1 // SPDX-License-Identifier: GPL-2.0
3 * Slab allocator functions that are independent of the allocator strategy
5 * (C) 2012 Christoph Lameter <cl@linux.com>
7 #include <linux/slab.h>
10 #include <linux/poison.h>
11 #include <linux/interrupt.h>
12 #include <linux/memory.h>
13 #include <linux/cache.h>
14 #include <linux/compiler.h>
15 #include <linux/kfence.h>
16 #include <linux/module.h>
17 #include <linux/cpu.h>
18 #include <linux/uaccess.h>
19 #include <linux/seq_file.h>
20 #include <linux/proc_fs.h>
21 #include <linux/debugfs.h>
22 #include <linux/kasan.h>
23 #include <asm/cacheflush.h>
24 #include <asm/tlbflush.h>
26 #include <linux/memcontrol.h>
27 #include <linux/stackdepot.h>
32 #define CREATE_TRACE_POINTS
33 #include <trace/events/kmem.h>
35 enum slab_state slab_state;
36 LIST_HEAD(slab_caches);
37 DEFINE_MUTEX(slab_mutex);
38 struct kmem_cache *kmem_cache;
40 static LIST_HEAD(slab_caches_to_rcu_destroy);
41 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work);
42 static DECLARE_WORK(slab_caches_to_rcu_destroy_work,
43 slab_caches_to_rcu_destroy_workfn);
46 * Set of flags that will prevent slab merging
48 #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
49 SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \
50 SLAB_FAILSLAB | kasan_never_merge())
52 #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \
53 SLAB_CACHE_DMA32 | SLAB_ACCOUNT)
56 * Merge control. If this is set then no merging of slab caches will occur.
58 static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT);
60 static int __init setup_slab_nomerge(char *str)
66 static int __init setup_slab_merge(char *str)
73 __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0);
74 __setup_param("slub_merge", slub_merge, setup_slab_merge, 0);
77 __setup("slab_nomerge", setup_slab_nomerge);
78 __setup("slab_merge", setup_slab_merge);
81 * Determine the size of a slab object
83 unsigned int kmem_cache_size(struct kmem_cache *s)
85 return s->object_size;
87 EXPORT_SYMBOL(kmem_cache_size);
89 #ifdef CONFIG_DEBUG_VM
90 static int kmem_cache_sanity_check(const char *name, unsigned int size)
92 if (!name || in_interrupt() || size > KMALLOC_MAX_SIZE) {
93 pr_err("kmem_cache_create(%s) integrity check failed\n", name);
97 WARN_ON(strchr(name, ' ')); /* It confuses parsers */
101 static inline int kmem_cache_sanity_check(const char *name, unsigned int size)
108 * Figure out what the alignment of the objects will be given a set of
109 * flags, a user specified alignment and the size of the objects.
111 static unsigned int calculate_alignment(slab_flags_t flags,
112 unsigned int align, unsigned int size)
115 * If the user wants hardware cache aligned objects then follow that
116 * suggestion if the object is sufficiently large.
118 * The hardware cache alignment cannot override the specified
119 * alignment though. If that is greater then use it.
121 if (flags & SLAB_HWCACHE_ALIGN) {
124 ralign = cache_line_size();
125 while (size <= ralign / 2)
127 align = max(align, ralign);
130 align = max(align, arch_slab_minalign());
132 return ALIGN(align, sizeof(void *));
136 * Find a mergeable slab cache
138 int slab_unmergeable(struct kmem_cache *s)
140 if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
150 * We may have set a slab to be unmergeable during bootstrap.
158 struct kmem_cache *find_mergeable(unsigned int size, unsigned int align,
159 slab_flags_t flags, const char *name, void (*ctor)(void *))
161 struct kmem_cache *s;
169 size = ALIGN(size, sizeof(void *));
170 align = calculate_alignment(flags, align, size);
171 size = ALIGN(size, align);
172 flags = kmem_cache_flags(size, flags, name);
174 if (flags & SLAB_NEVER_MERGE)
177 list_for_each_entry_reverse(s, &slab_caches, list) {
178 if (slab_unmergeable(s))
184 if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
187 * Check if alignment is compatible.
188 * Courtesy of Adrian Drzewiecki
190 if ((s->size & ~(align - 1)) != s->size)
193 if (s->size - size >= sizeof(void *))
196 if (IS_ENABLED(CONFIG_SLAB) && align &&
197 (align > s->align || s->align % align))
205 static struct kmem_cache *create_cache(const char *name,
206 unsigned int object_size, unsigned int align,
207 slab_flags_t flags, unsigned int useroffset,
208 unsigned int usersize, void (*ctor)(void *),
209 struct kmem_cache *root_cache)
211 struct kmem_cache *s;
214 if (WARN_ON(useroffset + usersize > object_size))
215 useroffset = usersize = 0;
218 s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
223 s->size = s->object_size = object_size;
226 s->useroffset = useroffset;
227 s->usersize = usersize;
229 err = __kmem_cache_create(s, flags);
234 list_add(&s->list, &slab_caches);
241 kmem_cache_free(kmem_cache, s);
246 * kmem_cache_create_usercopy - Create a cache with a region suitable
247 * for copying to userspace
248 * @name: A string which is used in /proc/slabinfo to identify this cache.
