zone_gc(), we didn't wait for
+ * zone_replenish_thread to finish. So we need to ensure that
+ * we will successfully grab an element.
+ *
+ * zones that have a replenish thread configured.
+ * The value of (refill_level / 2) in the previous bit of code should have
+ * given us headroom even though this thread didn't wait.
+ */
+ if (thr->options & TH_OPT_ZONE_PRIV) {
+ assert(zone->countfree != 0);
}
- return((void *)addr);
-}
-
-
-void *
-zalloc(zone_t zone)
-{
- return (zalloc_internal(zone, TRUE, FALSE));
-}
-
-void *
-zalloc_noblock(zone_t zone)
-{
- return (zalloc_internal(zone, FALSE, FALSE));
-}
-
-void *
-zalloc_nopagewait(zone_t zone)
-{
- return (zalloc_internal(zone, TRUE, TRUE));
}
-void *
-zalloc_canblock(zone_t zone, boolean_t canblock)
+#if ZONE_ENABLE_LOGGING || CONFIG_ZLEAKS
+__attribute__((noinline))
+static void
+zalloc_log_or_trace_leaks(zone_t zone, vm_offset_t addr)
{
- return (zalloc_internal(zone, canblock, FALSE));
-}
-
+ uintptr_t zbt[MAX_ZTRACE_DEPTH]; /* used in zone leak logging and zone leak detection */
+ unsigned int numsaved = 0;
+
+#if ZONE_ENABLE_LOGGING
+ if (DO_LOGGING(zone)) {
+ numsaved = backtrace_frame(zbt, MAX_ZTRACE_DEPTH,
+ __builtin_frame_address(0), NULL);
+ btlog_add_entry(zone->zlog_btlog, (void *)addr,
+ ZOP_ALLOC, (void **)zbt, numsaved);
+ }
+#endif
-void
-zalloc_async(
- __unused thread_call_param_t p0,
- __unused thread_call_param_t p1)
-{
- zone_t current_z = NULL, head_z;
- unsigned int max_zones, i;
- void *elt = NULL;
- boolean_t pending = FALSE;
-
- simple_lock(&all_zones_lock);
- head_z = first_zone;
- max_zones = num_zones;
- simple_unlock(&all_zones_lock);
- current_z = head_z;
- for (i = 0; i < max_zones; i++) {
- lock_zone(current_z);
- if (current_z->async_pending == TRUE) {
- current_z->async_pending = FALSE;
- pending = TRUE;
+#if CONFIG_ZLEAKS
+ /*
+ * Zone leak detection: capture a backtrace every zleak_sample_factor
+ * allocations in this zone.
+ */
+ if (__improbable(zone->zleak_on)) {
+ if (sample_counter(&zone->zleak_capture, zleak_sample_factor)) {
+ /* Avoid backtracing twice if zone logging is on */
+ if (numsaved == 0) {
+ numsaved = backtrace_frame(zbt, MAX_ZTRACE_DEPTH,
+ __builtin_frame_address(1), NULL);
+ }
+ /* Sampling can fail if another sample is happening at the same time in a different zone. */
+ if (!zleak_log(zbt, addr, numsaved, zone_elem_size(zone))) {
+ /* If it failed, roll back the counter so we sample the next allocation instead. */
+ zone->zleak_capture = zleak_sample_factor;
+ }
}
- unlock_zone(current_z);
+ }
- if (pending == TRUE) {
- elt = zalloc_canblock(current_z, TRUE);
- zfree(current_z, elt);
- pending = FALSE;
+ if (__improbable(zone_leaks_scan_enable &&
+ !(zone_elem_size(zone) & (sizeof(uintptr_t) - 1)))) {
+ unsigned int count, idx;
+ /* Fill element, from tail, with backtrace in reverse order */
+ if (numsaved == 0) {
+ numsaved = backtrace_frame(zbt, MAX_ZTRACE_DEPTH,
+ __builtin_frame_address(1), NULL);
+ }
+ count = (unsigned int)(zone_elem_size(zone) / sizeof(uintptr_t));
+ if (count >= numsaved) {
+ count = numsaved - 1;
+ }
+ for (idx = 0; idx < count; idx++) {
+ ((uintptr_t *)addr)[count - 1 - idx] = zbt[idx + 1];
}
- /*
- * This is based on assumption that zones never get
- * freed once allocated and linked.
- * Hence a read outside of lock is OK.
- */
- current_z = current_z->next_zone;
}
+#endif /* CONFIG_ZLEAKS */
}
-/*
- * zget returns an element from the specified zone
- * and immediately returns nothing if there is nothing there.
- *
- * This form should be used when you can not block (like when
- * processing an interrupt).
- *
- * XXX: It seems like only vm_page_grab_fictitious_common uses this, and its
- * friend vm_page_more_fictitious can block, so it doesn't seem like
- * this is used for interrupts any more....
- */
-void *
-zget(
- register zone_t zone)
+static inline bool
+zalloc_should_log_or_trace_leaks(zone_t zone, vm_size_t elem_size)
{
- vm_offset_t addr;
- boolean_t check_poison = FALSE;
-
-#if CONFIG_ZLEAKS
- uintptr_t zbt[MAX_ZTRACE_DEPTH]; /* used for zone leak detection */
- uint32_t zleak_tracedepth = 0; /* log this allocation if nonzero */
-#endif /* CONFIG_ZLEAKS */
-
- assert( zone != ZONE_NULL );
-
+#if ZONE_ENABLE_LOGGING
+ if (DO_LOGGING(zone)) {
+ return true;
+ }
+#endif
#if CONFIG_ZLEAKS
/*
- * Zone leak detection: capture a backtrace
+ * Zone leak detection: capture a backtrace every zleak_sample_factor
+ * allocations in this zone.
*/
- if (__improbable(zone->zleak_on && sample_counter(&zone->zleak_capture, zleak_sample_factor) == TRUE)) {
- zleak_tracedepth = fastbacktrace(zbt, MAX_ZTRACE_DEPTH);
+ if (zone->zleak_on) {
+ return true;
+ }
+ if (zone_leaks_scan_enable && !(elem_size & (sizeof(uintptr_t) - 1))) {
+ return true;
}
#endif /* CONFIG_ZLEAKS */
+ return false;
+}
+#endif /* ZONE_ENABLE_LOGGING || CONFIG_ZLEAKS */
+#if ZONE_ENABLE_LOGGING
- if (!lock_try_zone(zone))
- return NULL;
-
- addr = try_alloc_from_zone(zone, &check_poison);
-
- vm_offset_t inner_size = zone->elem_size;
-
-#if ZONE_DEBUG
- if (addr && zone_debug_enabled(zone)) {
- enqueue_tail(&zone->active_zones, (queue_entry_t)addr);
- addr += ZONE_DEBUG_OFFSET;
- inner_size -= ZONE_DEBUG_OFFSET;
- }
-#endif /* ZONE_DEBUG */
-
-#if CONFIG_ZLEAKS
+__attribute__((noinline))
+static void
+zfree_log_trace(zone_t zone, vm_offset_t addr)
+{
/*
- * Zone leak detection: record the allocation
+ * See if we're doing logging on this zone.
+ *
+ * There are two styles of logging used depending on
+ * whether we're trying to catch a leak or corruption.
*/
- if (zone->zleak_on && zleak_tracedepth > 0 && addr) {
- /* Sampling can fail if another sample is happening at the same time in a different zone. */
- if (!zleak_log(zbt, addr, zleak_tracedepth, zone->elem_size)) {
- /* If it failed, roll back the counter so we sample the next allocation instead. */
- zone->zleak_capture = zleak_sample_factor;
+ if (__improbable(DO_LOGGING(zone))) {
+ if (corruption_debug_flag) {
+ uintptr_t zbt[MAX_ZTRACE_DEPTH];
+ unsigned int numsaved;
+ /*
+ * We're logging to catch a corruption.
+ *
+ * Add a record of this zfree operation to log.
+ */
+ numsaved = backtrace_frame(zbt, MAX_ZTRACE_DEPTH,
+ __builtin_frame_address(1), NULL);
+ btlog_add_entry(zone->zlog_btlog, (void *)addr, ZOP_FREE,
+ (void **)zbt, numsaved);
+ } else {
+ /*
+ * We're logging to catch a leak.
+ *
+ * Remove any record we might have for this element
+ * since it's being freed. Note that we may not find it
+ * if the buffer overflowed and that's OK.
+ *
+ * Since the log is of a limited size, old records get
+ * overwritten if there are more zallocs than zfrees.
+ */
+ btlog_remove_entries_for_element(zone->zlog_btlog, (void *)addr);
}
}
-#endif /* CONFIG_ZLEAKS */
-
- unlock_zone(zone);
-
- if (__improbable(check_poison && addr)) {
- vm_offset_t *element_cursor = ((vm_offset_t *) addr) + 1;
- vm_offset_t *backup = get_backup_ptr(inner_size, (vm_offset_t *) addr);
-
- for ( ; element_cursor < backup ; element_cursor++)
- if (__improbable(*element_cursor != ZP_POISON))
- zone_element_was_modified_panic(zone,
- addr,
- *element_cursor,
- ZP_POISON,
- ((vm_offset_t)element_cursor) - addr);
- }
-
- if (addr) {
- /*
- * Clear out the old next pointer and backup to avoid leaking the cookie
- * and so that only values on the freelist have a valid cookie
- */
- vm_offset_t *primary = (vm_offset_t *) addr;
- vm_offset_t *backup = get_backup_ptr(inner_size, primary);
-
- *primary = ZP_POISON;
- *backup = ZP_POISON;
- }
-
- return((void *) addr);
}
+#endif /* ZONE_ENABLE_LOGGING */
-/* Keep this FALSE by default. Large memory machine run orders of magnitude
- slower in debug mode when true. Use debugger to enable if needed */
-/* static */ boolean_t zone_check = FALSE;
-
-static void zone_check_freelist(zone_t zone, vm_offset_t elem)
+/*
+ * Removes an element from the zone's free list, returning 0 if the free list is empty.
+ * Verifies that the next-pointer and backup next-pointer are intact,
+ * and verifies that a poisoned element hasn't been modified.
+ */
+vm_offset_t
+zalloc_direct_locked(
+ zone_t zone,
+ zalloc_flags_t flags __unused,
+ vm_size_t waste __unused)
{
- struct zone_free_element *this;
- struct zone_page_metadata *thispage;
+ struct zone_page_metadata *page_meta;
+ zone_addr_kind_t kind = ZONE_ADDR_NATIVE;
+ vm_offset_t element, page, validate_bit = 0;
- if (zone->use_page_list) {
- if (zone->allows_foreign) {
- for (thispage = (struct zone_page_metadata *)queue_first(&zone->pages.any_free_foreign);
- !queue_end(&zone->pages.any_free_foreign, (queue_entry_t)thispage);
- thispage = (struct zone_page_metadata *)queue_next((queue_chain_t *)thispage)) {
- for (this = thispage->elements;
- this != NULL;
- this = this->next) {
- if (!is_sane_zone_element(zone, (vm_address_t)this) || (vm_address_t)this == elem)
- panic("zone_check_freelist");
- }
- }
- }
- for (thispage = (struct zone_page_metadata *)queue_first(&zone->pages.all_free);
- !queue_end(&zone->pages.all_free, (queue_entry_t)thispage);
- thispage = (struct zone_page_metadata *)queue_next((queue_chain_t *)thispage)) {
- for (this = thispage->elements;
- this != NULL;
- this = this->next) {
- if (!is_sane_zone_element(zone, (vm_address_t)this) || (vm_address_t)this == elem)
- panic("zone_check_freelist");
- }
- }
- for (thispage = (struct zone_page_metadata *)queue_first(&zone->pages.intermediate);
- !queue_end(&zone->pages.intermediate, (queue_entry_t)thispage);
- thispage = (struct zone_page_metadata *)queue_next((queue_chain_t *)thispage)) {
- for (this = thispage->elements;
- this != NULL;
- this = this->next) {
- if (!is_sane_zone_element(zone, (vm_address_t)this) || (vm_address_t)this == elem)
- panic("zone_check_freelist");
- }
+ /* if zone is empty, bail */
+ if (!zone_pva_is_null(zone->pages_any_free_foreign)) {
+ kind = ZONE_ADDR_FOREIGN;
+ page_meta = zone_pva_to_meta(zone->pages_any_free_foreign, kind);
+ page = (vm_offset_t)page_meta;
+ } else if (!zone_pva_is_null(zone->pages_intermediate)) {
+ page_meta = zone_pva_to_meta(zone->pages_intermediate, kind);
+ page = zone_pva_to_addr(zone->pages_intermediate);
+ } else if (!zone_pva_is_null(zone->pages_all_free)) {
+ page_meta = zone_pva_to_meta(zone->pages_all_free, kind);
+ page = zone_pva_to_addr(zone->pages_all_free);
+ if (os_sub_overflow(zone->allfree_page_count,
+ page_meta->zm_page_count, &zone->allfree_page_count)) {
+ zone_accounting_panic(zone, "allfree_page_count wrap-around");
}
} else {
- for (this = zone->free_elements;
- this != NULL;
- this = this->next) {
- if (!is_sane_zone_element(zone, (vm_address_t)this) || (vm_address_t)this == elem)
- panic("zone_check_freelist");
- }
+ zone_accounting_panic(zone, "countfree corruption");
}
-}
-
-static zone_t zone_last_bogus_zone = ZONE_NULL;
-static vm_offset_t zone_last_bogus_elem = 0;
-void
-zfree(
- register zone_t zone,
- void *addr)
-{
- vm_offset_t elem = (vm_offset_t) addr;
- uintptr_t zbt[MAX_ZTRACE_DEPTH]; /* only used if zone logging is enabled via boot-args */
- int numsaved = 0;
- boolean_t gzfreed = FALSE;
- boolean_t poison = FALSE;
-
- assert(zone != ZONE_NULL);
-
-#if 1
- if (zone->use_page_list) {
- struct zone_page_metadata *page_meta = get_zone_page_metadata((struct zone_free_element *)addr);
- if (zone != page_meta->zone) {
- /*
- * Something bad has happened. Someone tried to zfree a pointer but the metadata says it is from
- * a different zone (or maybe it's from a zone that doesn't use page free lists at all). We can repair
- * some cases of this, if:
- * 1) The specified zone had use_page_list, and the true zone also has use_page_list set. In that case
- * we can swap the zone_t
- * 2) The specified zone had use_page_list, but the true zone does not. In this case page_meta is garbage,
- * and dereferencing page_meta->zone might panic.
- * To distinguish the two, we enumerate the zone list to match it up.
- * We do not handle the case where an incorrect zone is passed that does not have use_page_list set,
- * even if the true zone did have this set.
- */
- zone_t fixed_zone = NULL;
- int fixed_i, max_zones;
-
- simple_lock(&all_zones_lock);
- max_zones = num_zones;
- fixed_zone = first_zone;
- simple_unlock(&all_zones_lock);
-
- for (fixed_i=0; fixed_i < max_zones; fixed_i++, fixed_zone = fixed_zone->next_zone) {
- if (fixed_zone == page_meta->zone && fixed_zone->use_page_list) {
- /* we can fix this */
- printf("Fixing incorrect zfree from zone %s to zone %s\n", zone->zone_name, fixed_zone->zone_name);
- zone = fixed_zone;
- break;
- }
- }
- }
+ if (!zone_has_index(zone, page_meta->zm_index)) {
+ zone_page_metadata_index_confusion_panic(zone, page, page_meta);
}
-#endif
+
+ element = zone_page_meta_get_freelist(zone, page_meta, page);
+
+ vm_offset_t *primary = (vm_offset_t *) element;
+ vm_offset_t *backup = get_backup_ptr(zone_elem_size(zone), primary);
/*
- * If zone logging is turned on and this is the zone we're tracking, grab a backtrace.
