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/dma-mapping.h>
21 #include <linux/swiotlb.h>
22 #include <linux/proc_fs.h>
23 #include <linux/debugfs.h>
24 #include <linux/kasan.h>
25 #include <asm/cacheflush.h>
26 #include <asm/tlbflush.h>
28 #include <linux/memcontrol.h>
29 #include <linux/stackdepot.h>
34 #define CREATE_TRACE_POINTS
35 #include <trace/events/kmem.h>
37 enum slab_state slab_state;
38 LIST_HEAD(slab_caches);
39 DEFINE_MUTEX(slab_mutex);
40 struct kmem_cache *kmem_cache;
42 static LIST_HEAD(slab_caches_to_rcu_destroy);
43 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work);
44 static DECLARE_WORK(slab_caches_to_rcu_destroy_work,
45 slab_caches_to_rcu_destroy_workfn);
48 * Set of flags that will prevent slab merging
50 #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
51 SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \
52 SLAB_FAILSLAB | SLAB_NO_MERGE | kasan_never_merge())
54 #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \
55 SLAB_CACHE_DMA32 | SLAB_ACCOUNT)
58 * Merge control. If this is set then no merging of slab caches will occur.
60 static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT);
62 static int __init setup_slab_nomerge(char *str)
68 static int __init setup_slab_merge(char *str)
75 __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0);
76 __setup_param("slub_merge", slub_merge, setup_slab_merge, 0);
79 __setup("slab_nomerge", setup_slab_nomerge);
80 __setup("slab_merge", setup_slab_merge);
83 * Determine the size of a slab object
85 unsigned int kmem_cache_size(struct kmem_cache *s)
87 return s->object_size;
89 EXPORT_SYMBOL(kmem_cache_size);
91 #ifdef CONFIG_DEBUG_VM
92 static int kmem_cache_sanity_check(const char *name, unsigned int size)
94 if (!name || in_interrupt() || size > KMALLOC_MAX_SIZE) {
95 pr_err("kmem_cache_create(%s) integrity check failed\n", name);
99 WARN_ON(strchr(name, ' ')); /* It confuses parsers */
103 static inline int kmem_cache_sanity_check(const char *name, unsigned int size)
110 * Figure out what the alignment of the objects will be given a set of
111 * flags, a user specified alignment and the size of the objects.
113 static unsigned int calculate_alignment(slab_flags_t flags,
114 unsigned int align, unsigned int size)
117 * If the user wants hardware cache aligned objects then follow that
118 * suggestion if the object is sufficiently large.
120 * The hardware cache alignment cannot override the specified
121 * alignment though. If that is greater then use it.
123 if (flags & SLAB_HWCACHE_ALIGN) {
126 ralign = cache_line_size();
127 while (size <= ralign / 2)
129 align = max(align, ralign);
132 align = max(align, arch_slab_minalign());
134 return ALIGN(align, sizeof(void *));
138 * Find a mergeable slab cache
140 int slab_unmergeable(struct kmem_cache *s)
142 if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
148 #ifdef CONFIG_HARDENED_USERCOPY
154 * We may have set a slab to be unmergeable during bootstrap.
162 struct kmem_cache *find_mergeable(unsigned int size, unsigned int align,
163 slab_flags_t flags, const char *name, void (*ctor)(void *))
165 struct kmem_cache *s;
173 size = ALIGN(size, sizeof(void *));
174 align = calculate_alignment(flags, align, size);
175 size = ALIGN(size, align);
176 flags = kmem_cache_flags(size, flags, name);
178 if (flags & SLAB_NEVER_MERGE)
181 list_for_each_entry_reverse(s, &slab_caches, list) {
182 if (slab_unmergeable(s))
188 if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
191 * Check if alignment is compatible.
192 * Courtesy of Adrian Drzewiecki
194 if ((s->size & ~(align - 1)) != s->size)
197 if (s->size - size >= sizeof(void *))
200 if (IS_ENABLED(CONFIG_SLAB) && align &&
201 (align > s->align || s->align % align))
209 static struct kmem_cache *create_cache(const char *name,
210 unsigned int object_size, unsigned int align,
211 slab_flags_t flags, unsigned int useroffset,
212 unsigned int usersize, void (*ctor)(void *),
213 struct kmem_cache *root_cache)
215 struct kmem_cache *s;
218 if (WARN_ON(useroffset + usersize > object_size))
219 useroffset = usersize = 0;
222 s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
227 s->size = s->object_size = object_size;
230 #ifdef CONFIG_HARDENED_USERCOPY
231 s->useroffset = useroffset;
232 s->usersize = usersize;
235 err = __kmem_cache_create(s, flags);
240 list_add(&s->list, &slab_caches);
244 kmem_cache_free(kmem_cache, s);
250 * kmem_cache_create_usercopy - Create a cache with a region suitable
251 * for copying to userspace
252 * @name: A string which is used in /proc/slabinfo to identify this cache.
