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>
29 #define CREATE_TRACE_POINTS
30 #include <trace/events/kmem.h>
36 enum slab_state slab_state;
37 LIST_HEAD(slab_caches);
38 DEFINE_MUTEX(slab_mutex);
39 struct kmem_cache *kmem_cache;
41 static LIST_HEAD(slab_caches_to_rcu_destroy);
42 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work);
43 static DECLARE_WORK(slab_caches_to_rcu_destroy_work,
44 slab_caches_to_rcu_destroy_workfn);
47 * Set of flags that will prevent slab merging
49 #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
50 SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \
51 SLAB_FAILSLAB | kasan_never_merge())
53 #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \
54 SLAB_CACHE_DMA32 | SLAB_ACCOUNT)
57 * Merge control. If this is set then no merging of slab caches will occur.
59 static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT);
61 static int __init setup_slab_nomerge(char *str)
67 static int __init setup_slab_merge(char *str)
74 __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0);
75 __setup_param("slub_merge", slub_merge, setup_slab_merge, 0);
78 __setup("slab_nomerge", setup_slab_nomerge);
79 __setup("slab_merge", setup_slab_merge);
82 * Determine the size of a slab object
84 unsigned int kmem_cache_size(struct kmem_cache *s)
86 return s->object_size;
88 EXPORT_SYMBOL(kmem_cache_size);
90 #ifdef CONFIG_DEBUG_VM
91 static int kmem_cache_sanity_check(const char *name, unsigned int size)
93 if (!name || in_interrupt() || size > KMALLOC_MAX_SIZE) {
94 pr_err("kmem_cache_create(%s) integrity check failed\n", name);
98 WARN_ON(strchr(name, ' ')); /* It confuses parsers */
102 static inline int kmem_cache_sanity_check(const char *name, unsigned int size)
108 void __kmem_cache_free_bulk(struct kmem_cache *s, size_t nr, void **p)
112 for (i = 0; i < nr; i++) {
114 kmem_cache_free(s, p[i]);
120 int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t nr,
125 for (i = 0; i < nr; i++) {
126 void *x = p[i] = kmem_cache_alloc(s, flags);
128 __kmem_cache_free_bulk(s, i, p);
136 * Figure out what the alignment of the objects will be given a set of
137 * flags, a user specified alignment and the size of the objects.
139 static unsigned int calculate_alignment(slab_flags_t flags,
140 unsigned int align, unsigned int size)
143 * If the user wants hardware cache aligned objects then follow that
144 * suggestion if the object is sufficiently large.
146 * The hardware cache alignment cannot override the specified
147 * alignment though. If that is greater then use it.
149 if (flags & SLAB_HWCACHE_ALIGN) {
152 ralign = cache_line_size();
153 while (size <= ralign / 2)
155 align = max(align, ralign);
158 if (align < ARCH_SLAB_MINALIGN)
159 align = ARCH_SLAB_MINALIGN;
161 return ALIGN(align, sizeof(void *));
165 * Find a mergeable slab cache
167 int slab_unmergeable(struct kmem_cache *s)
169 if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
179 * We may have set a slab to be unmergeable during bootstrap.
187 struct kmem_cache *find_mergeable(unsigned int size, unsigned int align,
188 slab_flags_t flags, const char *name, void (*ctor)(void *))
190 struct kmem_cache *s;
198 size = ALIGN(size, sizeof(void *));
199 align = calculate_alignment(flags, align, size);
200 size = ALIGN(size, align);
201 flags = kmem_cache_flags(size, flags, name);
203 if (flags & SLAB_NEVER_MERGE)
206 list_for_each_entry_reverse(s, &slab_caches, list) {
207 if (slab_unmergeable(s))
213 if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
216 * Check if alignment is compatible.