+ * since the primary next pointer is xor'ed with zp_nopoison_cookie
+ * for obfuscation, retrieve the original value back
*/
+ vm_offset_t next_element = *primary ^ zp_nopoison_cookie;
+ vm_offset_t next_element_primary = *primary;
+ vm_offset_t next_element_backup = *backup;
- if (__improbable(DO_LOGGING(zone) && corruption_debug_flag))
- numsaved = OSBacktrace((void *)zbt, MAX_ZTRACE_DEPTH);
+ /*
+ * backup_ptr_mismatch_panic will determine what next_element
+ * should have been, and print it appropriately
+ */
+ if (!zone_page_meta_is_sane_element(zone, page_meta, page, next_element, kind)) {
+ backup_ptr_mismatch_panic(zone, page_meta, page, element);
+ }
-#if MACH_ASSERT
- /* Basic sanity checks */
- if (zone == ZONE_NULL || elem == (vm_offset_t)0)
- panic("zfree: NULL");
- /* zone_gc assumes zones are never freed */
- if (zone == zone_zone)
- panic("zfree: freeing to zone_zone breaks zone_gc!");
-#endif
+ /* Check the backup pointer for the regular cookie */
+ if (__improbable(next_element_primary != next_element_backup)) {
+ /* Check for the poisoned cookie instead */
+ if (__improbable(next_element != (next_element_backup ^ zp_poisoned_cookie))) {
+ /* Neither cookie is valid, corruption has occurred */
+ backup_ptr_mismatch_panic(zone, page_meta, page, element);
+ }
-#if CONFIG_GZALLOC
- gzfreed = gzalloc_free(zone, addr);
-#endif
+ /*
+ * Element was marked as poisoned, so check its integrity before using it.
+ */
+ validate_bit = ZALLOC_ELEMENT_NEEDS_VALIDATION;
+ } else if (zone->zfree_clear_mem) {
+ validate_bit = ZALLOC_ELEMENT_NEEDS_VALIDATION;
+ }
- TRACE_MACHLEAKS(ZFREE_CODE, ZFREE_CODE_2, zone->elem_size, (uintptr_t)addr);
+ /* Remove this element from the free list */
+ zone_page_meta_set_freelist(page_meta, page, next_element);
- if (__improbable(!gzfreed && zone->collectable && !zone->allows_foreign &&
- !from_zone_map(elem, zone->elem_size))) {
-#if MACH_ASSERT
- panic("zfree: non-allocated memory in collectable zone!");
-#endif
- zone_last_bogus_zone = zone;
- zone_last_bogus_elem = elem;
- return;
+ if (kind == ZONE_ADDR_FOREIGN) {
+ if (next_element == 0) {
+ /* last foreign element allocated on page, move to all_used_foreign */
+ zone_meta_requeue(zone, &zone->pages_all_used_foreign, page_meta, kind);
+ }
+ } else if (next_element == 0) {
+ zone_meta_requeue(zone, &zone->pages_all_used, page_meta, kind);
+ } else if (page_meta->zm_alloc_count == 0) {
+ /* remove from free, move to intermediate */
+ zone_meta_requeue(zone, &zone->pages_intermediate, page_meta, kind);
}
- if ((zp_factor != 0 || zp_tiny_zone_limit != 0) && !gzfreed) {
+ if (os_add_overflow(page_meta->zm_alloc_count, 1,
+ &page_meta->zm_alloc_count)) {
/*
- * Poison the memory before it ends up on the freelist to catch
- * use-after-free and use of uninitialized memory
+ * This will not catch a lot of errors, the proper check
+ * would be against the number of elements this run should
+ * have which is expensive to count.
*
- * Always poison tiny zones' elements (limit is 0 if -no-zp is set)
- * Also poison larger elements periodically
+ * But zm_alloc_count is a 16 bit number which could
+ * theoretically be valuable to cause to wrap around,
+ * so catch this.
*/
+ zone_page_meta_accounting_panic(zone, page_meta,
+ "zm_alloc_count overflow");
+ }
+ if (os_sub_overflow(zone->countfree, 1, &zone->countfree)) {
+ zone_accounting_panic(zone, "countfree wrap-around");
+ }
- vm_offset_t inner_size = zone->elem_size;
-
-#if ZONE_DEBUG
- if (!gzfreed && zone_debug_enabled(zone)) {
- inner_size -= ZONE_DEBUG_OFFSET;
+#if VM_MAX_TAG_ZONES
+ if (__improbable(zone->tags)) {
+ vm_tag_t tag = zalloc_flags_get_tag(flags);
+ // set the tag with b0 clear so the block remains inuse
+ ZTAG(zone, element)[0] = (vm_tag_t)(tag << 1);
+ vm_tag_update_zone_size(tag, zone->tag_zone_index,
+ zone_elem_size(zone), waste);
+ }
+#endif /* VM_MAX_TAG_ZONES */
+#if KASAN_ZALLOC
+ if (zone->percpu) {
+ zpercpu_foreach_cpu(i) {
+ kasan_poison_range(element + ptoa(i),
+ zone_elem_size(zone), ASAN_VALID);
}
+ } else {
+ kasan_poison_range(element, zone_elem_size(zone), ASAN_VALID);
+ }
+#endif
+
+ return element | validate_bit;
+}
+
+/*
+ * zalloc returns an element from the specified zone.
+ *
+ * The function is noinline when zlog can be used so that the backtracing can
+ * reliably skip the zalloc_ext() and zalloc_log_or_trace_leaks()
+ * boring frames.
+ */
+#if ZONE_ENABLE_LOGGING
+__attribute__((noinline))
#endif
- uint32_t sample_factor = zp_factor + (((uint32_t)inner_size) >> zp_scale);
+void *
+zalloc_ext(
+ zone_t zone,
+ zone_stats_t zstats,
+ zalloc_flags_t flags,
+ vm_size_t waste)
+{
+ vm_offset_t addr = 0;
+ vm_size_t elem_size = zone_elem_size(zone);
- if (inner_size <= zp_tiny_zone_limit)
- poison = TRUE;
- else if (zp_factor != 0 && sample_counter(&zone->zp_count, sample_factor) == TRUE)
- poison = TRUE;
+ /*
+ * KASan uses zalloc() for fakestack, which can be called anywhere.
+ * However, we make sure these calls can never block.
+ */
+ assert(zone->kasan_fakestacks ||
+ ml_get_interrupts_enabled() ||
+ ml_is_quiescing() ||
+ debug_mode_active() ||
+ startup_phase < STARTUP_SUB_EARLY_BOOT);
- if (__improbable(poison)) {
+ /*
+ * Make sure Z_NOFAIL was not obviously misused
+ */
+ if ((flags & Z_NOFAIL) && !zone->prio_refill_count) {
+ assert(!zone->exhaustible && (flags & (Z_NOWAIT | Z_NOPAGEWAIT)) == 0);
+ }
- /* memset_pattern{4|8} could help make this faster: <rdar://problem/4662004> */
- /* Poison everything but primary and backup */
- vm_offset_t *element_cursor = ((vm_offset_t *) elem) + 1;
- vm_offset_t *backup = get_backup_ptr(inner_size, (vm_offset_t *)elem);
+#if CONFIG_ZCACHE
+ /*
+ * Note: if zone caching is on, gzalloc and tags aren't used
+ * so we can always check this first
+ */
+ if (zone_caching_enabled(zone)) {
+ addr = zcache_alloc_from_cpu_cache(zone, zstats, waste);
+ if (__probable(addr)) {
+ goto allocated_from_cache;
+ }
+ }
+#endif /* CONFIG_ZCACHE */
- for ( ; element_cursor < backup; element_cursor++)
- *element_cursor = ZP_POISON;
+#if CONFIG_GZALLOC
+ if (__improbable(zone->gzalloc_tracked)) {
+ addr = gzalloc_alloc(zone, zstats, flags);
+ goto allocated_from_gzalloc;
+ }
+#endif /* CONFIG_GZALLOC */
+#if VM_MAX_TAG_ZONES
+ if (__improbable(zone->tags)) {
+ vm_tag_t tag = zalloc_flags_get_tag(flags);
+ if (tag == VM_KERN_MEMORY_NONE) {
+ /*
+ * zone views into heaps can lead to a site-less call
+ * and we fallback to KALLOC as a tag for those.
+ */
+ tag = VM_KERN_MEMORY_KALLOC;
+ flags |= Z_VM_TAG(tag);
}
+ vm_tag_will_update_zone(tag, zone->tag_zone_index);
}
+#endif /* VM_MAX_TAG_ZONES */
lock_zone(zone);
+ assert(zone->z_self == zone);
/*
- * See if we're doing logging on this zone. There are two styles of logging used depending on
- * whether we're trying to catch a leak or corruption. See comments above in zalloc for details.
+ * Check if we need another thread to replenish the zone or
+ * if we have to wait for a replenish thread to finish.
+ * This is used for elements, like vm_map_entry, which are
+ * needed themselves to implement zalloc().
*/
-
- if (__improbable(DO_LOGGING(zone))) {
- if (corruption_debug_flag) {
- /*
- * We're logging to catch a corruption. Add a record of this zfree operation
- * to log.
- */
- btlog_add_entry(zlog_btlog, (void *)addr, ZOP_FREE, (void **)zbt, numsaved);
- } else {
- /*
- * We're logging to catch a leak. Remove any record we might have for this
- * element since it's being freed. Note that we may not find it if the buffer
- * overflowed and that's OK. Since the log is of a limited size, old records
- * get overwritten if there are more zallocs than zfrees.
- */
- btlog_remove_entries_for_element(zlog_btlog, (void *)addr);
+ if (__improbable(zone->prio_refill_count &&
+ zone->countfree <= zone->prio_refill_count / 2)) {
+ zone_refill_asynchronously_locked(zone);
+ } else if (__improbable(zone->countfree == 0)) {
+ zone_refill_synchronously_locked(zone, flags);
+ if (__improbable(zone->countfree == 0)) {
+ unlock_zone(zone);
+ if (__improbable(flags & Z_NOFAIL)) {
+ zone_nofail_panic(zone);
+ }
+ goto out_nomem;
}
}
-#if ZONE_DEBUG
- if (!gzfreed && zone_debug_enabled(zone)) {
- queue_t tmp_elem;
-
- elem -= ZONE_DEBUG_OFFSET;
- if (zone_check) {
- /* check the zone's consistency */
-
- for (tmp_elem = queue_first(&zone->active_zones);
- !queue_end(tmp_elem, &zone->active_zones);
- tmp_elem = queue_next(tmp_elem))
- if (elem == (vm_offset_t)tmp_elem)
- break;
- if (elem != (vm_offset_t)tmp_elem)
- panic("zfree()ing element from wrong zone");
+ addr = zalloc_direct_locked(zone, flags, waste);
+ if (__probable(zstats != NULL)) {
+ /*
+ * The few vm zones used before zone_init() runs do not have
+ * per-cpu stats yet
+ */
+ int cpu = cpu_number();
+ zpercpu_get_cpu(zstats, cpu)->zs_mem_allocated += elem_size;
+#if ZALLOC_DETAILED_STATS
+ if (waste) {
+ zpercpu_get_cpu(zstats, cpu)->zs_mem_wasted += waste;
}
- remqueue((queue_t) elem);
- }
-#endif /* ZONE_DEBUG */
- if (zone_check) {
- zone_check_freelist(zone, elem);
+#endif /* ZALLOC_DETAILED_STATS */
}
- if (__probable(!gzfreed))
- free_to_zone(zone, elem, poison);
+ unlock_zone(zone);
-#if MACH_ASSERT
- if (zone->count < 0)
- panic("zfree: zone count underflow in zone %s while freeing element %p, possible cause: double frees or freeing memory that did not come from this zone",
- zone->zone_name, addr);
+#if ZALLOC_ENABLE_POISONING
+ bool validate = addr & ZALLOC_ELEMENT_NEEDS_VALIDATION;
#endif
-
-
-#if CONFIG_ZLEAKS
+ addr &= ~ZALLOC_ELEMENT_NEEDS_VALIDATION;
+ zone_clear_freelist_pointers(zone, addr);
+#if ZALLOC_ENABLE_POISONING
/*
- * Zone leak detection: un-track the allocation
+ * Note: percpu zones do not respect ZONE_MIN_ELEM_SIZE,
+ * so we will check the first word even if we just
+ * cleared it.
*/
- if (zone->zleak_on) {
- zleak_free(elem, zone->elem_size);
+ zalloc_validate_element(zone, addr, elem_size - sizeof(vm_offset_t),
+ validate);
+#endif /* ZALLOC_ENABLE_POISONING */
+
+allocated_from_cache:
+#if ZONE_ENABLE_LOGGING || CONFIG_ZLEAKS
+ if (__improbable(zalloc_should_log_or_trace_leaks(zone, elem_size))) {
+ zalloc_log_or_trace_leaks(zone, addr);
+ }
+#endif /* ZONE_ENABLE_LOGGING || CONFIG_ZLEAKS */
+
+#if CONFIG_GZALLOC
+allocated_from_gzalloc:
+#endif
+#if KASAN_ZALLOC
+ if (zone->kasan_redzone) {
+ addr = kasan_alloc(addr, elem_size,
+ elem_size - 2 * zone->kasan_redzone, zone->kasan_redzone);
+ elem_size -= 2 * zone->kasan_redzone;
}
-#endif /* CONFIG_ZLEAKS */
-
/*
- * If elements have one or more pages, and memory is low,
- * request to run the garbage collection in the zone the next
- * time the pageout thread runs.
+ * Initialize buffer with unique pattern only if memory
+ * wasn't expected to be zeroed.
*/
- if (zone->elem_size >= PAGE_SIZE &&
- vm_pool_low()){
- zone_gc_forced = TRUE;
+ if (!zone->zfree_clear_mem && !(flags & Z_ZERO)) {
+ kasan_leak_init(addr, elem_size);
+ }
+#endif /* KASAN_ZALLOC */
+ if ((flags & Z_ZERO) && !zone->zfree_clear_mem) {
+ bzero((void *)addr, elem_size);
}
- unlock_zone(zone);
-
- {
- thread_t thr = current_thread();
- task_t task;
- zinfo_usage_t zinfo;
- vm_size_t sz = zone->elem_size;
- if (zone->caller_acct)
- ledger_debit(thr->t_ledger, task_ledgers.tkm_private, sz);
- else
- ledger_debit(thr->t_ledger, task_ledgers.tkm_shared, sz);
+ TRACE_MACHLEAKS(ZALLOC_CODE, ZALLOC_CODE_2, elem_size, addr);
- if ((task = thr->task) != NULL && (zinfo = task->tkm_zinfo) != NULL)
- OSAddAtomic64(sz, (int64_t *)&zinfo[zone->index].free);
- }
+out_nomem:
+ DTRACE_VM2(zalloc, zone_t, zone, void*, addr);
+ return (void *)addr;
}
-
-/* Change a zone's flags.
- * This routine must be called immediately after zinit.