249 * @size: The size of objects to be created in this cache.
250 * @align: The required alignment for the objects.
252 * @useroffset: Usercopy region offset
253 * @usersize: Usercopy region size
254 * @ctor: A constructor for the objects.
256 * Cannot be called within a interrupt, but can be interrupted.
257 * The @ctor is run when new pages are allocated by the cache.
261 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
262 * to catch references to uninitialised memory.
264 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
265 * for buffer overruns.
267 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
268 * cacheline. This can be beneficial if you're counting cycles as closely
271 * Return: a pointer to the cache on success, NULL on failure.
274 kmem_cache_create_usercopy(const char *name,
275 unsigned int size, unsigned int align,
277 unsigned int useroffset, unsigned int usersize,
278 void (*ctor)(void *))
280 struct kmem_cache *s = NULL;
281 const char *cache_name;
284 #ifdef CONFIG_SLUB_DEBUG
286 * If no slub_debug was enabled globally, the static key is not yet
287 * enabled by setup_slub_debug(). Enable it if the cache is being
288 * created with any of the debugging flags passed explicitly.
289 * It's also possible that this is the first cache created with
290 * SLAB_STORE_USER and we should init stack_depot for it.
292 if (flags & SLAB_DEBUG_FLAGS)
293 static_branch_enable(&slub_debug_enabled);
294 if (flags & SLAB_STORE_USER)
298 mutex_lock(&slab_mutex);
300 err = kmem_cache_sanity_check(name, size);
305 /* Refuse requests with allocator specific flags */
306 if (flags & ~SLAB_FLAGS_PERMITTED) {
312 * Some allocators will constraint the set of valid flags to a subset
313 * of all flags. We expect them to define CACHE_CREATE_MASK in this
314 * case, and we'll just provide them with a sanitized version of the
317 flags &= CACHE_CREATE_MASK;
319 /* Fail closed on bad usersize of useroffset values. */
320 if (WARN_ON(!usersize && useroffset) ||
321 WARN_ON(size < usersize || size - usersize < useroffset))
322 usersize = useroffset = 0;
325 s = __kmem_cache_alias(name, size, align, flags, ctor);
329 cache_name = kstrdup_const(name, GFP_KERNEL);
335 s = create_cache(cache_name, size,
336 calculate_alignment(flags, align, size),
337 flags, useroffset, usersize, ctor, NULL);
340 kfree_const(cache_name);
344 mutex_unlock(&slab_mutex);
347 if (flags & SLAB_PANIC)
348 panic("%s: Failed to create slab '%s'. Error %d\n",
349 __func__, name, err);
351 pr_warn("%s(%s) failed with error %d\n",
352 __func__, name, err);
359 EXPORT_SYMBOL(kmem_cache_create_usercopy);
362 * kmem_cache_create - Create a cache.
363 * @name: A string which is used in /proc/slabinfo to identify this cache.
364 * @size: The size of objects to be created in this cache.
365 * @align: The required alignment for the objects.
367 * @ctor: A constructor for the objects.
369 * Cannot be called within a interrupt, but can be interrupted.
370 * The @ctor is run when new pages are allocated by the cache.
374 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
375 * to catch references to uninitialised memory.
377 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
378 * for buffer overruns.
380 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
381 * cacheline. This can be beneficial if you're counting cycles as closely
384 * Return: a pointer to the cache on success, NULL on failure.
387 kmem_cache_create(const char *name, unsigned int size, unsigned int align,
388 slab_flags_t flags, void (*ctor)(void *))
390 return kmem_cache_create_usercopy(name, size, align, flags, 0, 0,
393 EXPORT_SYMBOL(kmem_cache_create);
395 #ifdef SLAB_SUPPORTS_SYSFS
397 * For a given kmem_cache, kmem_cache_destroy() should only be called
398 * once or there will be a use-after-free problem. The actual deletion
399 * and release of the kobject does not need slab_mutex or cpu_hotplug_lock
400 * protection. So they are now done without holding those locks.
402 * Note that there will be a slight delay in the deletion of sysfs files
403 * if kmem_cache_release() is called indrectly from a work function.