217 * Courtesy of Adrian Drzewiecki
219 if ((s->size & ~(align - 1)) != s->size)
222 if (s->size - size >= sizeof(void *))
225 if (IS_ENABLED(CONFIG_SLAB) && align &&
226 (align > s->align || s->align % align))
234 static struct kmem_cache *create_cache(const char *name,
235 unsigned int object_size, unsigned int align,
236 slab_flags_t flags, unsigned int useroffset,
237 unsigned int usersize, void (*ctor)(void *),
238 struct kmem_cache *root_cache)
240 struct kmem_cache *s;
243 if (WARN_ON(useroffset + usersize > object_size))
244 useroffset = usersize = 0;
247 s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
252 s->size = s->object_size = object_size;
255 s->useroffset = useroffset;
256 s->usersize = usersize;
258 err = __kmem_cache_create(s, flags);
263 list_add(&s->list, &slab_caches);
270 kmem_cache_free(kmem_cache, s);
275 * kmem_cache_create_usercopy - Create a cache with a region suitable
276 * for copying to userspace
277 * @name: A string which is used in /proc/slabinfo to identify this cache.
278 * @size: The size of objects to be created in this cache.
279 * @align: The required alignment for the objects.
281 * @useroffset: Usercopy region offset
282 * @usersize: Usercopy region size
283 * @ctor: A constructor for the objects.
285 * Cannot be called within a interrupt, but can be interrupted.
286 * The @ctor is run when new pages are allocated by the cache.
290 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
291 * to catch references to uninitialised memory.
293 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
294 * for buffer overruns.
296 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
297 * cacheline. This can be beneficial if you're counting cycles as closely
300 * Return: a pointer to the cache on success, NULL on failure.
303 kmem_cache_create_usercopy(const char *name,
304 unsigned int size, unsigned int align,
306 unsigned int useroffset, unsigned int usersize,
307 void (*ctor)(void *))
309 struct kmem_cache *s = NULL;
310 const char *cache_name;
313 #ifdef CONFIG_SLUB_DEBUG
315 * If no slub_debug was enabled globally, the static key is not yet
316 * enabled by setup_slub_debug(). Enable it if the cache is being
317 * created with any of the debugging flags passed explicitly.
318 * It's also possible that this is the first cache created with
319 * SLAB_STORE_USER and we should init stack_depot for it.
321 if (flags & SLAB_DEBUG_FLAGS)
322 static_branch_enable(&slub_debug_enabled);
323 if (flags & SLAB_STORE_USER)
327 mutex_lock(&slab_mutex);
329 err = kmem_cache_sanity_check(name, size);
334 /* Refuse requests with allocator specific flags */
335 if (flags & ~SLAB_FLAGS_PERMITTED) {
341 * Some allocators will constraint the set of valid flags to a subset
342 * of all flags. We expect them to define CACHE_CREATE_MASK in this
343 * case, and we'll just provide them with a sanitized version of the
346 flags &= CACHE_CREATE_MASK;
348 /* Fail closed on bad usersize of useroffset values. */
349 if (WARN_ON(!usersize && useroffset) ||
350 WARN_ON(size < usersize || size - usersize < useroffset))
351 usersize = useroffset = 0;
354 s = __kmem_cache_alias(name, size, align, flags, ctor);
358 cache_name = kstrdup_const(name, GFP_KERNEL);
364 s = create_cache(cache_name, size,
365 calculate_alignment(flags, align, size),
366 flags, useroffset, usersize, ctor, NULL);
369 kfree_const(cache_name);
373 mutex_unlock(&slab_mutex);
376 if (flags & SLAB_PANIC)
377 panic("%s: Failed to create slab '%s'. Error %d\n",
378 __func__, name, err);
380 pr_warn("%s(%s) failed with error %d\n",
381 __func__, name, err);
388 EXPORT_SYMBOL(kmem_cache_create_usercopy);
391 * kmem_cache_create - Create a cache.
392 * @name: A string which is used in /proc/slabinfo to identify this cache.
393 * @size: The size of objects to be created in this cache.
394 * @align: The required alignment for the objects.
396 * @ctor: A constructor for the objects.
398 * Cannot be called within a interrupt, but can be interrupted.
399 * The @ctor is run when new pages are allocated by the cache.
403 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
404 * to catch references to uninitialised memory.
406 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
407 * for buffer overruns.
409 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
410 * cacheline. This can be beneficial if you're counting cycles as closely
413 * Return: a pointer to the cache on success, NULL on failure.