- */
-void
-zone_change(
- zone_t zone,
- unsigned int item,
- boolean_t value)
+void *
+zalloc(union zone_or_view zov)
{
- assert( zone != ZONE_NULL );
- assert( value == TRUE || value == FALSE );
-
- switch(item){
- case Z_NOENCRYPT:
- zone->noencrypt = value;
- break;
- case Z_EXHAUST:
- zone->exhaustible = value;
- break;
- case Z_COLLECT:
- zone->collectable = value;
- break;
- case Z_EXPAND:
- zone->expandable = value;
- break;
- case Z_FOREIGN:
- zone->allows_foreign = value;
- break;
- case Z_CALLERACCT:
- zone->caller_acct = value;
- break;
- case Z_NOCALLOUT:
- zone->no_callout = value;
- break;
- case Z_GZALLOC_EXEMPT:
- zone->gzalloc_exempt = value;
-#if CONFIG_GZALLOC
- gzalloc_reconfigure(zone);
-#endif
- break;
- case Z_ALIGNMENT_REQUIRED:
- zone->alignment_required = value;
- /*
- * Disable the page list optimization here to provide
- * more of an alignment guarantee. This prevents
- * the alignment from being modified by the metadata stored
- * at the beginning of the page.
- */
- zone->use_page_list = FALSE;
-#if ZONE_DEBUG
- zone_debug_disable(zone);
-#endif
-#if CONFIG_GZALLOC
- gzalloc_reconfigure(zone);
-#endif
- break;
- default:
- panic("Zone_change: Wrong Item Type!");
- /* break; */
- }
+ return zalloc_flags(zov, Z_WAITOK);
}
-/*
- * Return the expected number of free elements in the zone.
- * This calculation will be incorrect if items are zfree'd that
- * were never zalloc'd/zget'd. The correct way to stuff memory
- * into a zone is by zcram.
- */
-
-integer_t
-zone_free_count(zone_t zone)
+void *
+zalloc_noblock(union zone_or_view zov)
{
- integer_t free_count;
-
- lock_zone(zone);
- free_count = zone->countfree;
- unlock_zone(zone);
-
- assert(free_count >= 0);
-
- return(free_count);
+ return zalloc_flags(zov, Z_NOWAIT);
}
-/*
- * Zone garbage collection subroutines
- */
+void *
+zalloc_flags(union zone_or_view zov, zalloc_flags_t flags)
+{
+ zone_t zone = zov.zov_view->zv_zone;
+ zone_stats_t zstats = zov.zov_view->zv_stats;
+ assert(!zone->percpu);
+ return zalloc_ext(zone, zstats, flags, 0);
+}
-boolean_t
-zone_page_collectable(
- vm_offset_t addr,
- vm_size_t size)
+void *
+zalloc_percpu(union zone_or_view zov, zalloc_flags_t flags)
{
- struct zone_page_table_entry *zp;
- zone_page_index_t i, j;
+ zone_t zone = zov.zov_view->zv_zone;
+ zone_stats_t zstats = zov.zov_view->zv_stats;
+ assert(zone->percpu);
+ return (void *)__zpcpu_mangle(zalloc_ext(zone, zstats, flags, 0));
+}
-#if ZONE_ALIAS_ADDR
- addr = zone_virtual_addr(addr);
-#endif
-#if MACH_ASSERT
- if (!from_zone_map(addr, size))
- panic("zone_page_collectable");
-#endif
+static void *
+_zalloc_permanent(zone_t zone, vm_size_t size, vm_offset_t mask)
+{
+ const zone_addr_kind_t kind = ZONE_ADDR_NATIVE;
+ struct zone_page_metadata *page_meta;
+ vm_offset_t offs, addr;
+ zone_pva_t pva;
- i = (zone_page_index_t)atop_kernel(addr-zone_map_min_address);
- j = (zone_page_index_t)atop_kernel((addr+size-1) - zone_map_min_address);
+ assert(ml_get_interrupts_enabled() ||
+ ml_is_quiescing() ||
+ debug_mode_active() ||
+ startup_phase < STARTUP_SUB_EARLY_BOOT);
- for (; i <= j; i++) {
- zp = zone_page_table_lookup(i);
- if (zp->collect_count == zp->alloc_count)
- return (TRUE);
- }
+ size = (size + mask) & ~mask;
+ assert(size <= PAGE_SIZE);
- return (FALSE);
-}
+ lock_zone(zone);
+ assert(zone->z_self == zone);
-void
-zone_page_keep(
- vm_offset_t addr,
- vm_size_t size)
-{
- struct zone_page_table_entry *zp;
- zone_page_index_t i, j;
+ for (;;) {
+ pva = zone->pages_intermediate;
+ while (!zone_pva_is_null(pva)) {
+ page_meta = zone_pva_to_meta(pva, kind);
+ if (page_meta->zm_freelist_offs + size <= PAGE_SIZE) {
+ goto found;
+ }
+ pva = page_meta->zm_page_next;
+ }
-#if ZONE_ALIAS_ADDR
- addr = zone_virtual_addr(addr);
-#endif
-#if MACH_ASSERT
- if (!from_zone_map(addr, size))
- panic("zone_page_keep");
-#endif
+ zone_refill_synchronously_locked(zone, Z_WAITOK);
+ }
- i = (zone_page_index_t)atop_kernel(addr-zone_map_min_address);
- j = (zone_page_index_t)atop_kernel((addr+size-1) - zone_map_min_address);
+found:
+ offs = (page_meta->zm_freelist_offs + mask) & ~mask;
+ page_meta->zm_freelist_offs = offs + size;
+ page_meta->zm_alloc_count += size;
+ zone->countfree -= size;
+ if (__probable(zone->z_stats)) {
+ zpercpu_get(zone->z_stats)->zs_mem_allocated += size;
+ }
- for (; i <= j; i++) {
- zp = zone_page_table_lookup(i);
- zp->collect_count = 0;
+ if (page_meta->zm_alloc_count >= PAGE_SIZE - sizeof(vm_offset_t)) {
+ zone_meta_requeue(zone, &zone->pages_all_used, page_meta, kind);
}
-}
-void
-zone_page_collect(
- vm_offset_t addr,
- vm_size_t size)
-{
- struct zone_page_table_entry *zp;
- zone_page_index_t i, j;
+ unlock_zone(zone);
-#if ZONE_ALIAS_ADDR
- addr = zone_virtual_addr(addr);
-#endif
-#if MACH_ASSERT
- if (!from_zone_map(addr, size))
- panic("zone_page_collect");
-#endif
+ addr = offs + zone_pva_to_addr(pva);
- i = (zone_page_index_t)atop_kernel(addr-zone_map_min_address);
- j = (zone_page_index_t)atop_kernel((addr+size-1) - zone_map_min_address);
+ DTRACE_VM2(zalloc, zone_t, zone, void*, addr);
+ return (void *)addr;
+}
- for (; i <= j; i++) {
- zp = zone_page_table_lookup(i);
- ++zp->collect_count;
+static void *
+_zalloc_permanent_large(size_t size, vm_offset_t mask)
+{
+ kern_return_t kr;
+ vm_offset_t addr;
+
+ kr = kernel_memory_allocate(kernel_map, &addr, size, mask,
+ KMA_KOBJECT | KMA_PERMANENT | KMA_ZERO,
+ VM_KERN_MEMORY_KALLOC);
+ if (kr != 0) {
+ panic("zalloc_permanent: unable to allocate %zd bytes (%d)",
+ size, kr);
}
+ return (void *)addr;
}
-void
-zone_page_init(
- vm_offset_t addr,
- vm_size_t size)
+void *
+zalloc_permanent(vm_size_t size, vm_offset_t mask)
{
- struct zone_page_table_entry *zp;
- zone_page_index_t i, j;
-
-#if ZONE_ALIAS_ADDR
- addr = zone_virtual_addr(addr);
-#endif
-#if MACH_ASSERT
- if (!from_zone_map(addr, size))
- panic("zone_page_init");
-#endif
-
- i = (zone_page_index_t)atop_kernel(addr-zone_map_min_address);
- j = (zone_page_index_t)atop_kernel((addr+size-1) - zone_map_min_address);
-
- for (; i <= j; i++) {
- /* make sure entry exists before marking unused */
- zone_page_table_expand(i);
-
- zp = zone_page_table_lookup(i);
- assert(zp);
- zp->alloc_count = ZONE_PAGE_UNUSED;
- zp->collect_count = 0;
+ if (size <= PAGE_SIZE) {
+ zone_t zone = &zone_array[ZONE_ID_PERMANENT];
+ return _zalloc_permanent(zone, size, mask);
}
+ return _zalloc_permanent_large(size, mask);
}
-void
-zone_page_alloc(
- vm_offset_t addr,
- vm_size_t size)
+void *
+zalloc_percpu_permanent(vm_size_t size, vm_offset_t mask)
{
- struct zone_page_table_entry *zp;
- zone_page_index_t i, j;
-
-#if ZONE_ALIAS_ADDR
- addr = zone_virtual_addr(addr);
-#endif
-#if MACH_ASSERT
- if (!from_zone_map(addr, size))
- panic("zone_page_alloc");
-#endif
-
- i = (zone_page_index_t)atop_kernel(addr-zone_map_min_address);
- j = (zone_page_index_t)atop_kernel((addr+size-1) - zone_map_min_address);
-
- for (; i <= j; i++) {
- zp = zone_page_table_lookup(i);
- assert(zp);
-
- /*
- * Set alloc_count to ZONE_PAGE_USED if
- * it was previously set to ZONE_PAGE_UNUSED.
- */
- if (zp->alloc_count == ZONE_PAGE_UNUSED)
- zp->alloc_count = ZONE_PAGE_USED;
-
- ++zp->alloc_count;
- }
+ zone_t zone = &zone_array[ZONE_ID_PERCPU_PERMANENT];
+ return (void *)__zpcpu_mangle(_zalloc_permanent(zone, size, mask));
}
void
-zone_page_free_element(
- zone_page_index_t *free_page_head,
- zone_page_index_t *free_page_tail,
- vm_offset_t addr,
- vm_size_t size)
+zalloc_async(__unused thread_call_param_t p0, __unused thread_call_param_t p1)
{
- struct zone_page_table_entry *zp;
- zone_page_index_t i, j;
-
-#if ZONE_ALIAS_ADDR
- addr = zone_virtual_addr(addr);
-#endif
-#if MACH_ASSERT
- if (!from_zone_map(addr, size))
- panic("zone_page_free_element");
-#endif
-
- /* Clear out the old next and backup pointers */
- vm_offset_t *primary = (vm_offset_t *) addr;
- vm_offset_t *backup = get_backup_ptr(size, primary);
-
- *primary = ZP_POISON;
- *backup = ZP_POISON;
-
- i = (zone_page_index_t)atop_kernel(addr-zone_map_min_address);
- j = (zone_page_index_t)atop_kernel((addr+size-1) - zone_map_min_address);
-
- for (; i <= j; i++) {
- zp = zone_page_table_lookup(i);
-
- if (zp->collect_count > 0)
- --zp->collect_count;
- if (--zp->alloc_count == 0) {
- vm_address_t free_page_address;
- vm_address_t prev_free_page_address;
-
- zp->alloc_count = ZONE_PAGE_UNUSED;
- zp->collect_count = 0;
-
+ zone_index_foreach(i) {
+ zone_t z = &zone_array[i];
- /*
- * This element was the last one on this page, re-use the page's
- * storage for a page freelist
- */
- free_page_address = zone_map_min_address + PAGE_SIZE * ((vm_size_t)i);
- *(zone_page_index_t *)free_page_address = ZONE_PAGE_INDEX_INVALID;
+ if (z->no_callout) {
+ /* async_pending will never be set */
+ continue;
+ }
- if (*free_page_head == ZONE_PAGE_INDEX_INVALID) {
- *free_page_head = i;
- *free_page_tail = i;
- } else {
- prev_free_page_address = zone_map_min_address + PAGE_SIZE * ((vm_size_t)(*free_page_tail));
- *(zone_page_index_t *)prev_free_page_address = i;
- *free_page_tail = i;
- }
+ lock_zone(z);
+ if (z->z_self && z->async_pending) {
+ z->async_pending = false;
+ zone_refill_synchronously_locked(z, Z_WAITOK);
}
+ unlock_zone(z);
}
}
-
-#define ZONEGC_SMALL_ELEMENT_SIZE 4096
-
-struct {
- uint64_t zgc_invoked;
- uint64_t zgc_bailed;
- uint32_t pgs_freed;
-
- uint32_t elems_collected,
- elems_freed,
- elems_kept;
-} zgc_stats;
-
-/* Zone garbage collection
- *
- * zone_gc will walk through all the free elements in all the
- * zones that are marked collectable looking for reclaimable
- * pages. zone_gc is called by consider_zone_gc when the system
- * begins to run out of memory.
+/*
+ * Adds the element to the head of the zone's free list
+ * Keeps a backup next-pointer at the end of the element
*/
void
-zone_gc(boolean_t all_zones)
+zfree_direct_locked(zone_t zone, vm_offset_t element, bool poison)
{
- unsigned int max_zones;
- zone_t z;
- unsigned int i;
- uint32_t old_pgs_freed;
- zone_page_index_t zone_free_page_head;
- zone_page_index_t zone_free_page_tail;
- thread_t mythread = current_thread();
+ struct zone_page_metadata *page_meta;
+ vm_offset_t page, old_head;
+ zone_addr_kind_t kind;
+ vm_size_t elem_size = zone_elem_size(zone);
- lck_mtx_lock(&zone_gc_lock);
+ vm_offset_t *primary = (vm_offset_t *) element;
+ vm_offset_t *backup = get_backup_ptr(elem_size, primary);
- zgc_stats.zgc_invoked++;
- old_pgs_freed = zgc_stats.pgs_freed;
+ page_meta = zone_allocated_element_resolve(zone, element, &page, &kind);
+ old_head = zone_page_meta_get_freelist(zone, page_meta, page);
- simple_lock(&all_zones_lock);
- max_zones = num_zones;
- z = first_zone;
- simple_unlock(&all_zones_lock);
+ if (__improbable(old_head == element)) {
+ panic("zfree: double free of %p to zone %s%s\n",
+ (void *) element, zone_heap_name(zone), zone->z_name);
+ }
- if (zalloc_debug & ZALLOC_DEBUG_ZONEGC)
- kprintf("zone_gc(all_zones=%s) starting...\n", all_zones ? "TRUE" : "FALSE");
+#if ZALLOC_ENABLE_POISONING
+ if (poison && elem_size < ZONE_MIN_ELEM_SIZE) {
+ assert(zone->percpu);
+ poison = false;
+ }
+#else
+ poison = false;
+#endif
/*
- * it's ok to allow eager kernel preemption while
- * while holding a zone lock since it's taken
- * as a spin lock (which prevents preemption)
+ * Always write a redundant next pointer
+ * So that it is more difficult to forge, xor it with a random cookie
+ * A poisoned element is indicated by using zp_poisoned_cookie
+ * instead of zp_nopoison_cookie
*/
- thread_set_eager_preempt(mythread);
-
-#if MACH_ASSERT
- for (i = 0; i < zone_pages; i++) {
- struct zone_page_table_entry *zp;
-
- zp = zone_page_table_lookup(i);
- assert(!zp || (zp->collect_count == 0));
- }
-#endif /* MACH_ASSERT */
-
- for (i = 0; i < max_zones; i++, z = z->next_zone) {
- unsigned int n, m;
- vm_size_t elt_size, size_freed;
- struct zone_free_element *elt, *base_elt, *base_prev, *prev, *scan, *keep, *tail;
- int kmem_frees = 0, total_freed_pages = 0;
- struct zone_page_metadata *page_meta;
- queue_head_t page_meta_head;
-
- assert(z != ZONE_NULL);
-
- if (!z->collectable)
- continue;
- if (all_zones == FALSE && z->elem_size < ZONEGC_SMALL_ELEMENT_SIZE && !z->use_page_list)
- continue;
+ *backup = old_head ^ (poison ? zp_poisoned_cookie : zp_nopoison_cookie);
- lock_zone(z);
+ /*
+ * Insert this element at the head of the free list. We also xor the
+ * primary pointer with the zp_nopoison_cookie to make sure a free
+ * element does not provide the location of the next free element directly.