405 static void kmem_cache_release(struct kmem_cache *s)
407 sysfs_slab_unlink(s);
408 sysfs_slab_release(s);
411 static void kmem_cache_release(struct kmem_cache *s)
413 slab_kmem_cache_release(s);
417 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work)
419 LIST_HEAD(to_destroy);
420 struct kmem_cache *s, *s2;
423 * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the
424 * @slab_caches_to_rcu_destroy list. The slab pages are freed
425 * through RCU and the associated kmem_cache are dereferenced
426 * while freeing the pages, so the kmem_caches should be freed only
427 * after the pending RCU operations are finished. As rcu_barrier()
428 * is a pretty slow operation, we batch all pending destructions
431 mutex_lock(&slab_mutex);
432 list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy);
433 mutex_unlock(&slab_mutex);
435 if (list_empty(&to_destroy))
440 list_for_each_entry_safe(s, s2, &to_destroy, list) {
441 debugfs_slab_release(s);
442 kfence_shutdown_cache(s);
443 kmem_cache_release(s);
447 static int shutdown_cache(struct kmem_cache *s)
449 /* free asan quarantined objects */
450 kasan_cache_shutdown(s);
452 if (__kmem_cache_shutdown(s) != 0)
457 if (s->flags & SLAB_TYPESAFE_BY_RCU) {
458 list_add_tail(&s->list, &slab_caches_to_rcu_destroy);
459 schedule_work(&slab_caches_to_rcu_destroy_work);
461 kfence_shutdown_cache(s);
462 debugfs_slab_release(s);
468 void slab_kmem_cache_release(struct kmem_cache *s)
470 __kmem_cache_release(s);
471 kfree_const(s->name);
472 kmem_cache_free(kmem_cache, s);
475 void kmem_cache_destroy(struct kmem_cache *s)
480 if (unlikely(!s) || !kasan_check_byte(s))
484 mutex_lock(&slab_mutex);
486 rcu_set = s->flags & SLAB_TYPESAFE_BY_RCU;
488 refcnt = --s->refcount;
492 WARN(shutdown_cache(s),
493 "%s %s: Slab cache still has objects when called from %pS",
494 __func__, s->name, (void *)_RET_IP_);
496 mutex_unlock(&slab_mutex);
498 if (!refcnt && !rcu_set)
499 kmem_cache_release(s);
501 EXPORT_SYMBOL(kmem_cache_destroy);
504 * kmem_cache_shrink - Shrink a cache.
505 * @cachep: The cache to shrink.
507 * Releases as many slabs as possible for a cache.
508 * To help debugging, a zero exit status indicates all slabs were released.
510 * Return: %0 if all slabs were released, non-zero otherwise
512 int kmem_cache_shrink(struct kmem_cache *cachep)
514 kasan_cache_shrink(cachep);
516 return __kmem_cache_shrink(cachep);
518 EXPORT_SYMBOL(kmem_cache_shrink);
520 bool slab_is_available(void)
522 return slab_state >= UP;
527 * kmem_valid_obj - does the pointer reference a valid slab object?
528 * @object: pointer to query.
530 * Return: %true if the pointer is to a not-yet-freed object from
531 * kmalloc() or kmem_cache_alloc(), either %true or %false if the pointer
532 * is to an already-freed object, and %false otherwise.
534 bool kmem_valid_obj(void *object)
538 /* Some arches consider ZERO_SIZE_PTR to be a valid address. */
539 if (object < (void *)PAGE_SIZE || !virt_addr_valid(object))
541 folio = virt_to_folio(object);
542 return folio_test_slab(folio);
544 EXPORT_SYMBOL_GPL(kmem_valid_obj);
546 static void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab)
548 if (__kfence_obj_info(kpp, object, slab))
550 __kmem_obj_info(kpp, object, slab);
554 * kmem_dump_obj - Print available slab provenance information
555 * @object: slab object for which to find provenance information.
557 * This function uses pr_cont(), so that the caller is expected to have
558 * printed out whatever preamble is appropriate. The provenance information
559 * depends on the type of object and on how much debugging is enabled.
560 * For a slab-cache object, the fact that it is a slab object is printed,
561 * and, if available, the slab name, return address, and stack trace from
562 * the allocation and last free path of that object.
564 * This function will splat if passed a pointer to a non-slab object.
565 * If you are not sure what type of object you have, you should instead
566 * use mem_dump_obj().
568 void kmem_dump_obj(void *object)
570 char *cp = IS_ENABLED(CONFIG_MMU) ? "" : "/vmalloc";
573 unsigned long ptroffset;
574 struct kmem_obj_info kp = { };
576 if (WARN_ON_ONCE(!virt_addr_valid(object)))
578 slab = virt_to_slab(object);
579 if (WARN_ON_ONCE(!slab)) {
580 pr_cont(" non-slab memory.\n");
583 kmem_obj_info(&kp, object, slab);
584 if (kp.kp_slab_cache)
585 pr_cont(" slab%s %s", cp, kp.kp_slab_cache->name);
587 pr_cont(" slab%s", cp);
588 if (is_kfence_address(object))
589 pr_cont(" (kfence)");
591 pr_cont(" start %px", kp.kp_objp);
592 if (kp.kp_data_offset)
593 pr_cont(" data offset %lu", kp.kp_data_offset);
595 ptroffset = ((char *)object - (char *)kp.kp_objp) - kp.kp_data_offset;
596 pr_cont(" pointer offset %lu", ptroffset);
598 if (kp.kp_slab_cache && kp.kp_slab_cache->usersize)
599 pr_cont(" size %u", kp.kp_slab_cache->usersize);
601 pr_cont(" allocated at %pS\n", kp.kp_ret);
604 for (i = 0; i < ARRAY_SIZE(kp.kp_stack); i++) {
607 pr_info(" %pS\n", kp.kp_stack[i]);
610 if (kp.kp_free_stack[0])
611 pr_cont(" Free path:\n");
613 for (i = 0; i < ARRAY_SIZE(kp.kp_free_stack); i++) {
614 if (!kp.kp_free_stack[i])
616 pr_info(" %pS\n", kp.kp_free_stack[i]);
620 EXPORT_SYMBOL_GPL(kmem_dump_obj);
624 /* Create a cache during boot when no slab services are available yet */
625 void __init create_boot_cache(struct kmem_cache *s, const char *name,
626 unsigned int size, slab_flags_t flags,
627 unsigned int useroffset, unsigned int usersize)
630 unsigned int align = ARCH_KMALLOC_MINALIGN;
633 s->size = s->object_size = size;
636 * For power of two sizes, guarantee natural alignment for kmalloc
637 * caches, regardless of SL*B debugging options.