253 * @size: The size of objects to be created in this cache.
254 * @align: The required alignment for the objects.
256 * @useroffset: Usercopy region offset
257 * @usersize: Usercopy region size
258 * @ctor: A constructor for the objects.
260 * Cannot be called within a interrupt, but can be interrupted.
261 * The @ctor is run when new pages are allocated by the cache.
265 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
266 * to catch references to uninitialised memory.
268 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
269 * for buffer overruns.
271 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
272 * cacheline. This can be beneficial if you're counting cycles as closely
275 * Return: a pointer to the cache on success, NULL on failure.
278 kmem_cache_create_usercopy(const char *name,
279 unsigned int size, unsigned int align,
281 unsigned int useroffset, unsigned int usersize,
282 void (*ctor)(void *))
284 struct kmem_cache *s = NULL;
285 const char *cache_name;
288 #ifdef CONFIG_SLUB_DEBUG
290 * If no slub_debug was enabled globally, the static key is not yet
291 * enabled by setup_slub_debug(). Enable it if the cache is being
292 * created with any of the debugging flags passed explicitly.
293 * It's also possible that this is the first cache created with
294 * SLAB_STORE_USER and we should init stack_depot for it.
296 if (flags & SLAB_DEBUG_FLAGS)
297 static_branch_enable(&slub_debug_enabled);
298 if (flags & SLAB_STORE_USER)
302 mutex_lock(&slab_mutex);
304 err = kmem_cache_sanity_check(name, size);
309 /* Refuse requests with allocator specific flags */
310 if (flags & ~SLAB_FLAGS_PERMITTED) {
316 * Some allocators will constraint the set of valid flags to a subset
317 * of all flags. We expect them to define CACHE_CREATE_MASK in this
318 * case, and we'll just provide them with a sanitized version of the
321 flags &= CACHE_CREATE_MASK;
323 /* Fail closed on bad usersize of useroffset values. */
324 if (!IS_ENABLED(CONFIG_HARDENED_USERCOPY) ||
325 WARN_ON(!usersize && useroffset) ||
326 WARN_ON(size < usersize || size - usersize < useroffset))
327 usersize = useroffset = 0;
330 s = __kmem_cache_alias(name, size, align, flags, ctor);
334 cache_name = kstrdup_const(name, GFP_KERNEL);
340 s = create_cache(cache_name, size,
341 calculate_alignment(flags, align, size),
342 flags, useroffset, usersize, ctor, NULL);
345 kfree_const(cache_name);
349 mutex_unlock(&slab_mutex);
352 if (flags & SLAB_PANIC)
353 panic("%s: Failed to create slab '%s'. Error %d\n",
354 __func__, name, err);
356 pr_warn("%s(%s) failed with error %d\n",
357 __func__, name, err);
364 EXPORT_SYMBOL(kmem_cache_create_usercopy);
367 * kmem_cache_create - Create a cache.
368 * @name: A string which is used in /proc/slabinfo to identify this cache.
369 * @size: The size of objects to be created in this cache.
370 * @align: The required alignment for the objects.
372 * @ctor: A constructor for the objects.
374 * Cannot be called within a interrupt, but can be interrupted.
375 * The @ctor is run when new pages are allocated by the cache.
379 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
380 * to catch references to uninitialised memory.
382 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
383 * for buffer overruns.
385 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
386 * cacheline. This can be beneficial if you're counting cycles as closely
389 * Return: a pointer to the cache on success, NULL on failure.
392 kmem_cache_create(const char *name, unsigned int size, unsigned int align,
393 slab_flags_t flags, void (*ctor)(void *))
395 return kmem_cache_create_usercopy(name, size, align, flags, 0, 0,
398 EXPORT_SYMBOL(kmem_cache_create);
400 #ifdef SLAB_SUPPORTS_SYSFS
402 * For a given kmem_cache, kmem_cache_destroy() should only be called
403 * once or there will be a use-after-free problem. The actual deletion
404 * and release of the kobject does not need slab_mutex or cpu_hotplug_lock
405 * protection. So they are now done without holding those locks.
407 * Note that there will be a slight delay in the deletion of sysfs files
408 * if kmem_cache_release() is called indrectly from a work function.