+ */
+ *primary = old_head ^ zp_nopoison_cookie;
+
+#if VM_MAX_TAG_ZONES
+ if (__improbable(zone->tags)) {
+ vm_tag_t tag = (ZTAG(zone, element)[0] >> 1);
+ // set the tag with b0 clear so the block remains inuse
+ ZTAG(zone, element)[0] = 0xFFFE;
+ vm_tag_update_zone_size(tag, zone->tag_zone_index,
+ -((int64_t)elem_size), 0);
+ }
+#endif /* VM_MAX_TAG_ZONES */
- elt_size = z->elem_size;
+ zone_page_meta_set_freelist(page_meta, page, element);
+ if (os_sub_overflow(page_meta->zm_alloc_count, 1,
+ &page_meta->zm_alloc_count)) {
+ zone_page_meta_accounting_panic(zone, page_meta,
+ "alloc_count wrap-around");
+ }
+ zone->countfree++;
- /*
- * Do a quick feasibility check before we scan the zone:
- * skip unless there is likelihood of getting pages back
- * (i.e we need a whole allocation block's worth of free
- * elements before we can garbage collect) and
- * the zone has more than 10 percent of it's elements free
- * or the element size is a multiple of the PAGE_SIZE
- */
- if ((elt_size & PAGE_MASK) &&
- !z->use_page_list &&
- (((z->cur_size - z->count * elt_size) <= (2 * z->alloc_size)) ||
- ((z->cur_size - z->count * elt_size) <= (z->cur_size / 10)))) {
- unlock_zone(z);
- continue;
+ if (kind == ZONE_ADDR_FOREIGN) {
+ if (old_head == 0) {
+ /* first foreign element freed on page, move from all_used_foreign */
+ zone_meta_requeue(zone, &zone->pages_any_free_foreign, page_meta, kind);
}
+ } else if (page_meta->zm_alloc_count == 0) {
+ /* whether the page was on the intermediate or all_used, queue, move it to free */
+ zone_meta_requeue(zone, &zone->pages_all_free, page_meta, kind);
+ zone->allfree_page_count += page_meta->zm_page_count;
+ } else if (old_head == 0) {
+ /* first free element on page, move from all_used */
+ zone_meta_requeue(zone, &zone->pages_intermediate, page_meta, kind);
+ }
- z->doing_gc = TRUE;
-
- /*
- * Snatch all of the free elements away from the zone.
- */
-
- if (z->use_page_list) {
- queue_new_head(&z->pages.all_free, &page_meta_head, struct zone_page_metadata *, pages);
- queue_init(&z->pages.all_free);
- } else {
- scan = (void *)z->free_elements;
- z->free_elements = 0;
+#if KASAN_ZALLOC
+ if (zone->percpu) {
+ zpercpu_foreach_cpu(i) {
+ kasan_poison_range(element + ptoa(i), elem_size,
+ ASAN_HEAP_FREED);
}
-
- unlock_zone(z);
-
- if (z->use_page_list) {
- /*
- * For zones that maintain page lists (which in turn
- * track free elements on those pages), zone_gc()
- * is incredibly easy, and we bypass all the logic
- * for scanning elements and mapping them to
- * collectable pages
- */
-
- size_freed = 0;
-
- queue_iterate(&page_meta_head, page_meta, struct zone_page_metadata *, pages) {
- assert(from_zone_map((vm_address_t)page_meta, sizeof(*page_meta))); /* foreign elements should be in any_free_foreign */
-
- zgc_stats.elems_freed += page_meta->free_count;
- size_freed += elt_size * page_meta->free_count;
- zgc_stats.elems_collected += page_meta->free_count;
- }
-
- lock_zone(z);
-
- if (size_freed > 0) {
- z->cur_size -= size_freed;
- z->countfree -= size_freed/elt_size;
- }
-
- z->doing_gc = FALSE;
- if (z->waiting) {
- z->waiting = FALSE;
- zone_wakeup(z);
- }
-
- unlock_zone(z);
-
- if (queue_empty(&page_meta_head))
- continue;
-
- thread_clear_eager_preempt(mythread);
-
- while ((page_meta = (struct zone_page_metadata *)dequeue_head(&page_meta_head)) != NULL) {
- vm_address_t free_page_address;
-
- free_page_address = trunc_page((vm_address_t)page_meta);
-#if ZONE_ALIAS_ADDR
- free_page_address = zone_virtual_addr(free_page_address);
+ } else {
+ kasan_poison_range(element, elem_size, ASAN_HEAP_FREED);
+ }
#endif
- kmem_free(zone_map, free_page_address, PAGE_SIZE);
- ZONE_PAGE_COUNT_DECR(z, 1);
- total_freed_pages++;
- zgc_stats.pgs_freed += 1;
-
- if (++kmem_frees == 32) {
- thread_yield_internal(1);
- kmem_frees = 0;
- }
- }
-
- if (zalloc_debug & ZALLOC_DEBUG_ZONEGC)
- kprintf("zone_gc() of zone %s freed %lu elements, %d pages\n", z->zone_name, (unsigned long)size_freed/elt_size, total_freed_pages);
-
- thread_set_eager_preempt(mythread);
- continue; /* go to next zone */
- }
-
- /*
- * Pass 1:
- *
- * Determine which elements we can attempt to collect
- * and count them up in the page table. Foreign elements
- * are returned to the zone.
- */
-
- prev = (void *)&scan;
- elt = scan;
- n = 0; tail = keep = NULL;
-
- zone_free_page_head = ZONE_PAGE_INDEX_INVALID;
- zone_free_page_tail = ZONE_PAGE_INDEX_INVALID;
-
-
- while (elt != NULL) {
- if (from_zone_map(elt, elt_size)) {
- zone_page_collect((vm_offset_t)elt, elt_size);
-
- prev = elt;
- elt = elt->next;
-
- ++zgc_stats.elems_collected;
- }
- else {
- if (keep == NULL)
- keep = tail = elt;
- else {
- append_zone_element(z, tail, elt);
- tail = elt;
- }
-
- append_zone_element(z, prev, elt->next);
- elt = elt->next;
- append_zone_element(z, tail, NULL);
- }
-
- /*
- * Dribble back the elements we are keeping.
- * If there are none, give some elements that we haven't looked at yet
- * back to the freelist so that others waiting on the zone don't get stuck
- * for too long. This might prevent us from recovering some memory,
- * but allows us to avoid having to allocate new memory to serve requests
- * while zone_gc has all the free memory tied up.
- * <rdar://problem/3893406>
- */
-
- if (++n >= 50) {
- if (z->waiting == TRUE) {
- /* z->waiting checked without lock held, rechecked below after locking */
- lock_zone(z);
-
- if (keep != NULL) {
- add_list_to_zone(z, keep, tail);
- tail = keep = NULL;
- } else {
- m =0;
- base_elt = elt;
- base_prev = prev;
- while ((elt != NULL) && (++m < 50)) {
- prev = elt;
- elt = elt->next;
- }
- if (m !=0 ) {
- /* Extract the elements from the list and
- * give them back */
- append_zone_element(z, prev, NULL);
- add_list_to_zone(z, base_elt, prev);
- append_zone_element(z, base_prev, elt);
- prev = base_prev;
- }
- }
-
- if (z->waiting) {
- z->waiting = FALSE;
- zone_wakeup(z);
- }
-
- unlock_zone(z);
- }
- n =0;
- }
- }
-
- /*
- * Return any remaining elements.
- */
-
- if (keep != NULL) {
- lock_zone(z);
-
- add_list_to_zone(z, keep, tail);
+}
- if (z->waiting) {
- z->waiting = FALSE;
- zone_wakeup(z);
- }
+/*
+ * The function is noinline when zlog can be used so that the backtracing can
+ * reliably skip the zfree_ext() and zfree_log_trace()
+ * boring frames.
+ */
+#if ZONE_ENABLE_LOGGING
+__attribute__((noinline))
+#endif
+void
+zfree_ext(zone_t zone, zone_stats_t zstats, void *addr)
+{
+ vm_offset_t elem = (vm_offset_t)addr;
+ vm_size_t elem_size = zone_elem_size(zone);
+ bool poison = false;
- unlock_zone(z);
- }
+ DTRACE_VM2(zfree, zone_t, zone, void*, addr);
+ TRACE_MACHLEAKS(ZFREE_CODE, ZFREE_CODE_2, elem_size, elem);
- /*
- * Pass 2:
- *
- * Determine which pages we can reclaim and
- * free those elements.
- */
+#if KASAN_ZALLOC
+ if (kasan_quarantine_freed_element(&zone, &addr)) {
+ return;
+ }
+ /*
+ * kasan_quarantine_freed_element() might return a different
+ * {zone, addr} than the one being freed for kalloc heaps.
+ *
+ * Make sure we reload everything.
+ */
+ elem = (vm_offset_t)addr;
+ elem_size = zone_elem_size(zone);
+#endif
- size_freed = 0;
- elt = scan;
- n = 0; tail = keep = NULL;
+#if CONFIG_ZLEAKS
+ /*
+ * Zone leak detection: un-track the allocation
+ */
+ if (__improbable(zone->zleak_on)) {
+ zleak_free(elem, elem_size);
+ }
+#endif /* CONFIG_ZLEAKS */
- while (elt != NULL) {
- if (zone_page_collectable((vm_offset_t)elt, elt_size)) {
- struct zone_free_element *next_elt = elt->next;
+#if CONFIG_ZCACHE
+ /*
+ * Note: if zone caching is on, gzalloc and tags aren't used
+ * so we can always check this first
+ */
+ if (zone_caching_enabled(zone)) {
+ return zcache_free_to_cpu_cache(zone, zstats, (vm_offset_t)addr);
+ }
+#endif /* CONFIG_ZCACHE */
- size_freed += elt_size;
+#if CONFIG_GZALLOC
+ if (__improbable(zone->gzalloc_tracked)) {
+ return gzalloc_free(zone, zstats, addr);
+ }
+#endif /* CONFIG_GZALLOC */
- /*
- * If this is the last allocation on the page(s),
- * we may use their storage to maintain the linked
- * list of free-able pages. So store elt->next because
- * "elt" may be scribbled over.
- */
- zone_page_free_element(&zone_free_page_head, &zone_free_page_tail, (vm_offset_t)elt, elt_size);
+#if ZONE_ENABLE_LOGGING
+ if (__improbable(DO_LOGGING(zone))) {
+ zfree_log_trace(zone, elem);
+ }
+#endif /* ZONE_ENABLE_LOGGING */
- elt = next_elt;
+ if (zone->zfree_clear_mem) {
+ poison = zfree_clear(zone, elem, elem_size);
+ }
- ++zgc_stats.elems_freed;
- }
- else {
- zone_page_keep((vm_offset_t)elt, elt_size);
-
- if (keep == NULL)
- keep = tail = elt;
- else {
- append_zone_element(z, tail, elt);
- tail = elt;
- }
+ lock_zone(zone);
+ assert(zone->z_self == zone);
- elt = elt->next;
- append_zone_element(z, tail, NULL);
+ if (!poison) {
+ poison = zfree_poison_element(zone, &zone->zp_count, elem);
+ }
- ++zgc_stats.elems_kept;
- }
+ if (__probable(zstats != NULL)) {
+ /*
+ * The few vm zones used before zone_init() runs do not have
+ * per-cpu stats yet
+ */
+ zpercpu_get(zstats)->zs_mem_freed += elem_size;
+ }
- /*
- * Dribble back the elements we are keeping,
- * and update the zone size info.
- */
+ zfree_direct_locked(zone, elem, poison);
- if (++n >= 50) {
- lock_zone(z);
+ unlock_zone(zone);
+}
- z->cur_size -= size_freed;
- z->countfree -= size_freed/elt_size;
- size_freed = 0;
+void
+(zfree)(union zone_or_view zov, void *addr)
+{
+ zone_t zone = zov.zov_view->zv_zone;
+ zone_stats_t zstats = zov.zov_view->zv_stats;
+ assert(!zone->percpu);
+ zfree_ext(zone, zstats, addr);
+}
- if (keep != NULL) {
- add_list_to_zone(z, keep, tail);
- }
+void
+zfree_percpu(union zone_or_view zov, void *addr)
+{
+ zone_t zone = zov.zov_view->zv_zone;
+ zone_stats_t zstats = zov.zov_view->zv_stats;
+ assert(zone->percpu);
+ zfree_ext(zone, zstats, (void *)__zpcpu_demangle(addr));
+}
- if (z->waiting) {
- z->waiting = FALSE;
- zone_wakeup(z);
- }
+#pragma mark vm integration, MIG routines
+/*
+ * Drops (i.e. frees) the elements in the all free pages queue of a zone.
+ * Called by zone_gc() on each zone and when a zone is zdestroy()ed.
+ */
+static void
+zone_drop_free_elements(zone_t z)
+{
+ const zone_addr_kind_t kind = ZONE_ADDR_NATIVE;
+ unsigned int total_freed_pages = 0;
+ struct zone_page_metadata *page_meta, *seq_meta;
+ vm_address_t page_addr;
+ vm_size_t size_to_free;
+ vm_size_t free_count;
+ uint32_t page_count;
+
+ current_thread()->options |= TH_OPT_ZONE_PRIV;
+ lock_zone(z);
+
+ while (!zone_pva_is_null(z->pages_all_free)) {
+ /*
+ * If any replenishment threads are running, defer to them,
+ * so that we don't deplete reserved zones.
+ *
+ * The timing of the check isn't super important, as there are
+ * enough reserves to allow freeing an extra page_meta.
+ *
+ * Hence, we can check without grabbing the lock every time
+ * through the loop. We do need the lock however to avoid
+ * missing a wakeup when we decide to block.
+ */
+ if (zone_replenish_active > 0) {
+ lck_spin_lock(&zone_replenish_lock);
+ if (zone_replenish_active > 0) {
+ assert_wait(&zone_replenish_active, THREAD_UNINT);
+ lck_spin_unlock(&zone_replenish_lock);
unlock_zone(z);
-
- n = 0; tail = keep = NULL;
+ thread_block(THREAD_CONTINUE_NULL);
+ lock_zone(z);
+ continue;
}
+ lck_spin_unlock(&zone_replenish_lock);
}
+ page_meta = zone_pva_to_meta(z->pages_all_free, kind);
+ page_count = page_meta->zm_page_count;
+ free_count = zone_elem_count(z, ptoa(page_count), kind);
+
/*
- * Return any remaining elements, and update
- * the zone size info.
+ * Don't drain zones with async refill to below the refill
+ * threshold, as they need some reserve to function properly.