639 if (is_power_of_2(size))
640 align = max(align, size);
641 s->align = calculate_alignment(flags, align, size);
643 s->useroffset = useroffset;
644 s->usersize = usersize;
646 err = __kmem_cache_create(s, flags);
649 panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n",
652 s->refcount = -1; /* Exempt from merging for now */
655 struct kmem_cache *__init create_kmalloc_cache(const char *name,
656 unsigned int size, slab_flags_t flags,
657 unsigned int useroffset, unsigned int usersize)
659 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
662 panic("Out of memory when creating slab %s\n", name);
664 create_boot_cache(s, name, size, flags | SLAB_KMALLOC, useroffset,
666 kasan_cache_create_kmalloc(s);
667 list_add(&s->list, &slab_caches);
673 kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1] __ro_after_init =
674 { /* initialization for https://bugs.llvm.org/show_bug.cgi?id=42570 */ };
675 EXPORT_SYMBOL(kmalloc_caches);
678 * Conversion table for small slabs sizes / 8 to the index in the
679 * kmalloc array. This is necessary for slabs < 192 since we have non power
680 * of two cache sizes there. The size of larger slabs can be determined using
683 static u8 size_index[24] __ro_after_init = {
710 static inline unsigned int size_index_elem(unsigned int bytes)
712 return (bytes - 1) / 8;
716 * Find the kmem_cache structure that serves a given size of
719 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
725 return ZERO_SIZE_PTR;
727 index = size_index[size_index_elem(size)];
729 if (WARN_ON_ONCE(size > KMALLOC_MAX_CACHE_SIZE))
731 index = fls(size - 1);
734 return kmalloc_caches[kmalloc_type(flags)][index];
737 size_t kmalloc_size_roundup(size_t size)
739 struct kmem_cache *c;
741 /* Short-circuit the 0 size case. */
742 if (unlikely(size == 0))
744 /* Short-circuit saturated "too-large" case. */
745 if (unlikely(size == SIZE_MAX))
747 /* Above the smaller buckets, size is a multiple of page size. */
748 if (size > KMALLOC_MAX_CACHE_SIZE)
749 return PAGE_SIZE << get_order(size);
751 /* The flags don't matter since size_index is common to all. */
752 c = kmalloc_slab(size, GFP_KERNEL);
753 return c ? c->object_size : 0;
755 EXPORT_SYMBOL(kmalloc_size_roundup);
757 #ifdef CONFIG_ZONE_DMA
758 #define KMALLOC_DMA_NAME(sz) .name[KMALLOC_DMA] = "dma-kmalloc-" #sz,
760 #define KMALLOC_DMA_NAME(sz)
763 #ifdef CONFIG_MEMCG_KMEM
764 #define KMALLOC_CGROUP_NAME(sz) .name[KMALLOC_CGROUP] = "kmalloc-cg-" #sz,
766 #define KMALLOC_CGROUP_NAME(sz)
769 #define INIT_KMALLOC_INFO(__size, __short_size) \
771 .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \
772 .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #__short_size, \
773 KMALLOC_CGROUP_NAME(__short_size) \
774 KMALLOC_DMA_NAME(__short_size) \
779 * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
780 * kmalloc_index() supports up to 2^21=2MB, so the final entry of the table is
783 const struct kmalloc_info_struct kmalloc_info[] __initconst = {
784 INIT_KMALLOC_INFO(0, 0),
785 INIT_KMALLOC_INFO(96, 96),
786 INIT_KMALLOC_INFO(192, 192),
787 INIT_KMALLOC_INFO(8, 8),
788 INIT_KMALLOC_INFO(16, 16),
789 INIT_KMALLOC_INFO(32, 32),
790 INIT_KMALLOC_INFO(64, 64),
791 INIT_KMALLOC_INFO(128, 128),
792 INIT_KMALLOC_INFO(256, 256),
793 INIT_KMALLOC_INFO(512, 512),
794 INIT_KMALLOC_INFO(1024, 1k),
795 INIT_KMALLOC_INFO(2048, 2k),
796 INIT_KMALLOC_INFO(4096, 4k),
797 INIT_KMALLOC_INFO(8192, 8k),
798 INIT_KMALLOC_INFO(16384, 16k),
799 INIT_KMALLOC_INFO(32768, 32k),
800 INIT_KMALLOC_INFO(65536, 64k),
801 INIT_KMALLOC_INFO(131072, 128k),
802 INIT_KMALLOC_INFO(262144, 256k),
803 INIT_KMALLOC_INFO(524288, 512k),
804 INIT_KMALLOC_INFO(1048576, 1M),
805 INIT_KMALLOC_INFO(2097152, 2M)
809 * Patch up the size_index table if we have strange large alignment
810 * requirements for the kmalloc array. This is only the case for
811 * MIPS it seems. The standard arches will not generate any code here.