410 static void kmem_cache_release(struct kmem_cache *s)
412 sysfs_slab_unlink(s);
413 sysfs_slab_release(s);
416 static void kmem_cache_release(struct kmem_cache *s)
418 slab_kmem_cache_release(s);
422 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work)
424 LIST_HEAD(to_destroy);
425 struct kmem_cache *s, *s2;
428 * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the
429 * @slab_caches_to_rcu_destroy list. The slab pages are freed
430 * through RCU and the associated kmem_cache are dereferenced
431 * while freeing the pages, so the kmem_caches should be freed only
432 * after the pending RCU operations are finished. As rcu_barrier()
433 * is a pretty slow operation, we batch all pending destructions
436 mutex_lock(&slab_mutex);
437 list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy);
438 mutex_unlock(&slab_mutex);
440 if (list_empty(&to_destroy))
445 list_for_each_entry_safe(s, s2, &to_destroy, list) {
446 debugfs_slab_release(s);
447 kfence_shutdown_cache(s);
448 kmem_cache_release(s);
452 static int shutdown_cache(struct kmem_cache *s)
454 /* free asan quarantined objects */
455 kasan_cache_shutdown(s);
457 if (__kmem_cache_shutdown(s) != 0)
462 if (s->flags & SLAB_TYPESAFE_BY_RCU) {
463 list_add_tail(&s->list, &slab_caches_to_rcu_destroy);
464 schedule_work(&slab_caches_to_rcu_destroy_work);
466 kfence_shutdown_cache(s);
467 debugfs_slab_release(s);
473 void slab_kmem_cache_release(struct kmem_cache *s)
475 __kmem_cache_release(s);
476 kfree_const(s->name);
477 kmem_cache_free(kmem_cache, s);
480 void kmem_cache_destroy(struct kmem_cache *s)
485 if (unlikely(!s) || !kasan_check_byte(s))
489 mutex_lock(&slab_mutex);
491 rcu_set = s->flags & SLAB_TYPESAFE_BY_RCU;
493 refcnt = --s->refcount;
497 WARN(shutdown_cache(s),
498 "%s %s: Slab cache still has objects when called from %pS",
499 __func__, s->name, (void *)_RET_IP_);
501 mutex_unlock(&slab_mutex);
503 if (!refcnt && !rcu_set)
504 kmem_cache_release(s);
506 EXPORT_SYMBOL(kmem_cache_destroy);
509 * kmem_cache_shrink - Shrink a cache.
510 * @cachep: The cache to shrink.
512 * Releases as many slabs as possible for a cache.
513 * To help debugging, a zero exit status indicates all slabs were released.
515 * Return: %0 if all slabs were released, non-zero otherwise
517 int kmem_cache_shrink(struct kmem_cache *cachep)
519 kasan_cache_shrink(cachep);
521 return __kmem_cache_shrink(cachep);
523 EXPORT_SYMBOL(kmem_cache_shrink);
525 bool slab_is_available(void)
527 return slab_state >= UP;
532 * kmem_valid_obj - does the pointer reference a valid slab object?
533 * @object: pointer to query.
535 * Return: %true if the pointer is to a not-yet-freed object from
536 * kmalloc() or kmem_cache_alloc(), either %true or %false if the pointer
537 * is to an already-freed object, and %false otherwise.
539 bool kmem_valid_obj(void *object)
543 /* Some arches consider ZERO_SIZE_PTR to be a valid address. */
544 if (object < (void *)PAGE_SIZE || !virt_addr_valid(object))
546 folio = virt_to_folio(object);
547 return folio_test_slab(folio);
549 EXPORT_SYMBOL_GPL(kmem_valid_obj);
551 static void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab)
553 if (__kfence_obj_info(kpp, object, slab))
555 __kmem_obj_info(kpp, object, slab);
559 * kmem_dump_obj - Print available slab provenance information
560 * @object: slab object for which to find provenance information.
562 * This function uses pr_cont(), so that the caller is expected to have
563 * printed out whatever preamble is appropriate. The provenance information
564 * depends on the type of object and on how much debugging is enabled.
565 * For a slab-cache object, the fact that it is a slab object is printed,
566 * and, if available, the slab name, return address, and stack trace from
567 * the allocation and last free path of that object.
569 * This function will splat if passed a pointer to a non-slab object.
570 * If you are not sure what type of object you have, you should instead
571 * use mem_dump_obj().