*/
+ if (!z->destroyed && z->prio_refill_count &&
+ (vm_size_t)(z->countfree - free_count) < z->prio_refill_count) {
+ break;
+ }
- lock_zone(z);
-
- if (size_freed > 0 || keep != NULL) {
+ zone_meta_queue_pop(z, &z->pages_all_free, kind, &page_addr);
- z->cur_size -= size_freed;
- z->countfree -= size_freed/elt_size;
+ if (os_sub_overflow(z->countfree, free_count, &z->countfree)) {
+ zone_accounting_panic(z, "countfree wrap-around");
+ }
+ if (os_sub_overflow(z->countavail, free_count, &z->countavail)) {
+ zone_accounting_panic(z, "countavail wrap-around");
+ }
+ if (os_sub_overflow(z->allfree_page_count, page_count,
+ &z->allfree_page_count)) {
+ zone_accounting_panic(z, "allfree_page_count wrap-around");
+ }
+ if (os_sub_overflow(z->page_count, page_count, &z->page_count)) {
+ zone_accounting_panic(z, "page_count wrap-around");
+ }
- if (keep != NULL) {
- add_list_to_zone(z, keep, tail);
- }
+ os_atomic_sub(&zones_phys_page_count, page_count, relaxed);
+ os_atomic_sub(&zones_phys_page_mapped_count, page_count, relaxed);
- }
+ bzero(page_meta, sizeof(*page_meta) * page_count);
+ seq_meta = page_meta;
+ page_meta = NULL; /* page_meta fields are zeroed, prevent reuse */
- z->doing_gc = FALSE;
- if (z->waiting) {
- z->waiting = FALSE;
- zone_wakeup(z);
- }
unlock_zone(z);
- if (zone_free_page_head == ZONE_PAGE_INDEX_INVALID)
- continue;
+ /* Free the pages for metadata and account for them */
+ total_freed_pages += page_count;
+ size_to_free = ptoa(page_count);
+#if KASAN_ZALLOC
+ kasan_poison_range(page_addr, size_to_free, ASAN_VALID);
+#endif
+#if VM_MAX_TAG_ZONES
+ if (z->tags) {
+ ztMemoryRemove(z, page_addr, size_to_free);
+ }
+#endif /* VM_MAX_TAG_ZONES */
- /*
- * we don't want to allow eager kernel preemption while holding the
- * various locks taken in the kmem_free path of execution
- */
- thread_clear_eager_preempt(mythread);
+ if (z->va_sequester && z->alloc_pages == page_count) {
+ kernel_memory_depopulate(submap_for_zone(z), page_addr,
+ size_to_free, KMA_KOBJECT, VM_KERN_MEMORY_ZONE);
+ } else {
+ kmem_free(submap_for_zone(z), page_addr, size_to_free);
+ seq_meta = NULL;
+ }
+ thread_yield_to_preemption();
+ lock_zone(z);
- /*
- * This loop counts the number of pages that should be freed by the
- * next loop that tries to coalesce the kmem_frees()
- */
- uint32_t pages_to_free_count = 0;
- vm_address_t fpa;
- zone_page_index_t index;
- for (index = zone_free_page_head; index != ZONE_PAGE_INDEX_INVALID;) {
- pages_to_free_count++;
- fpa = zone_map_min_address + PAGE_SIZE * ((vm_size_t)index);
- index = *(zone_page_index_t *)fpa;
+ if (seq_meta) {
+ zone_meta_queue_push(z, &z->pages_sequester, seq_meta, kind);
+ z->sequester_page_count += page_count;
}
+ }
+ if (z->destroyed) {
+ assert(zone_pva_is_null(z->pages_all_free));
+ assert(z->allfree_page_count == 0);
+ }
+ unlock_zone(z);
+ current_thread()->options &= ~TH_OPT_ZONE_PRIV;
+
+#if DEBUG || DEVELOPMENT
+ if (zalloc_debug & ZALLOC_DEBUG_ZONEGC) {
+ kprintf("zone_gc() of zone %s%s freed %lu elements, %d pages\n",
+ zone_heap_name(z), z->z_name,
+ (unsigned long)(ptoa(total_freed_pages) / z->pcpu_elem_size),
+ total_freed_pages);
+ }
+#endif /* DEBUG || DEVELOPMENT */
+}
+/* Zone garbage collection
+ *
+ * zone_gc will walk through all the free elements in all the
+ * zones that are marked collectable looking for reclaimable
+ * pages. zone_gc is called by consider_zone_gc when the system
+ * begins to run out of memory.
+ *
+ * We should ensure that zone_gc never blocks.
+ */
+void
+zone_gc(boolean_t consider_jetsams)
+{
+ if (consider_jetsams) {
+ kill_process_in_largest_zone();
/*
- * Reclaim the pages we are freeing.
+ * If we do end up jetsamming something, we need to do a zone_gc so that
+ * we can reclaim free zone elements and update the zone map size.
+ * Fall through.
*/
- while (zone_free_page_head != ZONE_PAGE_INDEX_INVALID) {
- zone_page_index_t zind = zone_free_page_head;
- vm_address_t free_page_address;
- int page_count;
-
- /*
- * Use the first word of the page about to be freed to find the next free page
- */
- free_page_address = zone_map_min_address + PAGE_SIZE * ((vm_size_t)zind);
- zone_free_page_head = *(zone_page_index_t *)free_page_address;
-
- page_count = 1;
- total_freed_pages++;
+ }
- while (zone_free_page_head != ZONE_PAGE_INDEX_INVALID) {
- zone_page_index_t next_zind = zone_free_page_head;
- vm_address_t next_free_page_address;
+ lck_mtx_lock(&zone_gc_lock);
- next_free_page_address = zone_map_min_address + PAGE_SIZE * ((vm_size_t)next_zind);
+#if DEBUG || DEVELOPMENT
+ if (zalloc_debug & ZALLOC_DEBUG_ZONEGC) {
+ kprintf("zone_gc() starting...\n");
+ }
+#endif /* DEBUG || DEVELOPMENT */
- if (next_free_page_address == (free_page_address - PAGE_SIZE)) {
- free_page_address = next_free_page_address;
- } else if (next_free_page_address != (free_page_address + (PAGE_SIZE * page_count)))
- break;
+ zone_index_foreach(i) {
+ zone_t z = &zone_array[i];
- zone_free_page_head = *(zone_page_index_t *)next_free_page_address;
- page_count++;
- total_freed_pages++;
- }
- kmem_free(zone_map, free_page_address, page_count * PAGE_SIZE);
- ZONE_PAGE_COUNT_DECR(z, page_count);
- zgc_stats.pgs_freed += page_count;
- pages_to_free_count -= page_count;
-
- if (++kmem_frees == 32) {
- thread_yield_internal(1);
- kmem_frees = 0;
- }
+ if (!z->collectable) {
+ continue;
+ }
+#if CONFIG_ZCACHE
+ if (zone_caching_enabled(z)) {
+ zcache_drain_depot(z);
+ }
+#endif /* CONFIG_ZCACHE */
+ if (zone_pva_is_null(z->pages_all_free)) {
+ continue;
}
- /* Check that we actually free the exact number of pages we were supposed to */
- assert(pages_to_free_count == 0);
-
- if (zalloc_debug & ZALLOC_DEBUG_ZONEGC)
- kprintf("zone_gc() of zone %s freed %lu elements, %d pages\n", z->zone_name, (unsigned long)size_freed/elt_size, total_freed_pages);
-
- thread_set_eager_preempt(mythread);
+ zone_drop_free_elements(z);
}
- if (old_pgs_freed == zgc_stats.pgs_freed)
- zgc_stats.zgc_bailed++;
-
- thread_clear_eager_preempt(mythread);
-
lck_mtx_unlock(&zone_gc_lock);
-
}
-extern vm_offset_t kmapoff_kaddr;
-extern unsigned int kmapoff_pgcnt;
-
/*
* consider_zone_gc:
*
*/
void
-consider_zone_gc(boolean_t force)
+consider_zone_gc(boolean_t consider_jetsams)
+{
+ /*
+ * One-time reclaim of kernel_map resources we allocated in
+ * early boot.
+ *
+ * Use atomic exchange in case multiple threads race into here.
+ */
+ vm_offset_t deallocate_kaddr;
+ if (kmapoff_kaddr != 0 &&
+ (deallocate_kaddr = os_atomic_xchg(&kmapoff_kaddr, 0, relaxed)) != 0) {
+ vm_deallocate(kernel_map, deallocate_kaddr, ptoa_64(kmapoff_pgcnt));
+ }
+
+ zone_gc(consider_jetsams);
+}
+
+/*
+ * Creates a vm_map_copy_t to return to the caller of mach_* MIG calls
+ * requesting zone information.
+ * Frees unused pages towards the end of the region, and zero'es out unused
+ * space on the last page.
+ */
+static vm_map_copy_t
+create_vm_map_copy(
+ vm_offset_t start_addr,
+ vm_size_t total_size,
+ vm_size_t used_size)
+{
+ kern_return_t kr;
+ vm_offset_t end_addr;
+ vm_size_t free_size;
+ vm_map_copy_t copy;
+
+ if (used_size != total_size) {
+ end_addr = start_addr + used_size;
+ free_size = total_size - (round_page(end_addr) - start_addr);
+
+ if (free_size >= PAGE_SIZE) {
+ kmem_free(ipc_kernel_map,
+ round_page(end_addr), free_size);
+ }
+ bzero((char *) end_addr, round_page(end_addr) - end_addr);
+ }
+
+ kr = vm_map_copyin(ipc_kernel_map, (vm_map_address_t)start_addr,
+ (vm_map_size_t)used_size, TRUE, ©);
+ assert(kr == KERN_SUCCESS);
+
+ return copy;
+}
+
+static boolean_t
+get_zone_info(
+ zone_t z,
+ mach_zone_name_t *zn,
+ mach_zone_info_t *zi)
{
- boolean_t all_zones = FALSE;
+ struct zone zcopy;
+
+ assert(z != ZONE_NULL);
+ lock_zone(z);
+ if (!z->z_self) {
+ unlock_zone(z);
+ return FALSE;
+ }
+ zcopy = *z;
+ unlock_zone(z);
- if (kmapoff_kaddr != 0) {
+ if (zn != NULL) {
/*
- * One-time reclaim of kernel_map resources we allocated in
- * early boot.
+ * Append kalloc heap name to zone name (if zone is used by kalloc)
*/
- (void) vm_deallocate(kernel_map,
- kmapoff_kaddr, kmapoff_pgcnt * PAGE_SIZE_64);
- kmapoff_kaddr = 0;
+ char temp_zone_name[MAX_ZONE_NAME] = "";
+ snprintf(temp_zone_name, MAX_ZONE_NAME, "%s%s",
+ zone_heap_name(z), z->z_name);
+
+ /* assuming here the name data is static */
+ (void) __nosan_strlcpy(zn->mzn_name, temp_zone_name,
+ strlen(temp_zone_name) + 1);
}
- if (zone_gc_allowed &&
- (zone_gc_allowed_by_time_throttle ||
- zone_gc_forced ||
- force)) {
- if (zone_gc_allowed_by_time_throttle == TRUE) {
- zone_gc_allowed_by_time_throttle = FALSE;
- all_zones = TRUE;
+ if (zi != NULL) {
+ *zi = (mach_zone_info_t) {
+ .mzi_count = zone_count_allocated(&zcopy),
+ .mzi_cur_size = ptoa_64(zcopy.page_count),
+ // max_size for zprint is now high-watermark of pages used
+ .mzi_max_size = ptoa_64(zcopy.page_count_hwm),
+ .mzi_elem_size = zcopy.pcpu_elem_size,
+ .mzi_alloc_size = ptoa_64(zcopy.alloc_pages),
+ .mzi_exhaustible = (uint64_t)zcopy.exhaustible,
+ };
+ zpercpu_foreach(zs, zcopy.z_stats) {
+ zi->mzi_sum_size += zs->zs_mem_allocated;
+ }
+ if (zcopy.collectable) {
+ SET_MZI_COLLECTABLE_BYTES(zi->mzi_collectable,
+ ptoa_64(zcopy.allfree_page_count));
+ SET_MZI_COLLECTABLE_FLAG(zi->mzi_collectable, TRUE);
}
- zone_gc_forced = FALSE;
-
- zone_gc(all_zones);
}
+
+ return TRUE;
}
-/*
- * By default, don't attempt zone GC more frequently
- * than once / 1 minutes.
- */
-void
-compute_zone_gc_throttle(void *arg __unused)
+kern_return_t
+task_zone_info(
+ __unused task_t task,
+ __unused mach_zone_name_array_t *namesp,
+ __unused mach_msg_type_number_t *namesCntp,
+ __unused task_zone_info_array_t *infop,
+ __unused mach_msg_type_number_t *infoCntp)
{
- zone_gc_allowed_by_time_throttle = TRUE;
+ return KERN_FAILURE;
}
+kern_return_t
+mach_zone_info(
+ host_priv_t host,
+ mach_zone_name_array_t *namesp,
+ mach_msg_type_number_t *namesCntp,
+ mach_zone_info_array_t *infop,
+ mach_msg_type_number_t *infoCntp)
+{
+ return mach_memory_info(host, namesp, namesCntp, infop, infoCntp, NULL, NULL);
+}
-#if CONFIG_TASK_ZONE_INFO
kern_return_t
-task_zone_info(
- task_t task,
- mach_zone_name_array_t *namesp,
+mach_memory_info(
+ host_priv_t host,
+ mach_zone_name_array_t *namesp,
mach_msg_type_number_t *namesCntp,
- task_zone_info_array_t *infop,
- mach_msg_type_number_t *infoCntp)
+ mach_zone_info_array_t *infop,
+ mach_msg_type_number_t *infoCntp,
+ mach_memory_info_array_t *memoryInfop,
+ mach_msg_type_number_t *memoryInfoCntp)
{
- mach_zone_name_t *names;
- vm_offset_t names_addr;
- vm_size_t names_size;
- task_zone_info_t *info;
- vm_offset_t info_addr;
- vm_size_t info_size;
- unsigned int max_zones, i;
- zone_t z;
- mach_zone_name_t *zn;
- task_zone_info_t *zi;
- kern_return_t kr;
+ mach_zone_name_t *names;
+ vm_offset_t names_addr;
+ vm_size_t names_size;
+
+ mach_zone_info_t *info;
+ vm_offset_t info_addr;
+ vm_size_t info_size;
+
+ mach_memory_info_t *memory_info;
+ vm_offset_t memory_info_addr;
+ vm_size_t memory_info_size;
+ vm_size_t memory_info_vmsize;
+ unsigned int num_info;
- vm_size_t used;
- vm_map_copy_t copy;
+ unsigned int max_zones, used_zones, i;
+ mach_zone_name_t *zn;
+ mach_zone_info_t *zi;
+ kern_return_t kr;
+ uint64_t zones_collectable_bytes = 0;
- if (task == TASK_NULL)
- return KERN_INVALID_TASK;
+ if (host == HOST_NULL) {
+ return KERN_INVALID_HOST;
+ }
+#if CONFIG_DEBUGGER_FOR_ZONE_INFO
+ if (!PE_i_can_has_debugger(NULL)) {
+ return KERN_INVALID_HOST;
+ }
+#endif
/*
* We assume that zones aren't freed once allocated.
* We won't pick up any zones that are allocated later.