813 * Largest permitted alignment is 256 bytes due to the way we
814 * handle the index determination for the smaller caches.
816 * Make sure that nothing crazy happens if someone starts tinkering
817 * around with ARCH_KMALLOC_MINALIGN
819 void __init setup_kmalloc_cache_index_table(void)
823 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
824 !is_power_of_2(KMALLOC_MIN_SIZE));
826 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
827 unsigned int elem = size_index_elem(i);
829 if (elem >= ARRAY_SIZE(size_index))
831 size_index[elem] = KMALLOC_SHIFT_LOW;
834 if (KMALLOC_MIN_SIZE >= 64) {
836 * The 96 byte sized cache is not used if the alignment
839 for (i = 64 + 8; i <= 96; i += 8)
840 size_index[size_index_elem(i)] = 7;
844 if (KMALLOC_MIN_SIZE >= 128) {
846 * The 192 byte sized cache is not used if the alignment
847 * is 128 byte. Redirect kmalloc to use the 256 byte cache
850 for (i = 128 + 8; i <= 192; i += 8)
851 size_index[size_index_elem(i)] = 8;
856 new_kmalloc_cache(int idx, enum kmalloc_cache_type type, slab_flags_t flags)
858 if (type == KMALLOC_RECLAIM) {
859 flags |= SLAB_RECLAIM_ACCOUNT;
860 } else if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_CGROUP)) {
861 if (mem_cgroup_kmem_disabled()) {
862 kmalloc_caches[type][idx] = kmalloc_caches[KMALLOC_NORMAL][idx];
865 flags |= SLAB_ACCOUNT;
866 } else if (IS_ENABLED(CONFIG_ZONE_DMA) && (type == KMALLOC_DMA)) {
867 flags |= SLAB_CACHE_DMA;
870 kmalloc_caches[type][idx] = create_kmalloc_cache(
871 kmalloc_info[idx].name[type],
872 kmalloc_info[idx].size, flags, 0,
873 kmalloc_info[idx].size);
876 * If CONFIG_MEMCG_KMEM is enabled, disable cache merging for
877 * KMALLOC_NORMAL caches.
879 if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_NORMAL))
880 kmalloc_caches[type][idx]->refcount = -1;
884 * Create the kmalloc array. Some of the regular kmalloc arrays
885 * may already have been created because they were needed to
886 * enable allocations for slab creation.
888 void __init create_kmalloc_caches(slab_flags_t flags)
891 enum kmalloc_cache_type type;
894 * Including KMALLOC_CGROUP if CONFIG_MEMCG_KMEM defined
896 for (type = KMALLOC_NORMAL; type < NR_KMALLOC_TYPES; type++) {
897 for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
898 if (!kmalloc_caches[type][i])
899 new_kmalloc_cache(i, type, flags);
902 * Caches that are not of the two-to-the-power-of size.