573 void kmem_dump_obj(void *object)
575 char *cp = IS_ENABLED(CONFIG_MMU) ? "" : "/vmalloc";
578 unsigned long ptroffset;
579 struct kmem_obj_info kp = { };
581 if (WARN_ON_ONCE(!virt_addr_valid(object)))
583 slab = virt_to_slab(object);
584 if (WARN_ON_ONCE(!slab)) {
585 pr_cont(" non-slab memory.\n");
588 kmem_obj_info(&kp, object, slab);
589 if (kp.kp_slab_cache)
590 pr_cont(" slab%s %s", cp, kp.kp_slab_cache->name);
592 pr_cont(" slab%s", cp);
593 if (is_kfence_address(object))
594 pr_cont(" (kfence)");
596 pr_cont(" start %px", kp.kp_objp);
597 if (kp.kp_data_offset)
598 pr_cont(" data offset %lu", kp.kp_data_offset);
600 ptroffset = ((char *)object - (char *)kp.kp_objp) - kp.kp_data_offset;
601 pr_cont(" pointer offset %lu", ptroffset);
603 if (kp.kp_slab_cache && kp.kp_slab_cache->object_size)
604 pr_cont(" size %u", kp.kp_slab_cache->object_size);
606 pr_cont(" allocated at %pS\n", kp.kp_ret);
609 for (i = 0; i < ARRAY_SIZE(kp.kp_stack); i++) {
612 pr_info(" %pS\n", kp.kp_stack[i]);
615 if (kp.kp_free_stack[0])
616 pr_cont(" Free path:\n");
618 for (i = 0; i < ARRAY_SIZE(kp.kp_free_stack); i++) {
619 if (!kp.kp_free_stack[i])
621 pr_info(" %pS\n", kp.kp_free_stack[i]);
625 EXPORT_SYMBOL_GPL(kmem_dump_obj);
628 /* Create a cache during boot when no slab services are available yet */
629 void __init create_boot_cache(struct kmem_cache *s, const char *name,
630 unsigned int size, slab_flags_t flags,
631 unsigned int useroffset, unsigned int usersize)
634 unsigned int align = ARCH_KMALLOC_MINALIGN;
637 s->size = s->object_size = size;
640 * For power of two sizes, guarantee natural alignment for kmalloc
641 * caches, regardless of SL*B debugging options.
643 if (is_power_of_2(size))
644 align = max(align, size);
645 s->align = calculate_alignment(flags, align, size);
647 #ifdef CONFIG_HARDENED_USERCOPY
648 s->useroffset = useroffset;
649 s->usersize = usersize;
652 err = __kmem_cache_create(s, flags);
655 panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n",
658 s->refcount = -1; /* Exempt from merging for now */
661 static struct kmem_cache *__init create_kmalloc_cache(const char *name,
665 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
668 panic("Out of memory when creating slab %s\n", name);
670 create_boot_cache(s, name, size, flags | SLAB_KMALLOC, 0, size);
671 list_add(&s->list, &slab_caches);
677 kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1] __ro_after_init =
678 { /* initialization for https://bugs.llvm.org/show_bug.cgi?id=42570 */ };
679 EXPORT_SYMBOL(kmalloc_caches);
681 #ifdef CONFIG_RANDOM_KMALLOC_CACHES
682 unsigned long random_kmalloc_seed __ro_after_init;
683 EXPORT_SYMBOL(random_kmalloc_seed);
687 * Conversion table for small slabs sizes / 8 to the index in the
688 * kmalloc array. This is necessary for slabs < 192 since we have non power
689 * of two cache sizes there. The size of larger slabs can be determined using
692 static u8 size_index[24] __ro_after_init = {
719 static inline unsigned int size_index_elem(unsigned int bytes)
721 return (bytes - 1) / 8;
725 * Find the kmem_cache structure that serves a given size of
728 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags, unsigned long caller)
734 return ZERO_SIZE_PTR;
736 index = size_index[size_index_elem(size)];
738 if (WARN_ON_ONCE(size > KMALLOC_MAX_CACHE_SIZE))
740 index = fls(size - 1);
743 return kmalloc_caches[kmalloc_type(flags, caller)][index];
746 size_t kmalloc_size_roundup(size_t size)
748 struct kmem_cache *c;
750 /* Short-circuit the 0 size case. */
751 if (unlikely(size == 0))
753 /* Short-circuit saturated "too-large" case. */
754 if (unlikely(size == SIZE_MAX))
756 /* Above the smaller buckets, size is a multiple of page size. */
757 if (size > KMALLOC_MAX_CACHE_SIZE)
758 return PAGE_SIZE << get_order(size);
761 * The flags don't matter since size_index is common to all.
762 * Neither does the caller for just getting ->object_size.