*/
- simple_lock(&all_zones_lock);
- max_zones = (unsigned int)(num_zones + num_fake_zones);
- z = first_zone;
- simple_unlock(&all_zones_lock);
+ max_zones = os_atomic_load(&num_zones, relaxed);
names_size = round_page(max_zones * sizeof *names);
kr = kmem_alloc_pageable(ipc_kernel_map,
- &names_addr, names_size, VM_KERN_MEMORY_IPC);
- if (kr != KERN_SUCCESS)
+ &names_addr, names_size, VM_KERN_MEMORY_IPC);
+ if (kr != KERN_SUCCESS) {
return kr;
+ }
names = (mach_zone_name_t *) names_addr;
info_size = round_page(max_zones * sizeof *info);
kr = kmem_alloc_pageable(ipc_kernel_map,
- &info_addr, info_size, VM_KERN_MEMORY_IPC);
+ &info_addr, info_size, VM_KERN_MEMORY_IPC);
if (kr != KERN_SUCCESS) {
kmem_free(ipc_kernel_map,
- names_addr, names_size);
+ names_addr, names_size);
return kr;
}
-
- info = (task_zone_info_t *) info_addr;
+ info = (mach_zone_info_t *) info_addr;
zn = &names[0];
zi = &info[0];
- for (i = 0; i < max_zones - num_fake_zones; i++) {
- struct zone zcopy;
-
- assert(z != ZONE_NULL);
-
- lock_zone(z);
- zcopy = *z;
- unlock_zone(z);
-
- simple_lock(&all_zones_lock);
- z = z->next_zone;
- simple_unlock(&all_zones_lock);
-
- /* assuming here the name data is static */
- (void) strncpy(zn->mzn_name, zcopy.zone_name,
- sizeof zn->mzn_name);
- zn->mzn_name[sizeof zn->mzn_name - 1] = '\0';
-
- zi->tzi_count = (uint64_t)zcopy.count;
- zi->tzi_cur_size = ptoa_64(zcopy.page_count);
- zi->tzi_max_size = (uint64_t)zcopy.max_size;
- zi->tzi_elem_size = (uint64_t)zcopy.elem_size;
- zi->tzi_alloc_size = (uint64_t)zcopy.alloc_size;
- zi->tzi_sum_size = zcopy.sum_count * zcopy.elem_size;
- zi->tzi_exhaustible = (uint64_t)zcopy.exhaustible;
- zi->tzi_collectable = (uint64_t)zcopy.collectable;
- zi->tzi_caller_acct = (uint64_t)zcopy.caller_acct;
- if (task->tkm_zinfo != NULL) {
- zi->tzi_task_alloc = task->tkm_zinfo[zcopy.index].alloc;
- zi->tzi_task_free = task->tkm_zinfo[zcopy.index].free;
- } else {
- zi->tzi_task_alloc = 0;
- zi->tzi_task_free = 0;
+ used_zones = max_zones;
+ for (i = 0; i < max_zones; i++) {
+ if (!get_zone_info(&(zone_array[i]), zn, zi)) {
+ used_zones--;
+ continue;
}
+ zones_collectable_bytes += GET_MZI_COLLECTABLE_BYTES(zi->mzi_collectable);
zn++;
zi++;
}
- /*
- * loop through the fake zones and fill them using the specialized
- * functions
- */
- for (i = 0; i < num_fake_zones; i++) {
- int count, collectable, exhaustible, caller_acct, index;
- vm_size_t cur_size, max_size, elem_size, alloc_size;
- uint64_t sum_size;
-
- strncpy(zn->mzn_name, fake_zones[i].name, sizeof zn->mzn_name);
- zn->mzn_name[sizeof zn->mzn_name - 1] = '\0';
- fake_zones[i].query(&count, &cur_size,
- &max_size, &elem_size,
- &alloc_size, &sum_size,
- &collectable, &exhaustible, &caller_acct);
- zi->tzi_count = (uint64_t)count;
- zi->tzi_cur_size = (uint64_t)cur_size;
- zi->tzi_max_size = (uint64_t)max_size;
- zi->tzi_elem_size = (uint64_t)elem_size;
- zi->tzi_alloc_size = (uint64_t)alloc_size;
- zi->tzi_sum_size = sum_size;
- zi->tzi_collectable = (uint64_t)collectable;
- zi->tzi_exhaustible = (uint64_t)exhaustible;
- zi->tzi_caller_acct = (uint64_t)caller_acct;
- if (task->tkm_zinfo != NULL) {
- index = ZINFO_SLOTS - num_fake_zones + i;
- zi->tzi_task_alloc = task->tkm_zinfo[index].alloc;
- zi->tzi_task_free = task->tkm_zinfo[index].free;
- } else {
- zi->tzi_task_alloc = 0;
- zi->tzi_task_free = 0;
- }
- zn++;
- zi++;
- }
+ *namesp = (mach_zone_name_t *) create_vm_map_copy(names_addr, names_size, used_zones * sizeof *names);
+ *namesCntp = used_zones;
- used = max_zones * sizeof *names;
- if (used != names_size)
- bzero((char *) (names_addr + used), names_size - used);
+ *infop = (mach_zone_info_t *) create_vm_map_copy(info_addr, info_size, used_zones * sizeof *info);
+ *infoCntp = used_zones;
- kr = vm_map_copyin(ipc_kernel_map, (vm_map_address_t)names_addr,
- (vm_map_size_t)used, TRUE, ©);
- assert(kr == KERN_SUCCESS);
+ num_info = 0;
+ memory_info_addr = 0;
- *namesp = (mach_zone_name_t *) copy;
- *namesCntp = max_zones;
+ if (memoryInfop && memoryInfoCntp) {
+ vm_map_copy_t copy;
+ num_info = vm_page_diagnose_estimate();
+ memory_info_size = num_info * sizeof(*memory_info);
+ memory_info_vmsize = round_page(memory_info_size);
+ kr = kmem_alloc_pageable(ipc_kernel_map,
+ &memory_info_addr, memory_info_vmsize, VM_KERN_MEMORY_IPC);
+ if (kr != KERN_SUCCESS) {
+ return kr;
+ }
- used = max_zones * sizeof *info;
+ kr = vm_map_wire_kernel(ipc_kernel_map, memory_info_addr, memory_info_addr + memory_info_vmsize,
+ VM_PROT_READ | VM_PROT_WRITE, VM_KERN_MEMORY_IPC, FALSE);
+ assert(kr == KERN_SUCCESS);
- if (used != info_size)
- bzero((char *) (info_addr + used), info_size - used);
+ memory_info = (mach_memory_info_t *) memory_info_addr;
+ vm_page_diagnose(memory_info, num_info, zones_collectable_bytes);
- kr = vm_map_copyin(ipc_kernel_map, (vm_map_address_t)info_addr,
- (vm_map_size_t)used, TRUE, ©);
- assert(kr == KERN_SUCCESS);
+ kr = vm_map_unwire(ipc_kernel_map, memory_info_addr, memory_info_addr + memory_info_vmsize, FALSE);
+ assert(kr == KERN_SUCCESS);
+
+ kr = vm_map_copyin(ipc_kernel_map, (vm_map_address_t)memory_info_addr,
+ (vm_map_size_t)memory_info_size, TRUE, ©);
+ assert(kr == KERN_SUCCESS);
- *infop = (task_zone_info_t *) copy;
- *infoCntp = max_zones;
+ *memoryInfop = (mach_memory_info_t *) copy;
+ *memoryInfoCntp = num_info;
+ }
return KERN_SUCCESS;
}
-#else /* CONFIG_TASK_ZONE_INFO */
-
kern_return_t
-task_zone_info(
- __unused task_t task,
- __unused mach_zone_name_array_t *namesp,
- __unused mach_msg_type_number_t *namesCntp,
- __unused task_zone_info_array_t *infop,
- __unused mach_msg_type_number_t *infoCntp)
+mach_zone_info_for_zone(
+ host_priv_t host,
+ mach_zone_name_t name,
+ mach_zone_info_t *infop)
{
- return KERN_FAILURE;
-}
+ zone_t zone_ptr;
-#endif /* CONFIG_TASK_ZONE_INFO */
+ if (host == HOST_NULL) {
+ return KERN_INVALID_HOST;
+ }
+#if CONFIG_DEBUGGER_FOR_ZONE_INFO
+ if (!PE_i_can_has_debugger(NULL)) {
+ return KERN_INVALID_HOST;
+ }
+#endif
-kern_return_t
-mach_zone_info(
- host_priv_t host,
- mach_zone_name_array_t *namesp,
- mach_msg_type_number_t *namesCntp,
- mach_zone_info_array_t *infop,
- mach_msg_type_number_t *infoCntp)
-{
- return (mach_memory_info(host, namesp, namesCntp, infop, infoCntp, NULL, NULL));
+ if (infop == NULL) {
+ return KERN_INVALID_ARGUMENT;
+ }
+
+ zone_ptr = ZONE_NULL;
+ zone_index_foreach(i) {
+ zone_t z = &(zone_array[i]);
+ assert(z != ZONE_NULL);
+
+ /*
+ * Append kalloc heap name to zone name (if zone is used by kalloc)
+ */
+ char temp_zone_name[MAX_ZONE_NAME] = "";
+ snprintf(temp_zone_name, MAX_ZONE_NAME, "%s%s",
+ zone_heap_name(z), z->z_name);
+
+ /* Find the requested zone by name */
+ if (track_this_zone(temp_zone_name, name.mzn_name)) {
+ zone_ptr = z;
+ break;
+ }
+ }
+
+ /* No zones found with the requested zone name */
+ if (zone_ptr == ZONE_NULL) {
+ return KERN_INVALID_ARGUMENT;
+ }
+
+ if (get_zone_info(zone_ptr, NULL, infop)) {
+ return KERN_SUCCESS;
+ }
+ return KERN_FAILURE;
}
kern_return_t
-mach_memory_info(
- host_priv_t host,
- mach_zone_name_array_t *namesp,
- mach_msg_type_number_t *namesCntp,
- mach_zone_info_array_t *infop,
- mach_msg_type_number_t *infoCntp,
- mach_memory_info_array_t *memoryInfop,
- mach_msg_type_number_t *memoryInfoCntp)
+mach_zone_info_for_largest_zone(
+ host_priv_t host,
+ mach_zone_name_t *namep,
+ mach_zone_info_t *infop)
{
- mach_zone_name_t *names;
- vm_offset_t names_addr;
- vm_size_t names_size;
+ if (host == HOST_NULL) {
+ return KERN_INVALID_HOST;
+ }
+#if CONFIG_DEBUGGER_FOR_ZONE_INFO
+ if (!PE_i_can_has_debugger(NULL)) {
+ return KERN_INVALID_HOST;
+ }
+#endif
+
+ if (namep == NULL || infop == NULL) {
+ return KERN_INVALID_ARGUMENT;
+ }
- mach_zone_info_t *info;
- vm_offset_t info_addr;
- vm_size_t info_size;
+ if (get_zone_info(zone_find_largest(), namep, infop)) {
+ return KERN_SUCCESS;
+ }
+ return KERN_FAILURE;
+}
- mach_memory_info_t *memory_info;
- vm_offset_t memory_info_addr;
- vm_size_t memory_info_size;
- vm_size_t memory_info_vmsize;
- unsigned int num_sites;
+uint64_t
+get_zones_collectable_bytes(void)
+{
+ uint64_t zones_collectable_bytes = 0;
+ mach_zone_info_t zi;
- unsigned int max_zones, i;
- zone_t z;
- mach_zone_name_t *zn;
- mach_zone_info_t *zi;
- kern_return_t kr;
-
- vm_size_t used;
- vm_map_copy_t copy;
+ zone_index_foreach(i) {
+ if (get_zone_info(&zone_array[i], NULL, &zi)) {
+ zones_collectable_bytes +=
+ GET_MZI_COLLECTABLE_BYTES(zi.mzi_collectable);
+ }
+ }
+ return zones_collectable_bytes;
+}
- if (host == HOST_NULL)
- return KERN_INVALID_HOST;
-#if CONFIG_DEBUGGER_FOR_ZONE_INFO
- if (!PE_i_can_has_debugger(NULL))
+kern_return_t
+mach_zone_get_zlog_zones(
+ host_priv_t host,
+ mach_zone_name_array_t *namesp,
+ mach_msg_type_number_t *namesCntp)
+{
+#if ZONE_ENABLE_LOGGING
+ unsigned int max_zones, logged_zones, i;
+ kern_return_t kr;
+ zone_t zone_ptr;
+ mach_zone_name_t *names;
+ vm_offset_t names_addr;
+ vm_size_t names_size;
+
+ if (host == HOST_NULL) {
return KERN_INVALID_HOST;
-#endif
+ }
- /*
- * We assume that zones aren't freed once allocated.
- * We won't pick up any zones that are allocated later.