416 kmem_cache_create(const char *name, unsigned int size, unsigned int align,
417 slab_flags_t flags, void (*ctor)(void *))
419 return kmem_cache_create_usercopy(name, size, align, flags, 0, 0,
422 EXPORT_SYMBOL(kmem_cache_create);
424 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work)
426 LIST_HEAD(to_destroy);
427 struct kmem_cache *s, *s2;
430 * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the
431 * @slab_caches_to_rcu_destroy list. The slab pages are freed
432 * through RCU and the associated kmem_cache are dereferenced
433 * while freeing the pages, so the kmem_caches should be freed only
434 * after the pending RCU operations are finished. As rcu_barrier()
435 * is a pretty slow operation, we batch all pending destructions
438 mutex_lock(&slab_mutex);
439 list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy);
440 mutex_unlock(&slab_mutex);
442 if (list_empty(&to_destroy))
447 list_for_each_entry_safe(s, s2, &to_destroy, list) {
448 debugfs_slab_release(s);
449 kfence_shutdown_cache(s);
450 #ifdef SLAB_SUPPORTS_SYSFS
451 sysfs_slab_release(s);
453 slab_kmem_cache_release(s);
458 static int shutdown_cache(struct kmem_cache *s)
460 /* free asan quarantined objects */
461 kasan_cache_shutdown(s);
463 if (__kmem_cache_shutdown(s) != 0)
468 if (s->flags & SLAB_TYPESAFE_BY_RCU) {
469 #ifdef SLAB_SUPPORTS_SYSFS
470 sysfs_slab_unlink(s);
472 list_add_tail(&s->list, &slab_caches_to_rcu_destroy);
473 schedule_work(&slab_caches_to_rcu_destroy_work);
475 kfence_shutdown_cache(s);
476 debugfs_slab_release(s);
477 #ifdef SLAB_SUPPORTS_SYSFS
478 sysfs_slab_unlink(s);
479 sysfs_slab_release(s);
481 slab_kmem_cache_release(s);
488 void slab_kmem_cache_release(struct kmem_cache *s)
490 __kmem_cache_release(s);
491 kfree_const(s->name);
492 kmem_cache_free(kmem_cache, s);
495 void kmem_cache_destroy(struct kmem_cache *s)
497 if (unlikely(!s) || !kasan_check_byte(s))
501 mutex_lock(&slab_mutex);
507 WARN(shutdown_cache(s),
508 "%s %s: Slab cache still has objects when called from %pS",
509 __func__, s->name, (void *)_RET_IP_);
511 mutex_unlock(&slab_mutex);
514 EXPORT_SYMBOL(kmem_cache_destroy);
517 * kmem_cache_shrink - Shrink a cache.
518 * @cachep: The cache to shrink.
520 * Releases as many slabs as possible for a cache.
521 * To help debugging, a zero exit status indicates all slabs were released.
523 * Return: %0 if all slabs were released, non-zero otherwise
525 int kmem_cache_shrink(struct kmem_cache *cachep)
530 kasan_cache_shrink(cachep);
531 ret = __kmem_cache_shrink(cachep);
535 EXPORT_SYMBOL(kmem_cache_shrink);
537 bool slab_is_available(void)
539 return slab_state >= UP;
544 * kmem_valid_obj - does the pointer reference a valid slab object?
545 * @object: pointer to query.
547 * Return: %true if the pointer is to a not-yet-freed object from
548 * kmalloc() or kmem_cache_alloc(), either %true or %false if the pointer
549 * is to an already-freed object, and %false otherwise.
551 bool kmem_valid_obj(void *object)
555 /* Some arches consider ZERO_SIZE_PTR to be a valid address. */
556 if (object < (void *)PAGE_SIZE || !virt_addr_valid(object))
558 folio = virt_to_folio(object);
559 return folio_test_slab(folio);
561 EXPORT_SYMBOL_GPL(kmem_valid_obj);
563 static void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab)
565 if (__kfence_obj_info(kpp, object, slab))
567 __kmem_obj_info(kpp, object, slab);
571 * kmem_dump_obj - Print available slab provenance information
572 * @object: slab object for which to find provenance information.
574 * This function uses pr_cont(), so that the caller is expected to have
575 * printed out whatever preamble is appropriate. The provenance information
576 * depends on the type of object and on how much debugging is enabled.
577 * For a slab-cache object, the fact that it is a slab object is printed,
578 * and, if available, the slab name, return address, and stack trace from
579 * the allocation and last free path of that object.
581 * This function will splat if passed a pointer to a non-slab object.
582 * If you are not sure what type of object you have, you should instead
583 * use mem_dump_obj().