764 c = kmalloc_slab(size, GFP_KERNEL, 0);
765 return c ? c->object_size : 0;
767 EXPORT_SYMBOL(kmalloc_size_roundup);
769 #ifdef CONFIG_ZONE_DMA
770 #define KMALLOC_DMA_NAME(sz) .name[KMALLOC_DMA] = "dma-kmalloc-" #sz,
772 #define KMALLOC_DMA_NAME(sz)
775 #ifdef CONFIG_MEMCG_KMEM
776 #define KMALLOC_CGROUP_NAME(sz) .name[KMALLOC_CGROUP] = "kmalloc-cg-" #sz,
778 #define KMALLOC_CGROUP_NAME(sz)
781 #ifndef CONFIG_SLUB_TINY
782 #define KMALLOC_RCL_NAME(sz) .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #sz,
784 #define KMALLOC_RCL_NAME(sz)
787 #ifdef CONFIG_RANDOM_KMALLOC_CACHES
788 #define __KMALLOC_RANDOM_CONCAT(a, b) a ## b
789 #define KMALLOC_RANDOM_NAME(N, sz) __KMALLOC_RANDOM_CONCAT(KMA_RAND_, N)(sz)
790 #define KMA_RAND_1(sz) .name[KMALLOC_RANDOM_START + 1] = "kmalloc-rnd-01-" #sz,
791 #define KMA_RAND_2(sz) KMA_RAND_1(sz) .name[KMALLOC_RANDOM_START + 2] = "kmalloc-rnd-02-" #sz,
792 #define KMA_RAND_3(sz) KMA_RAND_2(sz) .name[KMALLOC_RANDOM_START + 3] = "kmalloc-rnd-03-" #sz,
793 #define KMA_RAND_4(sz) KMA_RAND_3(sz) .name[KMALLOC_RANDOM_START + 4] = "kmalloc-rnd-04-" #sz,
794 #define KMA_RAND_5(sz) KMA_RAND_4(sz) .name[KMALLOC_RANDOM_START + 5] = "kmalloc-rnd-05-" #sz,
795 #define KMA_RAND_6(sz) KMA_RAND_5(sz) .name[KMALLOC_RANDOM_START + 6] = "kmalloc-rnd-06-" #sz,
796 #define KMA_RAND_7(sz) KMA_RAND_6(sz) .name[KMALLOC_RANDOM_START + 7] = "kmalloc-rnd-07-" #sz,
797 #define KMA_RAND_8(sz) KMA_RAND_7(sz) .name[KMALLOC_RANDOM_START + 8] = "kmalloc-rnd-08-" #sz,
798 #define KMA_RAND_9(sz) KMA_RAND_8(sz) .name[KMALLOC_RANDOM_START + 9] = "kmalloc-rnd-09-" #sz,
799 #define KMA_RAND_10(sz) KMA_RAND_9(sz) .name[KMALLOC_RANDOM_START + 10] = "kmalloc-rnd-10-" #sz,
800 #define KMA_RAND_11(sz) KMA_RAND_10(sz) .name[KMALLOC_RANDOM_START + 11] = "kmalloc-rnd-11-" #sz,
801 #define KMA_RAND_12(sz) KMA_RAND_11(sz) .name[KMALLOC_RANDOM_START + 12] = "kmalloc-rnd-12-" #sz,
802 #define KMA_RAND_13(sz) KMA_RAND_12(sz) .name[KMALLOC_RANDOM_START + 13] = "kmalloc-rnd-13-" #sz,
803 #define KMA_RAND_14(sz) KMA_RAND_13(sz) .name[KMALLOC_RANDOM_START + 14] = "kmalloc-rnd-14-" #sz,
804 #define KMA_RAND_15(sz) KMA_RAND_14(sz) .name[KMALLOC_RANDOM_START + 15] = "kmalloc-rnd-15-" #sz,
805 #else // CONFIG_RANDOM_KMALLOC_CACHES
806 #define KMALLOC_RANDOM_NAME(N, sz)
809 #define INIT_KMALLOC_INFO(__size, __short_size) \
811 .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \
812 KMALLOC_RCL_NAME(__short_size) \
813 KMALLOC_CGROUP_NAME(__short_size) \
814 KMALLOC_DMA_NAME(__short_size) \
815 KMALLOC_RANDOM_NAME(RANDOM_KMALLOC_CACHES_NR, __short_size) \
820 * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
821 * kmalloc_index() supports up to 2^21=2MB, so the final entry of the table is
824 const struct kmalloc_info_struct kmalloc_info[] __initconst = {
825 INIT_KMALLOC_INFO(0, 0),
826 INIT_KMALLOC_INFO(96, 96),
827 INIT_KMALLOC_INFO(192, 192),
828 INIT_KMALLOC_INFO(8, 8),
829 INIT_KMALLOC_INFO(16, 16),
830 INIT_KMALLOC_INFO(32, 32),
831 INIT_KMALLOC_INFO(64, 64),
832 INIT_KMALLOC_INFO(128, 128),
833 INIT_KMALLOC_INFO(256, 256),
834 INIT_KMALLOC_INFO(512, 512),
835 INIT_KMALLOC_INFO(1024, 1k),
836 INIT_KMALLOC_INFO(2048, 2k),
837 INIT_KMALLOC_INFO(4096, 4k),
838 INIT_KMALLOC_INFO(8192, 8k),
839 INIT_KMALLOC_INFO(16384, 16k),
840 INIT_KMALLOC_INFO(32768, 32k),
841 INIT_KMALLOC_INFO(65536, 64k),
842 INIT_KMALLOC_INFO(131072, 128k),
843 INIT_KMALLOC_INFO(262144, 256k),
844 INIT_KMALLOC_INFO(524288, 512k),
845 INIT_KMALLOC_INFO(1048576, 1M),
846 INIT_KMALLOC_INFO(2097152, 2M)
850 * Patch up the size_index table if we have strange large alignment
851 * requirements for the kmalloc array. This is only the case for
852 * MIPS it seems. The standard arches will not generate any code here.