- */
+ if (namesp == NULL || namesCntp == NULL) {
+ return KERN_INVALID_ARGUMENT;
+ }
- simple_lock(&all_zones_lock);
- max_zones = (unsigned int)(num_zones + num_fake_zones);
- z = first_zone;
- simple_unlock(&all_zones_lock);
+ max_zones = os_atomic_load(&num_zones, relaxed);
names_size = round_page(max_zones * sizeof *names);
kr = kmem_alloc_pageable(ipc_kernel_map,
- &names_addr, names_size, VM_KERN_MEMORY_IPC);
- if (kr != KERN_SUCCESS)
- return kr;
- names = (mach_zone_name_t *) names_addr;
-
- info_size = round_page(max_zones * sizeof *info);
- kr = kmem_alloc_pageable(ipc_kernel_map,
- &info_addr, info_size, VM_KERN_MEMORY_IPC);
+ &names_addr, names_size, VM_KERN_MEMORY_IPC);
if (kr != KERN_SUCCESS) {
- kmem_free(ipc_kernel_map,
- names_addr, names_size);
return kr;
}
- info = (mach_zone_info_t *) info_addr;
+ names = (mach_zone_name_t *) names_addr;
- num_sites = 0;
- memory_info_addr = 0;
- if (memoryInfop && memoryInfoCntp)
- {
- num_sites = VM_KERN_MEMORY_COUNT + VM_KERN_COUNTER_COUNT;
- memory_info_size = num_sites * sizeof(*info);
- memory_info_vmsize = round_page(memory_info_size);
- kr = kmem_alloc_pageable(ipc_kernel_map,
- &memory_info_addr, memory_info_vmsize, VM_KERN_MEMORY_IPC);
- if (kr != KERN_SUCCESS) {
- kmem_free(ipc_kernel_map,
- names_addr, names_size);
- kmem_free(ipc_kernel_map,
- info_addr, info_size);
- return kr;
+ zone_ptr = ZONE_NULL;
+ logged_zones = 0;
+ for (i = 0; i < max_zones; i++) {
+ zone_t z = &(zone_array[i]);
+ assert(z != ZONE_NULL);
+
+ /* Copy out the zone name if zone logging is enabled */
+ if (z->zlog_btlog) {
+ get_zone_info(z, &names[logged_zones], NULL);
+ logged_zones++;
}
+ }
- kr = vm_map_wire(ipc_kernel_map, memory_info_addr, memory_info_addr + memory_info_vmsize,
- VM_PROT_READ|VM_PROT_WRITE|VM_PROT_MEMORY_TAG_MAKE(VM_KERN_MEMORY_IPC), FALSE);
- assert(kr == KERN_SUCCESS);
+ *namesp = (mach_zone_name_t *) create_vm_map_copy(names_addr, names_size, logged_zones * sizeof *names);
+ *namesCntp = logged_zones;
- memory_info = (mach_memory_info_t *) memory_info_addr;
- vm_page_diagnose(memory_info, num_sites);
+ return KERN_SUCCESS;
- kr = vm_map_unwire(ipc_kernel_map, memory_info_addr, memory_info_addr + memory_info_vmsize, FALSE);
- assert(kr == KERN_SUCCESS);
+#else /* ZONE_ENABLE_LOGGING */
+#pragma unused(host, namesp, namesCntp)
+ return KERN_FAILURE;
+#endif /* ZONE_ENABLE_LOGGING */
+}
+
+kern_return_t
+mach_zone_get_btlog_records(
+ host_priv_t host,
+ mach_zone_name_t name,
+ zone_btrecord_array_t *recsp,
+ mach_msg_type_number_t *recsCntp)
+{
+#if DEBUG || DEVELOPMENT
+ unsigned int numrecs = 0;
+ zone_btrecord_t *recs;
+ kern_return_t kr;
+ zone_t zone_ptr;
+ vm_offset_t recs_addr;
+ vm_size_t recs_size;
+
+ if (host == HOST_NULL) {
+ return KERN_INVALID_HOST;
}
- zn = &names[0];
- zi = &info[0];
+ if (recsp == NULL || recsCntp == NULL) {
+ return KERN_INVALID_ARGUMENT;
+ }
- for (i = 0; i < max_zones - num_fake_zones; i++) {
- struct zone zcopy;
+ zone_ptr = ZONE_NULL;
+ zone_index_foreach(i) {
+ zone_t z = &zone_array[i];
- assert(z != ZONE_NULL);
+ /*
+ * Append kalloc heap name to zone name (if zone is used by kalloc)
+ */
+ char temp_zone_name[MAX_ZONE_NAME] = "";
+ snprintf(temp_zone_name, MAX_ZONE_NAME, "%s%s",
+ zone_heap_name(z), z->z_name);
- lock_zone(z);
- zcopy = *z;
- unlock_zone(z);
+ /* Find the requested zone by name */
+ if (track_this_zone(temp_zone_name, name.mzn_name)) {
+ zone_ptr = z;
+ break;
+ }
+ }
- simple_lock(&all_zones_lock);
- z = z->next_zone;
- simple_unlock(&all_zones_lock);
+ /* No zones found with the requested zone name */
+ if (zone_ptr == ZONE_NULL) {
+ return KERN_INVALID_ARGUMENT;
+ }
- /* assuming here the name data is static */
- (void) strncpy(zn->mzn_name, zcopy.zone_name,
- sizeof zn->mzn_name);
- zn->mzn_name[sizeof zn->mzn_name - 1] = '\0';
-
- zi->mzi_count = (uint64_t)zcopy.count;
- zi->mzi_cur_size = ptoa_64(zcopy.page_count);
- zi->mzi_max_size = (uint64_t)zcopy.max_size;
- zi->mzi_elem_size = (uint64_t)zcopy.elem_size;
- zi->mzi_alloc_size = (uint64_t)zcopy.alloc_size;
- zi->mzi_sum_size = zcopy.sum_count * zcopy.elem_size;
- zi->mzi_exhaustible = (uint64_t)zcopy.exhaustible;
- zi->mzi_collectable = (uint64_t)zcopy.collectable;
- zn++;
- zi++;
+ /* Logging not turned on for the requested zone */
+ if (!DO_LOGGING(zone_ptr)) {
+ return KERN_FAILURE;
}
- /*
- * loop through the fake zones and fill them using the specialized
- * functions
- */
- for (i = 0; i < num_fake_zones; i++) {
- int count, collectable, exhaustible, caller_acct;
- vm_size_t cur_size, max_size, elem_size, alloc_size;
- uint64_t sum_size;
-
- strncpy(zn->mzn_name, fake_zones[i].name, sizeof zn->mzn_name);
- zn->mzn_name[sizeof zn->mzn_name - 1] = '\0';
- fake_zones[i].query(&count, &cur_size,
- &max_size, &elem_size,
- &alloc_size, &sum_size,
- &collectable, &exhaustible, &caller_acct);
- zi->mzi_count = (uint64_t)count;
- zi->mzi_cur_size = (uint64_t)cur_size;
- zi->mzi_max_size = (uint64_t)max_size;
- zi->mzi_elem_size = (uint64_t)elem_size;
- zi->mzi_alloc_size = (uint64_t)alloc_size;
- zi->mzi_sum_size = sum_size;
- zi->mzi_collectable = (uint64_t)collectable;
- zi->mzi_exhaustible = (uint64_t)exhaustible;
+ /* Allocate memory for btlog records */
+ numrecs = (unsigned int)(get_btlog_records_count(zone_ptr->zlog_btlog));
+ recs_size = round_page(numrecs * sizeof *recs);
- zn++;
- zi++;
+ kr = kmem_alloc_pageable(ipc_kernel_map, &recs_addr, recs_size, VM_KERN_MEMORY_IPC);
+ if (kr != KERN_SUCCESS) {
+ return kr;
}
- used = max_zones * sizeof *names;
- if (used != names_size)
- bzero((char *) (names_addr + used), names_size - used);
+ /*
+ * We will call get_btlog_records() below which populates this region while holding a spinlock
+ * (the btlog lock). So these pages need to be wired.
+ */
+ kr = vm_map_wire_kernel(ipc_kernel_map, recs_addr, recs_addr + recs_size,
+ VM_PROT_READ | VM_PROT_WRITE, VM_KERN_MEMORY_IPC, FALSE);
+ assert(kr == KERN_SUCCESS);
+
+ recs = (zone_btrecord_t *)recs_addr;
+ get_btlog_records(zone_ptr->zlog_btlog, recs, &numrecs);
- kr = vm_map_copyin(ipc_kernel_map, (vm_map_address_t)names_addr,
- (vm_map_size_t)used, TRUE, ©);
+ kr = vm_map_unwire(ipc_kernel_map, recs_addr, recs_addr + recs_size, FALSE);
assert(kr == KERN_SUCCESS);
- *namesp = (mach_zone_name_t *) copy;
- *namesCntp = max_zones;
+ *recsp = (zone_btrecord_t *) create_vm_map_copy(recs_addr, recs_size, numrecs * sizeof *recs);
+ *recsCntp = numrecs;
+
+ return KERN_SUCCESS;
+
+#else /* DEBUG || DEVELOPMENT */
+#pragma unused(host, name, recsp, recsCntp)
+ return KERN_FAILURE;
+#endif /* DEBUG || DEVELOPMENT */
+}
- used = max_zones * sizeof *info;
- if (used != info_size)
- bzero((char *) (info_addr + used), info_size - used);
+#if DEBUG || DEVELOPMENT
- kr = vm_map_copyin(ipc_kernel_map, (vm_map_address_t)info_addr,
- (vm_map_size_t)used, TRUE, ©);
+kern_return_t
+mach_memory_info_check(void)
+{
+ mach_memory_info_t * memory_info;
+ mach_memory_info_t * info;
+ unsigned int num_info;
+ vm_offset_t memory_info_addr;
+ kern_return_t kr;
+ size_t memory_info_size, memory_info_vmsize;
+ uint64_t top_wired, zonestotal, total;
+
+ num_info = vm_page_diagnose_estimate();
+ memory_info_size = num_info * sizeof(*memory_info);
+ memory_info_vmsize = round_page(memory_info_size);
+ kr = kmem_alloc(kernel_map, &memory_info_addr, memory_info_vmsize, VM_KERN_MEMORY_DIAG);
assert(kr == KERN_SUCCESS);
- *infop = (mach_zone_info_t *) copy;
- *infoCntp = max_zones;
+ memory_info = (mach_memory_info_t *) memory_info_addr;
+ vm_page_diagnose(memory_info, num_info, 0);
- if (memoryInfop && memoryInfoCntp)
- {
- kr = vm_map_copyin(ipc_kernel_map, (vm_map_address_t)memory_info_addr,
- (vm_map_size_t)memory_info_size, TRUE, ©);
- assert(kr == KERN_SUCCESS);
+ top_wired = total = zonestotal = 0;
+ zone_index_foreach(idx) {
+ zonestotal += zone_size_wired(&zone_array[idx]);
+ }
- *memoryInfop = (mach_memory_info_t *) copy;
- *memoryInfoCntp = num_sites;
+ for (uint32_t idx = 0; idx < num_info; idx++) {
+ info = &memory_info[idx];
+ if (!info->size) {
+ continue;
+ }
+ if (VM_KERN_COUNT_WIRED == info->site) {
+ top_wired = info->size;
+ }
+ if (VM_KERN_SITE_HIDE & info->flags) {
+ continue;
+ }
+ if (!(VM_KERN_SITE_WIRED & info->flags)) {
+ continue;
+ }
+ total += info->size;
}
+ total += zonestotal;
- return KERN_SUCCESS;
+ printf("vm_page_diagnose_check %qd of %qd, zones %qd, short 0x%qx\n",
+ total, top_wired, zonestotal, top_wired - total);
+
+ kmem_free(kernel_map, memory_info_addr, memory_info_vmsize);
+
+ return kr;
}
-/*
- * host_zone_info - LEGACY user interface for Mach zone information
- * Should use mach_zone_info() instead!
- */
+extern boolean_t(*volatile consider_buffer_cache_collect)(int);
+
+#endif /* DEBUG || DEVELOPMENT */
+
kern_return_t
-host_zone_info(
- host_priv_t host,
- zone_name_array_t *namesp,
- mach_msg_type_number_t *namesCntp,
- zone_info_array_t *infop,
- mach_msg_type_number_t *infoCntp)
+mach_zone_force_gc(
+ host_t host)
{
- zone_name_t *names;
- vm_offset_t names_addr;
- vm_size_t names_size;
- zone_info_t *info;
- vm_offset_t info_addr;
- vm_size_t info_size;
- unsigned int max_zones, i;
- zone_t z;
- zone_name_t *zn;
- zone_info_t *zi;
- kern_return_t kr;
-
- vm_size_t used;
- vm_map_copy_t copy;
+ if (host == HOST_NULL) {
+ return KERN_INVALID_HOST;
+ }
+#if DEBUG || DEVELOPMENT
+ /* Callout to buffer cache GC to drop elements in the apfs zones */
+ if (consider_buffer_cache_collect != NULL) {
+ (void)(*consider_buffer_cache_collect)(0);
+ }
+ consider_zone_gc(FALSE);
+#endif /* DEBUG || DEVELOPMENT */
+ return KERN_SUCCESS;
+}
- if (host == HOST_NULL)
- return KERN_INVALID_HOST;
-#if CONFIG_DEBUGGER_FOR_ZONE_INFO
- if (!PE_i_can_has_debugger(NULL))
- return KERN_INVALID_HOST;
-#endif
+zone_t
+zone_find_largest(void)
+{
+ uint32_t largest_idx = 0;
+ vm_offset_t largest_size = zone_size_wired(&zone_array[0]);
+
+ zone_index_foreach(i) {
+ vm_offset_t size = zone_size_wired(&zone_array[i]);
+ if (size > largest_size) {
+ largest_idx = i;
+ largest_size = size;
+ }
+ }
-#if defined(__LP64__)
- if (!thread_is_64bit(current_thread()))
- return KERN_NOT_SUPPORTED;
-#else
- if (thread_is_64bit(current_thread()))
- return KERN_NOT_SUPPORTED;
-#endif
+ return &zone_array[largest_idx];
+}
- /*
- * We assume that zones aren't freed once allocated.
- * We won't pick up any zones that are allocated later.
- */
+#pragma mark - tests
+#if DEBUG || DEVELOPMENT
- simple_lock(&all_zones_lock);
- max_zones = (unsigned int)(num_zones + num_fake_zones);
- z = first_zone;
- simple_unlock(&all_zones_lock);
+/*
+ * Used for sysctl kern.run_zone_test which is not thread-safe. Ensure only one
+ * thread goes through at a time. Or we can end up with multiple test zones (if
+ * a second zinit() comes through before zdestroy()), which could lead us to
+ * run out of zones.