585 void kmem_dump_obj(void *object)
587 char *cp = IS_ENABLED(CONFIG_MMU) ? "" : "/vmalloc";
590 unsigned long ptroffset;
591 struct kmem_obj_info kp = { };
593 if (WARN_ON_ONCE(!virt_addr_valid(object)))
595 slab = virt_to_slab(object);
596 if (WARN_ON_ONCE(!slab)) {
597 pr_cont(" non-slab memory.\n");
600 kmem_obj_info(&kp, object, slab);
601 if (kp.kp_slab_cache)
602 pr_cont(" slab%s %s", cp, kp.kp_slab_cache->name);
604 pr_cont(" slab%s", cp);
605 if (is_kfence_address(object))
606 pr_cont(" (kfence)");
608 pr_cont(" start %px", kp.kp_objp);
609 if (kp.kp_data_offset)
610 pr_cont(" data offset %lu", kp.kp_data_offset);
612 ptroffset = ((char *)object - (char *)kp.kp_objp) - kp.kp_data_offset;
613 pr_cont(" pointer offset %lu", ptroffset);
615 if (kp.kp_slab_cache && kp.kp_slab_cache->usersize)
616 pr_cont(" size %u", kp.kp_slab_cache->usersize);
618 pr_cont(" allocated at %pS\n", kp.kp_ret);
621 for (i = 0; i < ARRAY_SIZE(kp.kp_stack); i++) {
624 pr_info(" %pS\n", kp.kp_stack[i]);
627 if (kp.kp_free_stack[0])
628 pr_cont(" Free path:\n");
630 for (i = 0; i < ARRAY_SIZE(kp.kp_free_stack); i++) {
631 if (!kp.kp_free_stack[i])
633 pr_info(" %pS\n", kp.kp_free_stack[i]);
637 EXPORT_SYMBOL_GPL(kmem_dump_obj);
641 /* Create a cache during boot when no slab services are available yet */
642 void __init create_boot_cache(struct kmem_cache *s, const char *name,
643 unsigned int size, slab_flags_t flags,
644 unsigned int useroffset, unsigned int usersize)
647 unsigned int align = ARCH_KMALLOC_MINALIGN;
650 s->size = s->object_size = size;
653 * For power of two sizes, guarantee natural alignment for kmalloc
654 * caches, regardless of SL*B debugging options.
656 if (is_power_of_2(size))
657 align = max(align, size);
658 s->align = calculate_alignment(flags, align, size);
660 s->useroffset = useroffset;
661 s->usersize = usersize;
663 err = __kmem_cache_create(s, flags);
666 panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n",
669 s->refcount = -1; /* Exempt from merging for now */
672 struct kmem_cache *__init create_kmalloc_cache(const char *name,
673 unsigned int size, slab_flags_t flags,
674 unsigned int useroffset, unsigned int usersize)
676 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
679 panic("Out of memory when creating slab %s\n", name);
681 create_boot_cache(s, name, size, flags, useroffset, usersize);
682 kasan_cache_create_kmalloc(s);
683 list_add(&s->list, &slab_caches);
689 kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1] __ro_after_init =
690 { /* initialization for https://bugs.llvm.org/show_bug.cgi?id=42570 */ };
691 EXPORT_SYMBOL(kmalloc_caches);
694 * Conversion table for small slabs sizes / 8 to the index in the
695 * kmalloc array. This is necessary for slabs < 192 since we have non power
696 * of two cache sizes there. The size of larger slabs can be determined using
699 static u8 size_index[24] __ro_after_init = {
726 static inline unsigned int size_index_elem(unsigned int bytes)
728 return (bytes - 1) / 8;
732 * Find the kmem_cache structure that serves a given size of
735 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
741 return ZERO_SIZE_PTR;
743 index = size_index[size_index_elem(size)];
745 if (WARN_ON_ONCE(size > KMALLOC_MAX_CACHE_SIZE))
747 index = fls(size - 1);
750 return kmalloc_caches[kmalloc_type(flags)][index];
753 #ifdef CONFIG_ZONE_DMA
754 #define KMALLOC_DMA_NAME(sz) .name[KMALLOC_DMA] = "dma-kmalloc-" #sz,
756 #define KMALLOC_DMA_NAME(sz)
759 #ifdef CONFIG_MEMCG_KMEM
760 #define KMALLOC_CGROUP_NAME(sz) .name[KMALLOC_CGROUP] = "kmalloc-cg-" #sz,
762 #define KMALLOC_CGROUP_NAME(sz)
765 #define INIT_KMALLOC_INFO(__size, __short_size) \
767 .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \
768 .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #__short_size, \
769 KMALLOC_CGROUP_NAME(__short_size) \
770 KMALLOC_DMA_NAME(__short_size) \
775 * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
776 * kmalloc_index() supports up to 2^25=32MB, so the final entry of the table is
779 const struct kmalloc_info_struct kmalloc_info[] __initconst = {
780 INIT_KMALLOC_INFO(0, 0),
781 INIT_KMALLOC_INFO(96, 96),
782 INIT_KMALLOC_INFO(192, 192),
783 INIT_KMALLOC_INFO(8, 8),
784 INIT_KMALLOC_INFO(16, 16),
785 INIT_KMALLOC_INFO(32, 32),
786 INIT_KMALLOC_INFO(64, 64),
787 INIT_KMALLOC_INFO(128, 128),
788 INIT_KMALLOC_INFO(256, 256),
789 INIT_KMALLOC_INFO(512, 512),
790 INIT_KMALLOC_INFO(1024, 1k),
791 INIT_KMALLOC_INFO(2048, 2k),
792 INIT_KMALLOC_INFO(4096, 4k),
793 INIT_KMALLOC_INFO(8192, 8k),
794 INIT_KMALLOC_INFO(16384, 16k),
795 INIT_KMALLOC_INFO(32768, 32k),
796 INIT_KMALLOC_INFO(65536, 64k),
797 INIT_KMALLOC_INFO(131072, 128k),
798 INIT_KMALLOC_INFO(262144, 256k),
799 INIT_KMALLOC_INFO(524288, 512k),
800 INIT_KMALLOC_INFO(1048576, 1M),
801 INIT_KMALLOC_INFO(2097152, 2M),
802 INIT_KMALLOC_INFO(4194304, 4M),
803 INIT_KMALLOC_INFO(8388608, 8M),
804 INIT_KMALLOC_INFO(16777216, 16M),
805 INIT_KMALLOC_INFO(33554432, 32M)
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 */
918 #endif /* !CONFIG_SLOB */
920 gfp_t kmalloc_fix_flags(gfp_t flags)
922 gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
924 flags &= ~GFP_SLAB_BUG_MASK;
925 pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
926 invalid_mask, &invalid_mask, flags, &flags);
933 * To avoid unnecessary overhead, we pass through large allocation requests
934 * directly to the page allocator. We use __GFP_COMP, because we will need to
935 * know the allocation order to free the pages properly in kfree.
937 void *kmalloc_order(size_t size, gfp_t flags, unsigned int order)
942 if (unlikely(flags & GFP_SLAB_BUG_MASK))
943 flags = kmalloc_fix_flags(flags);
946 page = alloc_pages(flags, order);
948 ret = page_address(page);
949 mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B,
952 ret = kasan_kmalloc_large(ret, size, flags);
953 /* As ret might get tagged, call kmemleak hook after KASAN. */
954 kmemleak_alloc(ret, size, 1, flags);
957 EXPORT_SYMBOL(kmalloc_order);
959 #ifdef CONFIG_TRACING
960 void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
962 void *ret = kmalloc_order(size, flags, order);
963 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
966 EXPORT_SYMBOL(kmalloc_order_trace);
969 #ifdef CONFIG_SLAB_FREELIST_RANDOM
970 /* Randomize a generic freelist */
971 static void freelist_randomize(struct rnd_state *state, unsigned int *list,
977 for (i = 0; i < count; i++)
980 /* Fisher-Yates shuffle */
981 for (i = count - 1; i > 0; i--) {
982 rand = prandom_u32_state(state);
984 swap(list[i], list[rand]);
988 /* Create a random sequence per cache */
989 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
992 struct rnd_state state;
994 if (count < 2 || cachep->random_seq)
997 cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp);
998 if (!cachep->random_seq)
1001 /* Get best entropy at this stage of boot */
1002 prandom_seed_state(&state, get_random_long());
1004 freelist_randomize(&state, cachep->random_seq, count);
1008 /* Destroy the per-cache random freelist sequence */
1009 void cache_random_seq_destroy(struct kmem_cache *cachep)
1011 kfree(cachep->random_seq);
1012 cachep->random_seq = NULL;
1014 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
1016 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
1018 #define SLABINFO_RIGHTS (0600)
1020 #define SLABINFO_RIGHTS (0400)
1023 static void print_slabinfo_header(struct seq_file *m)
1026 * Output format version, so at least we can change it
1027 * without _too_ many complaints.