854 * Largest permitted alignment is 256 bytes due to the way we
855 * handle the index determination for the smaller caches.
857 * Make sure that nothing crazy happens if someone starts tinkering
858 * around with ARCH_KMALLOC_MINALIGN
860 void __init setup_kmalloc_cache_index_table(void)
864 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
865 !is_power_of_2(KMALLOC_MIN_SIZE));
867 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
868 unsigned int elem = size_index_elem(i);
870 if (elem >= ARRAY_SIZE(size_index))
872 size_index[elem] = KMALLOC_SHIFT_LOW;
875 if (KMALLOC_MIN_SIZE >= 64) {
877 * The 96 byte sized cache is not used if the alignment
880 for (i = 64 + 8; i <= 96; i += 8)
881 size_index[size_index_elem(i)] = 7;
885 if (KMALLOC_MIN_SIZE >= 128) {
887 * The 192 byte sized cache is not used if the alignment
888 * is 128 byte. Redirect kmalloc to use the 256 byte cache
891 for (i = 128 + 8; i <= 192; i += 8)
892 size_index[size_index_elem(i)] = 8;
896 static unsigned int __kmalloc_minalign(void)
898 if (IS_ENABLED(CONFIG_DMA_BOUNCE_UNALIGNED_KMALLOC) &&
899 is_swiotlb_allocated())
900 return ARCH_KMALLOC_MINALIGN;
901 return dma_get_cache_alignment();
905 new_kmalloc_cache(int idx, enum kmalloc_cache_type type, slab_flags_t flags)
907 unsigned int minalign = __kmalloc_minalign();
908 unsigned int aligned_size = kmalloc_info[idx].size;
909 int aligned_idx = idx;
911 if ((KMALLOC_RECLAIM != KMALLOC_NORMAL) && (type == KMALLOC_RECLAIM)) {
912 flags |= SLAB_RECLAIM_ACCOUNT;
913 } else if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_CGROUP)) {
914 if (mem_cgroup_kmem_disabled()) {
915 kmalloc_caches[type][idx] = kmalloc_caches[KMALLOC_NORMAL][idx];
918 flags |= SLAB_ACCOUNT;
919 } else if (IS_ENABLED(CONFIG_ZONE_DMA) && (type == KMALLOC_DMA)) {
920 flags |= SLAB_CACHE_DMA;
923 #ifdef CONFIG_RANDOM_KMALLOC_CACHES
924 if (type >= KMALLOC_RANDOM_START && type <= KMALLOC_RANDOM_END)
925 flags |= SLAB_NO_MERGE;
929 * If CONFIG_MEMCG_KMEM is enabled, disable cache merging for
930 * KMALLOC_NORMAL caches.
932 if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_NORMAL))
933 flags |= SLAB_NO_MERGE;
935 if (minalign > ARCH_KMALLOC_MINALIGN) {
936 aligned_size = ALIGN(aligned_size, minalign);
937 aligned_idx = __kmalloc_index(aligned_size, false);
940 if (!kmalloc_caches[type][aligned_idx])
941 kmalloc_caches[type][aligned_idx] = create_kmalloc_cache(
942 kmalloc_info[aligned_idx].name[type],
943 aligned_size, flags);
944 if (idx != aligned_idx)
945 kmalloc_caches[type][idx] = kmalloc_caches[type][aligned_idx];
949 * Create the kmalloc array. Some of the regular kmalloc arrays
950 * may already have been created because they were needed to
951 * enable allocations for slab creation.
953 void __init create_kmalloc_caches(slab_flags_t flags)
956 enum kmalloc_cache_type type;
959 * Including KMALLOC_CGROUP if CONFIG_MEMCG_KMEM defined
961 for (type = KMALLOC_NORMAL; type < NR_KMALLOC_TYPES; type++) {
962 for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
963 if (!kmalloc_caches[type][i])
964 new_kmalloc_cache(i, type, flags);
967 * Caches that are not of the two-to-the-power-of size.