903 * These have to be created immediately after the
904 * earlier power of two caches
906 if (KMALLOC_MIN_SIZE <= 32 && i == 6 &&
907 !kmalloc_caches[type][1])
908 new_kmalloc_cache(1, type, flags);
909 if (KMALLOC_MIN_SIZE <= 64 && i == 7 &&
910 !kmalloc_caches[type][2])
911 new_kmalloc_cache(2, type, flags);
915 /* Kmalloc array is now usable */
919 void free_large_kmalloc(struct folio *folio, void *object)
921 unsigned int order = folio_order(folio);
923 if (WARN_ON_ONCE(order == 0))
924 pr_warn_once("object pointer: 0x%p\n", object);
926 kmemleak_free(object);
927 kasan_kfree_large(object);
928 kmsan_kfree_large(object);
930 mod_lruvec_page_state(folio_page(folio, 0), NR_SLAB_UNRECLAIMABLE_B,
931 -(PAGE_SIZE << order));
932 __free_pages(folio_page(folio, 0), order);
935 static void *__kmalloc_large_node(size_t size, gfp_t flags, int node);
936 static __always_inline
937 void *__do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller)
939 struct kmem_cache *s;
942 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
943 ret = __kmalloc_large_node(size, flags, node);
944 trace_kmalloc(_RET_IP_, ret, size,
945 PAGE_SIZE << get_order(size), flags, node);
949 s = kmalloc_slab(size, flags);
951 if (unlikely(ZERO_OR_NULL_PTR(s)))
954 ret = __kmem_cache_alloc_node(s, flags, node, size, caller);
955 ret = kasan_kmalloc(s, ret, size, flags);
956 trace_kmalloc(_RET_IP_, ret, size, s->size, flags, node);
960 void *__kmalloc_node(size_t size, gfp_t flags, int node)
962 return __do_kmalloc_node(size, flags, node, _RET_IP_);
964 EXPORT_SYMBOL(__kmalloc_node);
966 void *__kmalloc(size_t size, gfp_t flags)
968 return __do_kmalloc_node(size, flags, NUMA_NO_NODE, _RET_IP_);
970 EXPORT_SYMBOL(__kmalloc);
972 void *__kmalloc_node_track_caller(size_t size, gfp_t flags,
973 int node, unsigned long caller)
975 return __do_kmalloc_node(size, flags, node, caller);
977 EXPORT_SYMBOL(__kmalloc_node_track_caller);
980 * kfree - free previously allocated memory
981 * @object: pointer returned by kmalloc.
983 * If @object is NULL, no operation is performed.
985 * Don't free memory not originally allocated by kmalloc()
986 * or you will run into trouble.
988 void kfree(const void *object)
992 struct kmem_cache *s;
994 trace_kfree(_RET_IP_, object);
996 if (unlikely(ZERO_OR_NULL_PTR(object)))
999 folio = virt_to_folio(object);
1000 if (unlikely(!folio_test_slab(folio))) {
1001 free_large_kmalloc(folio, (void *)object);
1005 slab = folio_slab(folio);
1006 s = slab->slab_cache;
1007 __kmem_cache_free(s, (void *)object, _RET_IP_);
1009 EXPORT_SYMBOL(kfree);
1012 * __ksize -- Report full size of underlying allocation
1013 * @objp: pointer to the object
1015 * This should only be used internally to query the true size of allocations.
1016 * It is not meant to be a way to discover the usable size of an allocation
1017 * after the fact. Instead, use kmalloc_size_roundup(). Using memory beyond
1018 * the originally requested allocation size may trigger KASAN, UBSAN_BOUNDS,
1019 * and/or FORTIFY_SOURCE.
1021 * Return: size of the actual memory used by @objp in bytes
1023 size_t __ksize(const void *object)
1025 struct folio *folio;
1027 if (unlikely(object == ZERO_SIZE_PTR))
1030 folio = virt_to_folio(object);
1032 if (unlikely(!folio_test_slab(folio))) {
1033 if (WARN_ON(folio_size(folio) <= KMALLOC_MAX_CACHE_SIZE))
1035 if (WARN_ON(object != folio_address(folio)))
1037 return folio_size(folio);
1040 return slab_ksize(folio_slab(folio)->slab_cache);
1043 #ifdef CONFIG_TRACING
1044 void *kmalloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
1046 void *ret = __kmem_cache_alloc_node(s, gfpflags, NUMA_NO_NODE,
1049 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags, NUMA_NO_NODE);
1051 ret = kasan_kmalloc(s, ret, size, gfpflags);
1054 EXPORT_SYMBOL(kmalloc_trace);
1056 void *kmalloc_node_trace(struct kmem_cache *s, gfp_t gfpflags,
1057 int node, size_t size)
1059 void *ret = __kmem_cache_alloc_node(s, gfpflags, node, size, _RET_IP_);
1061 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags, node);
1063 ret = kasan_kmalloc(s, ret, size, gfpflags);
1066 EXPORT_SYMBOL(kmalloc_node_trace);
1067 #endif /* !CONFIG_TRACING */
1068 #endif /* !CONFIG_SLOB */
1070 gfp_t kmalloc_fix_flags(gfp_t flags)
1072 gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
1074 flags &= ~GFP_SLAB_BUG_MASK;
1075 pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
1076 invalid_mask, &invalid_mask, flags, &flags);
1083 * To avoid unnecessary overhead, we pass through large allocation requests
1084 * directly to the page allocator. We use __GFP_COMP, because we will need to
1085 * know the allocation order to free the pages properly in kfree.