1029 #ifdef CONFIG_DEBUG_SLAB
1030 seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
1032 seq_puts(m, "slabinfo - version: 2.1\n");
1034 seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
1035 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
1036 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
1037 #ifdef CONFIG_DEBUG_SLAB
1038 seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
1039 seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
1044 static void *slab_start(struct seq_file *m, loff_t *pos)
1046 mutex_lock(&slab_mutex);
1047 return seq_list_start(&slab_caches, *pos);
1050 static void *slab_next(struct seq_file *m, void *p, loff_t *pos)
1052 return seq_list_next(p, &slab_caches, pos);
1055 static void slab_stop(struct seq_file *m, void *p)
1057 mutex_unlock(&slab_mutex);
1060 static void cache_show(struct kmem_cache *s, struct seq_file *m)
1062 struct slabinfo sinfo;
1064 memset(&sinfo, 0, sizeof(sinfo));
1065 get_slabinfo(s, &sinfo);
1067 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
1068 s->name, sinfo.active_objs, sinfo.num_objs, s->size,
1069 sinfo.objects_per_slab, (1 << sinfo.cache_order));
1071 seq_printf(m, " : tunables %4u %4u %4u",
1072 sinfo.limit, sinfo.batchcount, sinfo.shared);
1073 seq_printf(m, " : slabdata %6lu %6lu %6lu",
1074 sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
1075 slabinfo_show_stats(m, s);
1079 static int slab_show(struct seq_file *m, void *p)
1081 struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
1083 if (p == slab_caches.next)
1084 print_slabinfo_header(m);
1089 void dump_unreclaimable_slab(void)
1091 struct kmem_cache *s;
1092 struct slabinfo sinfo;
1095 * Here acquiring slab_mutex is risky since we don't prefer to get
1096 * sleep in oom path. But, without mutex hold, it may introduce a
1098 * Use mutex_trylock to protect the list traverse, dump nothing
1099 * without acquiring the mutex.
1101 if (!mutex_trylock(&slab_mutex)) {
1102 pr_warn("excessive unreclaimable slab but cannot dump stats\n");
1106 pr_info("Unreclaimable slab info:\n");
1107 pr_info("Name Used Total\n");
1109 list_for_each_entry(s, &slab_caches, list) {
1110 if (s->flags & SLAB_RECLAIM_ACCOUNT)
1113 get_slabinfo(s, &sinfo);
1115 if (sinfo.num_objs > 0)
1116 pr_info("%-17s %10luKB %10luKB\n", s->name,
1117 (sinfo.active_objs * s->size) / 1024,
1118 (sinfo.num_objs * s->size) / 1024);
1120 mutex_unlock(&slab_mutex);
1124 * slabinfo_op - iterator that generates /proc/slabinfo
1133 * num-pages-per-slab
1134 * + further values on SMP and with statistics enabled
1136 static const struct seq_operations slabinfo_op = {
1137 .start = slab_start,
1143 static int slabinfo_open(struct inode *inode, struct file *file)
1145 return seq_open(file, &slabinfo_op);
1148 static const struct proc_ops slabinfo_proc_ops = {
1149 .proc_flags = PROC_ENTRY_PERMANENT,
1150 .proc_open = slabinfo_open,
1151 .proc_read = seq_read,
1152 .proc_write = slabinfo_write,
1153 .proc_lseek = seq_lseek,
1154 .proc_release = seq_release,
1157 static int __init slab_proc_init(void)
1159 proc_create("slabinfo", SLABINFO_RIGHTS, NULL, &slabinfo_proc_ops);
1162 module_init(slab_proc_init);
1164 #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */
1166 static __always_inline void *__do_krealloc(const void *p, size_t new_size,
1172 /* Don't use instrumented ksize to allow precise KASAN poisoning. */
1173 if (likely(!ZERO_OR_NULL_PTR(p))) {
1174 if (!kasan_check_byte(p))
1176 ks = kfence_ksize(p) ?: __ksize(p);
1180 /* If the object still fits, repoison it precisely. */
1181 if (ks >= new_size) {
1182 p = kasan_krealloc((void *)p, new_size, flags);
1186 ret = kmalloc_track_caller(new_size, flags);
1188 /* Disable KASAN checks as the object's redzone is accessed. */
1189 kasan_disable_current();
1190 memcpy(ret, kasan_reset_tag(p), ks);
1191 kasan_enable_current();
1198 * krealloc - reallocate memory. The contents will remain unchanged.