968 * These have to be created immediately after the
969 * earlier power of two caches
971 if (KMALLOC_MIN_SIZE <= 32 && i == 6 &&
972 !kmalloc_caches[type][1])
973 new_kmalloc_cache(1, type, flags);
974 if (KMALLOC_MIN_SIZE <= 64 && i == 7 &&
975 !kmalloc_caches[type][2])
976 new_kmalloc_cache(2, type, flags);
979 #ifdef CONFIG_RANDOM_KMALLOC_CACHES
980 random_kmalloc_seed = get_random_u64();
983 /* Kmalloc array is now usable */
987 void free_large_kmalloc(struct folio *folio, void *object)
989 unsigned int order = folio_order(folio);
991 if (WARN_ON_ONCE(order == 0))
992 pr_warn_once("object pointer: 0x%p\n", object);
994 kmemleak_free(object);
995 kasan_kfree_large(object);
996 kmsan_kfree_large(object);
998 mod_lruvec_page_state(folio_page(folio, 0), NR_SLAB_UNRECLAIMABLE_B,
999 -(PAGE_SIZE << order));
1000 __free_pages(folio_page(folio, 0), order);
1003 static void *__kmalloc_large_node(size_t size, gfp_t flags, int node);
1004 static __always_inline
1005 void *__do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller)
1007 struct kmem_cache *s;
1010 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
1011 ret = __kmalloc_large_node(size, flags, node);
1012 trace_kmalloc(caller, ret, size,
1013 PAGE_SIZE << get_order(size), flags, node);
1017 s = kmalloc_slab(size, flags, caller);
1019 if (unlikely(ZERO_OR_NULL_PTR(s)))
1022 ret = __kmem_cache_alloc_node(s, flags, node, size, caller);
1023 ret = kasan_kmalloc(s, ret, size, flags);
1024 trace_kmalloc(caller, ret, size, s->size, flags, node);
1028 void *__kmalloc_node(size_t size, gfp_t flags, int node)
1030 return __do_kmalloc_node(size, flags, node, _RET_IP_);
1032 EXPORT_SYMBOL(__kmalloc_node);
1034 void *__kmalloc(size_t size, gfp_t flags)
1036 return __do_kmalloc_node(size, flags, NUMA_NO_NODE, _RET_IP_);
1038 EXPORT_SYMBOL(__kmalloc);
1040 void *__kmalloc_node_track_caller(size_t size, gfp_t flags,
1041 int node, unsigned long caller)
1043 return __do_kmalloc_node(size, flags, node, caller);
1045 EXPORT_SYMBOL(__kmalloc_node_track_caller);
1048 * kfree - free previously allocated memory
1049 * @object: pointer returned by kmalloc() or kmem_cache_alloc()
1051 * If @object is NULL, no operation is performed.
1053 void kfree(const void *object)
1055 struct folio *folio;
1057 struct kmem_cache *s;
1059 trace_kfree(_RET_IP_, object);
1061 if (unlikely(ZERO_OR_NULL_PTR(object)))
1064 folio = virt_to_folio(object);
1065 if (unlikely(!folio_test_slab(folio))) {
1066 free_large_kmalloc(folio, (void *)object);
1070 slab = folio_slab(folio);
1071 s = slab->slab_cache;
1072 __kmem_cache_free(s, (void *)object, _RET_IP_);
1074 EXPORT_SYMBOL(kfree);
1077 * __ksize -- Report full size of underlying allocation
1078 * @object: pointer to the object
1080 * This should only be used internally to query the true size of allocations.
1081 * It is not meant to be a way to discover the usable size of an allocation
1082 * after the fact. Instead, use kmalloc_size_roundup(). Using memory beyond
1083 * the originally requested allocation size may trigger KASAN, UBSAN_BOUNDS,
1084 * and/or FORTIFY_SOURCE.
1086 * Return: size of the actual memory used by @object in bytes
1088 size_t __ksize(const void *object)
1090 struct folio *folio;
1092 if (unlikely(object == ZERO_SIZE_PTR))
1095 folio = virt_to_folio(object);
1097 if (unlikely(!folio_test_slab(folio))) {
1098 if (WARN_ON(folio_size(folio) <= KMALLOC_MAX_CACHE_SIZE))
1100 if (WARN_ON(object != folio_address(folio)))
1102 return folio_size(folio);
1105 #ifdef CONFIG_SLUB_DEBUG
1106 skip_orig_size_check(folio_slab(folio)->slab_cache, object);
1109 return slab_ksize(folio_slab(folio)->slab_cache);
1112 void *kmalloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
1114 void *ret = __kmem_cache_alloc_node(s, gfpflags, NUMA_NO_NODE,
1117 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags, NUMA_NO_NODE);
1119 ret = kasan_kmalloc(s, ret, size, gfpflags);
1122 EXPORT_SYMBOL(kmalloc_trace);
1124 void *kmalloc_node_trace(struct kmem_cache *s, gfp_t gfpflags,
1125 int node, size_t size)
1127 void *ret = __kmem_cache_alloc_node(s, gfpflags, node, size, _RET_IP_);
1129 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags, node);
1131 ret = kasan_kmalloc(s, ret, size, gfpflags);
1134 EXPORT_SYMBOL(kmalloc_node_trace);
1136 gfp_t kmalloc_fix_flags(gfp_t flags)
1138 gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
1140 flags &= ~GFP_SLAB_BUG_MASK;
1141 pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
1142 invalid_mask, &invalid_mask, flags, &flags);
1149 * To avoid unnecessary overhead, we pass through large allocation requests
1150 * directly to the page allocator. We use __GFP_COMP, because we will need to
1151 * know the allocation order to free the pages properly in kfree.