1088 static void *__kmalloc_large_node(size_t size, gfp_t flags, int node)
1092 unsigned int order = get_order(size);
1094 if (unlikely(flags & GFP_SLAB_BUG_MASK))
1095 flags = kmalloc_fix_flags(flags);
1097 flags |= __GFP_COMP;
1098 page = alloc_pages_node(node, flags, order);
1100 ptr = page_address(page);
1101 mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B,
1102 PAGE_SIZE << order);
1105 ptr = kasan_kmalloc_large(ptr, size, flags);
1106 /* As ptr might get tagged, call kmemleak hook after KASAN. */
1107 kmemleak_alloc(ptr, size, 1, flags);
1108 kmsan_kmalloc_large(ptr, size, flags);
1113 void *kmalloc_large(size_t size, gfp_t flags)
1115 void *ret = __kmalloc_large_node(size, flags, NUMA_NO_NODE);
1117 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << get_order(size),
1118 flags, NUMA_NO_NODE);
1121 EXPORT_SYMBOL(kmalloc_large);
1123 void *kmalloc_large_node(size_t size, gfp_t flags, int node)
1125 void *ret = __kmalloc_large_node(size, flags, node);
1127 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << get_order(size),
1131 EXPORT_SYMBOL(kmalloc_large_node);
1133 #ifdef CONFIG_SLAB_FREELIST_RANDOM
1134 /* Randomize a generic freelist */
1135 static void freelist_randomize(struct rnd_state *state, unsigned int *list,
1141 for (i = 0; i < count; i++)
1144 /* Fisher-Yates shuffle */
1145 for (i = count - 1; i > 0; i--) {
1146 rand = prandom_u32_state(state);
1148 swap(list[i], list[rand]);
1152 /* Create a random sequence per cache */
1153 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
1156 struct rnd_state state;
1158 if (count < 2 || cachep->random_seq)
1161 cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp);
1162 if (!cachep->random_seq)
1165 /* Get best entropy at this stage of boot */
1166 prandom_seed_state(&state, get_random_long());
1168 freelist_randomize(&state, cachep->random_seq, count);
1172 /* Destroy the per-cache random freelist sequence */
1173 void cache_random_seq_destroy(struct kmem_cache *cachep)
1175 kfree(cachep->random_seq);
1176 cachep->random_seq = NULL;
1178 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
1180 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
1182 #define SLABINFO_RIGHTS (0600)
1184 #define SLABINFO_RIGHTS (0400)
1187 static void print_slabinfo_header(struct seq_file *m)
1190 * Output format version, so at least we can change it
1191 * without _too_ many complaints.
1193 #ifdef CONFIG_DEBUG_SLAB
1194 seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
1196 seq_puts(m, "slabinfo - version: 2.1\n");
1198 seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
1199 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
1200 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
1201 #ifdef CONFIG_DEBUG_SLAB
1202 seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
1203 seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
1208 static void *slab_start(struct seq_file *m, loff_t *pos)
1210 mutex_lock(&slab_mutex);
1211 return seq_list_start(&slab_caches, *pos);
1214 static void *slab_next(struct seq_file *m, void *p, loff_t *pos)
1216 return seq_list_next(p, &slab_caches, pos);
1219 static void slab_stop(struct seq_file *m, void *p)
1221 mutex_unlock(&slab_mutex);
1224 static void cache_show(struct kmem_cache *s, struct seq_file *m)
1226 struct slabinfo sinfo;
1228 memset(&sinfo, 0, sizeof(sinfo));
1229 get_slabinfo(s, &sinfo);
1231 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
1232 s->name, sinfo.active_objs, sinfo.num_objs, s->size,
1233 sinfo.objects_per_slab, (1 << sinfo.cache_order));
1235 seq_printf(m, " : tunables %4u %4u %4u",
1236 sinfo.limit, sinfo.batchcount, sinfo.shared);
1237 seq_printf(m, " : slabdata %6lu %6lu %6lu",
1238 sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
1239 slabinfo_show_stats(m, s);
1243 static int slab_show(struct seq_file *m, void *p)
1245 struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
1247 if (p == slab_caches.next)
1248 print_slabinfo_header(m);
1253 void dump_unreclaimable_slab(void)
1255 struct kmem_cache *s;
1256 struct slabinfo sinfo;
1259 * Here acquiring slab_mutex is risky since we don't prefer to get
1260 * sleep in oom path. But, without mutex hold, it may introduce a
1262 * Use mutex_trylock to protect the list traverse, dump nothing
1263 * without acquiring the mutex.