1199 * @p: object to reallocate memory for.
1200 * @new_size: how many bytes of memory are required.
1201 * @flags: the type of memory to allocate.
1203 * The contents of the object pointed to are preserved up to the
1204 * lesser of the new and old sizes (__GFP_ZERO flag is effectively ignored).
1205 * If @p is %NULL, krealloc() behaves exactly like kmalloc(). If @new_size
1206 * is 0 and @p is not a %NULL pointer, the object pointed to is freed.
1208 * Return: pointer to the allocated memory or %NULL in case of error
1210 void *krealloc(const void *p, size_t new_size, gfp_t flags)
1214 if (unlikely(!new_size)) {
1216 return ZERO_SIZE_PTR;
1219 ret = __do_krealloc(p, new_size, flags);
1220 if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret))
1225 EXPORT_SYMBOL(krealloc);
1228 * kfree_sensitive - Clear sensitive information in memory before freeing
1229 * @p: object to free memory of
1231 * The memory of the object @p points to is zeroed before freed.
1232 * If @p is %NULL, kfree_sensitive() does nothing.
1234 * Note: this function zeroes the whole allocated buffer which can be a good
1235 * deal bigger than the requested buffer size passed to kmalloc(). So be
1236 * careful when using this function in performance sensitive code.
1238 void kfree_sensitive(const void *p)
1241 void *mem = (void *)p;
1245 memzero_explicit(mem, ks);
1248 EXPORT_SYMBOL(kfree_sensitive);
1251 * ksize - get the actual amount of memory allocated for a given object
1252 * @objp: Pointer to the object
1254 * kmalloc may internally round up allocations and return more memory
1255 * than requested. ksize() can be used to determine the actual amount of
1256 * memory allocated. The caller may use this additional memory, even though
1257 * a smaller amount of memory was initially specified with the kmalloc call.
1258 * The caller must guarantee that objp points to a valid object previously
1259 * allocated with either kmalloc() or kmem_cache_alloc(). The object
1260 * must not be freed during the duration of the call.
1262 * Return: size of the actual memory used by @objp in bytes
1264 size_t ksize(const void *objp)
1269 * We need to first check that the pointer to the object is valid, and
1270 * only then unpoison the memory. The report printed from ksize() is
1271 * more useful, then when it's printed later when the behaviour could
1272 * be undefined due to a potential use-after-free or double-free.
1274 * We use kasan_check_byte(), which is supported for the hardware
1275 * tag-based KASAN mode, unlike kasan_check_read/write().
1277 * If the pointed to memory is invalid, we return 0 to avoid users of
1278 * ksize() writing to and potentially corrupting the memory region.
1280 * We want to perform the check before __ksize(), to avoid potentially
1281 * crashing in __ksize() due to accessing invalid metadata.
1283 if (unlikely(ZERO_OR_NULL_PTR(objp)) || !kasan_check_byte(objp))
1286 size = kfence_ksize(objp) ?: __ksize(objp);
1288 * We assume that ksize callers could use whole allocated area,
1289 * so we need to unpoison this area.
1291 kasan_unpoison_range(objp, size);
1294 EXPORT_SYMBOL(ksize);
1296 /* Tracepoints definitions. */
1297 EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1298 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1299 EXPORT_TRACEPOINT_SYMBOL(kmalloc_node);
1300 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node);
1301 EXPORT_TRACEPOINT_SYMBOL(kfree);
1302 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);
1304 int should_failslab(struct kmem_cache *s, gfp_t gfpflags)
1306 if (__should_failslab(s, gfpflags))
1310 ALLOW_ERROR_INJECTION(should_failslab, ERRNO);