+ */
+SIMPLE_LOCK_DECLARE(zone_test_lock, 0);
+static boolean_t zone_test_running = FALSE;
+static zone_t test_zone_ptr = NULL;
- names_size = round_page(max_zones * sizeof *names);
- kr = kmem_alloc_pageable(ipc_kernel_map,
- &names_addr, names_size, VM_KERN_MEMORY_IPC);
- if (kr != KERN_SUCCESS)
- return kr;
- names = (zone_name_t *) names_addr;
+static uintptr_t *
+zone_copy_allocations(zone_t z, uintptr_t *elems, bitmap_t *bits,
+ zone_pva_t page_index, zone_addr_kind_t kind)
+{
+ vm_offset_t free, first, end, page;
+ struct zone_page_metadata *meta;
- info_size = round_page(max_zones * sizeof *info);
- kr = kmem_alloc_pageable(ipc_kernel_map,
- &info_addr, info_size, VM_KERN_MEMORY_IPC);
- if (kr != KERN_SUCCESS) {
- kmem_free(ipc_kernel_map,
- names_addr, names_size);
- return kr;
- }
+ while (!zone_pva_is_null(page_index)) {
+ page = zone_pva_to_addr(page_index);
+ meta = zone_pva_to_meta(page_index, kind);
+ end = page + ptoa(meta->zm_percpu ? 1 : meta->zm_page_count);
+ first = page + ZONE_PAGE_FIRST_OFFSET(kind);
- info = (zone_info_t *) info_addr;
+ bitmap_clear(bits, (uint32_t)((end - first) / zone_elem_size(z)));
- zn = &names[0];
- zi = &info[0];
+ // construct bitmap of all freed elements
+ free = zone_page_meta_get_freelist(z, meta, page);
+ while (free) {
+ bitmap_set(bits, (uint32_t)((free - first) / zone_elem_size(z)));
- for (i = 0; i < max_zones - num_fake_zones; i++) {
- struct zone zcopy;
+ // next free element
+ free = *(vm_offset_t *)free ^ zp_nopoison_cookie;
+ }
- assert(z != ZONE_NULL);
+ for (unsigned i = 0; first < end; i++, first += zone_elem_size(z)) {
+ if (!bitmap_test(bits, i)) {
+ *elems++ = INSTANCE_PUT(first);
+ }
+ }
- lock_zone(z);
- zcopy = *z;
- unlock_zone(z);
+ page_index = meta->zm_page_next;
+ }
+ return elems;
+}
- simple_lock(&all_zones_lock);
- z = z->next_zone;
- simple_unlock(&all_zones_lock);
+kern_return_t
+zone_leaks(const char * zoneName, uint32_t nameLen, leak_site_proc proc, void * refCon)
+{
+ uintptr_t zbt[MAX_ZTRACE_DEPTH];
+ zone_t zone = NULL;
+ uintptr_t * array;
+ uintptr_t * next;
+ uintptr_t element, bt;
+ uint32_t idx, count, found;
+ uint32_t btidx, btcount, nobtcount, btfound;
+ uint32_t elemSize;
+ uint64_t maxElems;
+ kern_return_t kr;
+ bitmap_t *bits;
+
+ zone_index_foreach(i) {
+ if (!strncmp(zoneName, zone_array[i].z_name, nameLen)) {
+ zone = &zone_array[i];
+ break;
+ }
+ }
+ if (zone == NULL) {
+ return KERN_INVALID_NAME;
+ }
- /* assuming here the name data is static */
- (void) strncpy(zn->zn_name, zcopy.zone_name,
- sizeof zn->zn_name);
- zn->zn_name[sizeof zn->zn_name - 1] = '\0';
-
- zi->zi_count = zcopy.count;
- zi->zi_cur_size = ptoa(zcopy.page_count);
- zi->zi_max_size = zcopy.max_size;
- zi->zi_elem_size = zcopy.elem_size;
- zi->zi_alloc_size = zcopy.alloc_size;
- zi->zi_exhaustible = zcopy.exhaustible;
- zi->zi_collectable = zcopy.collectable;
+ elemSize = zone_elem_size(zone);
+ maxElems = (zone->countavail + 1) & ~1ul;
- zn++;
- zi++;
+ if ((ptoa(zone->percpu ? 1 : zone->alloc_pages) % elemSize) &&
+ !zone_leaks_scan_enable) {
+ return KERN_INVALID_CAPABILITY;
}
- /*
- * loop through the fake zones and fill them using the specialized
- * functions
- */
- for (i = 0; i < num_fake_zones; i++) {
- int caller_acct;
- uint64_t sum_space;
- strncpy(zn->zn_name, fake_zones[i].name, sizeof zn->zn_name);
- zn->zn_name[sizeof zn->zn_name - 1] = '\0';
- fake_zones[i].query(&zi->zi_count, &zi->zi_cur_size,
- &zi->zi_max_size, &zi->zi_elem_size,
- &zi->zi_alloc_size, &sum_space,
- &zi->zi_collectable, &zi->zi_exhaustible, &caller_acct);
- zn++;
- zi++;
+ kr = kmem_alloc_kobject(kernel_map, (vm_offset_t *) &array,
+ maxElems * sizeof(uintptr_t) + BITMAP_LEN(ZONE_CHUNK_MAXELEMENTS),
+ VM_KERN_MEMORY_DIAG);
+ if (KERN_SUCCESS != kr) {
+ return kr;
}
- used = max_zones * sizeof *names;
- if (used != names_size)
- bzero((char *) (names_addr + used), names_size - used);
+ /* maxElems is a 2-multiple so we're always aligned */
+ bits = CAST_DOWN_EXPLICIT(bitmap_t *, array + maxElems);
- kr = vm_map_copyin(ipc_kernel_map, (vm_map_address_t)names_addr,
- (vm_map_size_t)used, TRUE, ©);
- assert(kr == KERN_SUCCESS);
+ lock_zone(zone);
- *namesp = (zone_name_t *) copy;
- *namesCntp = max_zones;
+ next = array;
+ next = zone_copy_allocations(zone, next, bits,
+ zone->pages_any_free_foreign, ZONE_ADDR_FOREIGN);
+ next = zone_copy_allocations(zone, next, bits,
+ zone->pages_all_used_foreign, ZONE_ADDR_FOREIGN);
+ next = zone_copy_allocations(zone, next, bits,
+ zone->pages_intermediate, ZONE_ADDR_NATIVE);
+ next = zone_copy_allocations(zone, next, bits,
+ zone->pages_all_used, ZONE_ADDR_NATIVE);
+ count = (uint32_t)(next - array);
- used = max_zones * sizeof *info;
- if (used != info_size)
- bzero((char *) (info_addr + used), info_size - used);
+ unlock_zone(zone);
- kr = vm_map_copyin(ipc_kernel_map, (vm_map_address_t)info_addr,
- (vm_map_size_t)used, TRUE, ©);
- assert(kr == KERN_SUCCESS);
+ zone_leaks_scan(array, count, zone_elem_size(zone), &found);
+ assert(found <= count);
- *infop = (zone_info_t *) copy;
- *infoCntp = max_zones;
+ for (idx = 0; idx < count; idx++) {
+ element = array[idx];
+ if (kInstanceFlagReferenced & element) {
+ continue;
+ }
+ element = INSTANCE_PUT(element) & ~kInstanceFlags;
+ }
- return KERN_SUCCESS;
-}
+#if ZONE_ENABLE_LOGGING
+ if (zone->zlog_btlog && !corruption_debug_flag) {
+ // btlog_copy_backtraces_for_elements will set kInstanceFlagReferenced on elements it found
+ btlog_copy_backtraces_for_elements(zone->zlog_btlog, array, &count, elemSize, proc, refCon);
+ }
+#endif /* ZONE_ENABLE_LOGGING */
-kern_return_t
-mach_zone_force_gc(
- host_t host)
-{
+ for (nobtcount = idx = 0; idx < count; idx++) {
+ element = array[idx];
+ if (!element) {
+ continue;
+ }
+ if (kInstanceFlagReferenced & element) {
+ continue;
+ }
+ element = INSTANCE_PUT(element) & ~kInstanceFlags;
- if (host == HOST_NULL)
- return KERN_INVALID_HOST;
+ // see if we can find any backtrace left in the element
+ btcount = (typeof(btcount))(zone_elem_size(zone) / sizeof(uintptr_t));
+ if (btcount >= MAX_ZTRACE_DEPTH) {
+ btcount = MAX_ZTRACE_DEPTH - 1;
+ }
+ for (btfound = btidx = 0; btidx < btcount; btidx++) {
+ bt = ((uintptr_t *)element)[btcount - 1 - btidx];
+ if (!VM_KERNEL_IS_SLID(bt)) {
+ break;
+ }
+ zbt[btfound++] = bt;
+ }
+ if (btfound) {
+ (*proc)(refCon, 1, elemSize, &zbt[0], btfound);
+ } else {
+ nobtcount++;
+ }
+ }
+ if (nobtcount) {
+ // fake backtrace when we found nothing
+ zbt[0] = (uintptr_t) &zalloc;
+ (*proc)(refCon, nobtcount, elemSize, &zbt[0], 1);
+ }
- consider_zone_gc(TRUE);
+ kmem_free(kernel_map, (vm_offset_t) array, maxElems * sizeof(uintptr_t));
- return (KERN_SUCCESS);
+ return KERN_SUCCESS;
}
-extern unsigned int stack_total;
-extern unsigned long long stack_allocs;
-
-#if defined(__i386__) || defined (__x86_64__)
-extern unsigned int inuse_ptepages_count;
-extern long long alloc_ptepages_count;
-#endif
-
-void zone_display_zprint()
+boolean_t
+run_zone_test(void)
{
- unsigned int i;
- zone_t the_zone;
+ unsigned int i = 0, max_iter = 5;
+ void * test_ptr;
+ zone_t test_zone;
- if(first_zone!=NULL) {
- the_zone = first_zone;
- for (i = 0; i < num_zones; i++) {
- if(the_zone->cur_size > (1024*1024)) {
- printf("%.20s:\t%lu\n",the_zone->zone_name,(uintptr_t)the_zone->cur_size);
- }
+ simple_lock(&zone_test_lock, &zone_locks_grp);
+ if (!zone_test_running) {
+ zone_test_running = TRUE;
+ } else {
+ simple_unlock(&zone_test_lock);
+ printf("run_zone_test: Test already running.\n");
+ return FALSE;
+ }
+ simple_unlock(&zone_test_lock);
- if(the_zone->next_zone == NULL) {
- break;
- }
+ printf("run_zone_test: Testing zinit(), zalloc(), zfree() and zdestroy() on zone \"test_zone_sysctl\"\n");
- the_zone = the_zone->next_zone;
+ /* zinit() and zdestroy() a zone with the same name a bunch of times, verify that we get back the same zone each time */
+ do {
+ test_zone = zinit(sizeof(uint64_t), 100 * sizeof(uint64_t), sizeof(uint64_t), "test_zone_sysctl");
+ if (test_zone == NULL) {
+ printf("run_zone_test: zinit() failed\n");
+ return FALSE;
}
- }
-
- printf("Kernel Stacks:\t%lu\n",(uintptr_t)(kernel_stack_size * stack_total));
-#if defined(__i386__) || defined (__x86_64__)
- printf("PageTables:\t%lu\n",(uintptr_t)(PAGE_SIZE * inuse_ptepages_count));
+#if KASAN_ZALLOC
+ if (test_zone_ptr == NULL && test_zone->countfree != 0) {
+#else
+ if (test_zone->countfree != 0) {
#endif
+ printf("run_zone_test: free count is not zero\n");
+ return FALSE;
+ }
- printf("Kalloc.Large:\t%lu\n",(uintptr_t)kalloc_large_total);
-}
-
-zone_t
-zone_find_largest(void)
-{
- unsigned int i;
- unsigned int max_zones;
- zone_t the_zone;
- zone_t zone_largest;
+ if (test_zone_ptr == NULL) {
+ /* Stash the zone pointer returned on the fist zinit */
+ printf("run_zone_test: zone created for the first time\n");
+ test_zone_ptr = test_zone;
+ } else if (test_zone != test_zone_ptr) {
+ printf("run_zone_test: old zone pointer and new zone pointer don't match\n");
+ return FALSE;
+ }
- simple_lock(&all_zones_lock);
- the_zone = first_zone;
- max_zones = num_zones;
- simple_unlock(&all_zones_lock);
-
- zone_largest = the_zone;
- for (i = 0; i < max_zones; i++) {
- if (the_zone->cur_size > zone_largest->cur_size) {
- zone_largest = the_zone;
+ test_ptr = zalloc(test_zone);
+ if (test_ptr == NULL) {
+ printf("run_zone_test: zalloc() failed\n");
+ return FALSE;
+ }
+ zfree(test_zone, test_ptr);
+
+ zdestroy(test_zone);
+ i++;
+
+ printf("run_zone_test: Iteration %d successful\n", i);
+ } while (i < max_iter);
+
+ /* test Z_VA_SEQUESTER */
+ if (zsecurity_options & ZSECURITY_OPTIONS_SEQUESTER) {
+ int idx, num_allocs = 8;
+ vm_size_t elem_size = 2 * PAGE_SIZE / num_allocs;
+ void *allocs[num_allocs];
+ vm_offset_t phys_pages = os_atomic_load(&zones_phys_page_count, relaxed);
+ vm_size_t zone_map_size = zone_range_size(&zone_info.zi_map_range);
+
+ test_zone = zone_create("test_zone_sysctl", elem_size,
+ ZC_DESTRUCTIBLE | ZC_SEQUESTER);
+ if (test_zone == NULL) {
+ printf("run_zone_test: zinit() failed\n");
+ return FALSE;
}
- if (the_zone->next_zone == NULL) {
- break;
+ for (idx = 0; idx < num_allocs; idx++) {
+ allocs[idx] = zalloc(test_zone);
+ assert(NULL != allocs[idx]);
+ printf("alloc[%d] %p\n", idx, allocs[idx]);
+ }
+ for (idx = 0; idx < num_allocs; idx++) {
+ zfree(test_zone, allocs[idx]);
+ }
+ assert(!zone_pva_is_null(test_zone->pages_all_free));
+
+ printf("vm_page_wire_count %d, vm_page_free_count %d, p to v %qd%%\n",
+ vm_page_wire_count, vm_page_free_count,
+ (100ULL * ptoa_64(phys_pages)) / zone_map_size);
+ zone_gc(FALSE);
+ printf("vm_page_wire_count %d, vm_page_free_count %d, p to v %qd%%\n",
+ vm_page_wire_count, vm_page_free_count,
+ (100ULL * ptoa_64(phys_pages)) / zone_map_size);
+ unsigned int allva = 0;
+ zone_index_foreach(zidx) {
+ zone_t z = &zone_array[zidx];
+ lock_zone(z);
+ allva += z->page_count;
+ if (!z->sequester_page_count) {
+ unlock_zone(z);
+ continue;
+ }
+ unsigned count = 0;
+ uint64_t size;
+ zone_pva_t pg = z->pages_sequester;
+ struct zone_page_metadata *page_meta;
+ while (pg.packed_address) {
+ page_meta = zone_pva_to_meta(pg, ZONE_ADDR_NATIVE);
+ count += z->alloc_pages;
+ pg = page_meta->zm_page_next;
+ }
+ assert(count == z->sequester_page_count);
+ size = zone_size_wired(z);
+ if (!size) {
+ size = 1;
+ }
+ printf("%s%s: seq %d, res %d, %qd %%\n",
+ zone_heap_name(z), z->z_name, z->sequester_page_count,
+ z->page_count, zone_size_allocated(z) * 100ULL / size);
+ unlock_zone(z);
}
- the_zone = the_zone->next_zone;
+ printf("total va: %d\n", allva);
+
+ assert(zone_pva_is_null(test_zone->pages_all_free));
+ assert(!zone_pva_is_null(test_zone->pages_sequester));
+ assert(2 == test_zone->sequester_page_count);
+ for (idx = 0; idx < num_allocs; idx++) {
+ assert(0 == pmap_find_phys(kernel_pmap, (addr64_t)(uintptr_t) allocs[idx]));
+ }
+ for (idx = 0; idx < num_allocs; idx++) {
+ allocs[idx] = zalloc(test_zone);
+ assert(allocs[idx]);
+ printf("alloc[%d] %p\n", idx, allocs[idx]);
+ }
+ assert(zone_pva_is_null(test_zone->pages_sequester));
+ assert(0 == test_zone->sequester_page_count);
+ for (idx = 0; idx < num_allocs; idx++) {
+ zfree(test_zone, allocs[idx]);
+ }
+ zdestroy(test_zone);
+ } else {
+ printf("run_zone_test: skipping sequester test (not enabled)\n");
}
- return zone_largest;
-}
-#if ZONE_DEBUG
+ printf("run_zone_test: Test passed\n");
+
+ simple_lock(&zone_test_lock, &zone_locks_grp);
+ zone_test_running = FALSE;
+ simple_unlock(&zone_test_lock);
-/* should we care about locks here ? */
+ return TRUE;
+}
-#define zone_in_use(z) ( z->count || z->free_elements \
- || !queue_empty(&z->pages.all_free) \
- || !queue_empty(&z->pages.intermediate) \
- || (z->allows_foreign && !queue_empty(&z->pages.any_free_foreign)))
+/*
+ * Routines to test that zone garbage collection and zone replenish threads
+ * running at the same time don't cause problems.
+ */
void
-zone_debug_enable(
- zone_t z)
+zone_gc_replenish_test(void)
{
- if (zone_debug_enabled(z) || zone_in_use(z) ||
- z->alloc_size < (z->elem_size + ZONE_DEBUG_OFFSET))
- return;
- queue_init(&z->active_zones);
- z->elem_size += ZONE_DEBUG_OFFSET;
+ zone_gc(FALSE);
}
+
void
-zone_debug_disable(
- zone_t z)
+zone_alloc_replenish_test(void)
{
- if (!zone_debug_enabled(z) || zone_in_use(z))
+ zone_t z = NULL;
+ struct data { struct data *next; } *node, *list = NULL;
+
+ /*
+ * Find a zone that has a replenish thread
+ */
+ zone_index_foreach(i) {
+ z = &zone_array[i];
+ if (z->prio_refill_count &&
+ zone_elem_size(z) >= sizeof(struct data)) {
+ z = &zone_array[i];
+ break;
+ }
+ }
+ if (z == NULL) {
+ printf("Couldn't find a replenish zone\n");
return;
- z->elem_size -= ZONE_DEBUG_OFFSET;
- z->active_zones.next = z->active_zones.prev = NULL;
-}
+ }
+
+ for (uint32_t i = 0; i < 2000; ++i) { /* something big enough to go past replenishment */
+ node = zalloc(z);
+ node->next = list;
+ list = node;
+ }
+ /*
+ * release the memory we allocated
+ */
+ while (list != NULL) {
+ node = list;
+ list = list->next;
+ zfree(z, node);
+ }
+}
-#endif /* ZONE_DEBUG */
+#endif /* DEBUG || DEVELOPMENT */
projects / apple / xnu.git / blobdiff