1265 if (!mutex_trylock(&slab_mutex)) {
1266 pr_warn("excessive unreclaimable slab but cannot dump stats\n");
1270 pr_info("Unreclaimable slab info:\n");
1271 pr_info("Name Used Total\n");
1273 list_for_each_entry(s, &slab_caches, list) {
1274 if (s->flags & SLAB_RECLAIM_ACCOUNT)
1277 get_slabinfo(s, &sinfo);
1279 if (sinfo.num_objs > 0)
1280 pr_info("%-17s %10luKB %10luKB\n", s->name,
1281 (sinfo.active_objs * s->size) / 1024,
1282 (sinfo.num_objs * s->size) / 1024);
1284 mutex_unlock(&slab_mutex);
1288 * slabinfo_op - iterator that generates /proc/slabinfo
1297 * num-pages-per-slab
1298 * + further values on SMP and with statistics enabled
1300 static const struct seq_operations slabinfo_op = {
1301 .start = slab_start,
1307 static int slabinfo_open(struct inode *inode, struct file *file)
1309 return seq_open(file, &slabinfo_op);
1312 static const struct proc_ops slabinfo_proc_ops = {
1313 .proc_flags = PROC_ENTRY_PERMANENT,
1314 .proc_open = slabinfo_open,
1315 .proc_read = seq_read,
1316 .proc_write = slabinfo_write,
1317 .proc_lseek = seq_lseek,
1318 .proc_release = seq_release,
1321 static int __init slab_proc_init(void)
1323 proc_create("slabinfo", SLABINFO_RIGHTS, NULL, &slabinfo_proc_ops);
1326 module_init(slab_proc_init);
1328 #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */
1330 static __always_inline __realloc_size(2) void *
1331 __do_krealloc(const void *p, size_t new_size, gfp_t flags)
1336 /* Don't use instrumented ksize to allow precise KASAN poisoning. */
1337 if (likely(!ZERO_OR_NULL_PTR(p))) {
1338 if (!kasan_check_byte(p))
1340 ks = kfence_ksize(p) ?: __ksize(p);
1344 /* If the object still fits, repoison it precisely. */
1345 if (ks >= new_size) {
1346 p = kasan_krealloc((void *)p, new_size, flags);
1350 ret = kmalloc_track_caller(new_size, flags);
1352 /* Disable KASAN checks as the object's redzone is accessed. */
1353 kasan_disable_current();
1354 memcpy(ret, kasan_reset_tag(p), ks);
1355 kasan_enable_current();
1362 * krealloc - reallocate memory. The contents will remain unchanged.
1363 * @p: object to reallocate memory for.
1364 * @new_size: how many bytes of memory are required.
1365 * @flags: the type of memory to allocate.
1367 * The contents of the object pointed to are preserved up to the
1368 * lesser of the new and old sizes (__GFP_ZERO flag is effectively ignored).
1369 * If @p is %NULL, krealloc() behaves exactly like kmalloc(). If @new_size
1370 * is 0 and @p is not a %NULL pointer, the object pointed to is freed.
1372 * Return: pointer to the allocated memory or %NULL in case of error
1374 void *krealloc(const void *p, size_t new_size, gfp_t flags)
1378 if (unlikely(!new_size)) {
1380 return ZERO_SIZE_PTR;
1383 ret = __do_krealloc(p, new_size, flags);
1384 if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret))
1389 EXPORT_SYMBOL(krealloc);
1392 * kfree_sensitive - Clear sensitive information in memory before freeing
1393 * @p: object to free memory of
1395 * The memory of the object @p points to is zeroed before freed.
1396 * If @p is %NULL, kfree_sensitive() does nothing.
1398 * Note: this function zeroes the whole allocated buffer which can be a good
1399 * deal bigger than the requested buffer size passed to kmalloc(). So be
1400 * careful when using this function in performance sensitive code.
1402 void kfree_sensitive(const void *p)
1405 void *mem = (void *)p;
1409 memzero_explicit(mem, ks);
1412 EXPORT_SYMBOL(kfree_sensitive);
1415 * ksize - get the actual amount of memory allocated for a given object
1416 * @objp: Pointer to the object
1418 * kmalloc may internally round up allocations and return more memory
1419 * than requested. ksize() can be used to determine the actual amount of
1420 * memory allocated. The caller may use this additional memory, even though
1421 * a smaller amount of memory was initially specified with the kmalloc call.
1422 * The caller must guarantee that objp points to a valid object previously
1423 * allocated with either kmalloc() or kmem_cache_alloc(). The object
1424 * must not be freed during the duration of the call.
1426 * Return: size of the actual memory used by @objp in bytes
1428 size_t ksize(const void *objp)
1433 * We need to first check that the pointer to the object is valid, and
1434 * only then unpoison the memory. The report printed from ksize() is
1435 * more useful, then when it's printed later when the behaviour could
1436 * be undefined due to a potential use-after-free or double-free.
1438 * We use kasan_check_byte(), which is supported for the hardware
1439 * tag-based KASAN mode, unlike kasan_check_read/write().
1441 * If the pointed to memory is invalid, we return 0 to avoid users of
1442 * ksize() writing to and potentially corrupting the memory region.
1444 * We want to perform the check before __ksize(), to avoid potentially
1445 * crashing in __ksize() due to accessing invalid metadata.
1447 if (unlikely(ZERO_OR_NULL_PTR(objp)) || !kasan_check_byte(objp))
1450 size = kfence_ksize(objp) ?: __ksize(objp);
1452 * We assume that ksize callers could use whole allocated area,
1453 * so we need to unpoison this area.
1455 kasan_unpoison_range(objp, size);
1458 EXPORT_SYMBOL(ksize);
1460 /* Tracepoints definitions. */
1461 EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1462 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1463 EXPORT_TRACEPOINT_SYMBOL(kfree);
1464 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);
1466 int should_failslab(struct kmem_cache *s, gfp_t gfpflags)
1468 if (__should_failslab(s, gfpflags))
1472 ALLOW_ERROR_INJECTION(should_failslab, ERRNO);