1154 static void *__kmalloc_large_node(size_t size, gfp_t flags, int node)
1158 unsigned int order = get_order(size);
1160 if (unlikely(flags & GFP_SLAB_BUG_MASK))
1161 flags = kmalloc_fix_flags(flags);
1163 flags |= __GFP_COMP;
1164 page = alloc_pages_node(node, flags, order);
1166 ptr = page_address(page);
1167 mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B,
1168 PAGE_SIZE << order);
1171 ptr = kasan_kmalloc_large(ptr, size, flags);
1172 /* As ptr might get tagged, call kmemleak hook after KASAN. */
1173 kmemleak_alloc(ptr, size, 1, flags);
1174 kmsan_kmalloc_large(ptr, size, flags);
1179 void *kmalloc_large(size_t size, gfp_t flags)
1181 void *ret = __kmalloc_large_node(size, flags, NUMA_NO_NODE);
1183 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << get_order(size),
1184 flags, NUMA_NO_NODE);
1187 EXPORT_SYMBOL(kmalloc_large);
1189 void *kmalloc_large_node(size_t size, gfp_t flags, int node)
1191 void *ret = __kmalloc_large_node(size, flags, node);
1193 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << get_order(size),
1197 EXPORT_SYMBOL(kmalloc_large_node);
1199 #ifdef CONFIG_SLAB_FREELIST_RANDOM
1200 /* Randomize a generic freelist */
1201 static void freelist_randomize(unsigned int *list,
1207 for (i = 0; i < count; i++)
1210 /* Fisher-Yates shuffle */
1211 for (i = count - 1; i > 0; i--) {
1212 rand = get_random_u32_below(i + 1);
1213 swap(list[i], list[rand]);
1217 /* Create a random sequence per cache */
1218 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
1222 if (count < 2 || cachep->random_seq)
1225 cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp);
1226 if (!cachep->random_seq)
1229 freelist_randomize(cachep->random_seq, count);
1233 /* Destroy the per-cache random freelist sequence */
1234 void cache_random_seq_destroy(struct kmem_cache *cachep)
1236 kfree(cachep->random_seq);
1237 cachep->random_seq = NULL;
1239 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
1241 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
1243 #define SLABINFO_RIGHTS (0600)
1245 #define SLABINFO_RIGHTS (0400)
1248 static void print_slabinfo_header(struct seq_file *m)
1251 * Output format version, so at least we can change it
1252 * without _too_ many complaints.
1254 #ifdef CONFIG_DEBUG_SLAB
1255 seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
1257 seq_puts(m, "slabinfo - version: 2.1\n");
1259 seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
1260 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
1261 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
1262 #ifdef CONFIG_DEBUG_SLAB
1263 seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
1264 seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
1269 static void *slab_start(struct seq_file *m, loff_t *pos)
1271 mutex_lock(&slab_mutex);
1272 return seq_list_start(&slab_caches, *pos);
1275 static void *slab_next(struct seq_file *m, void *p, loff_t *pos)
1277 return seq_list_next(p, &slab_caches, pos);
1280 static void slab_stop(struct seq_file *m, void *p)
1282 mutex_unlock(&slab_mutex);
1285 static void cache_show(struct kmem_cache *s, struct seq_file *m)
1287 struct slabinfo sinfo;
1289 memset(&sinfo, 0, sizeof(sinfo));
1290 get_slabinfo(s, &sinfo);
1292 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
1293 s->name, sinfo.active_objs, sinfo.num_objs, s->size,
1294 sinfo.objects_per_slab, (1 << sinfo.cache_order));
1296 seq_printf(m, " : tunables %4u %4u %4u",
1297 sinfo.limit, sinfo.batchcount, sinfo.shared);
1298 seq_printf(m, " : slabdata %6lu %6lu %6lu",
1299 sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
1300 slabinfo_show_stats(m, s);
1304 static int slab_show(struct seq_file *m, void *p)
1306 struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
1308 if (p == slab_caches.next)
1309 print_slabinfo_header(m);
1314 void dump_unreclaimable_slab(void)
1316 struct kmem_cache *s;
1317 struct slabinfo sinfo;
1320 * Here acquiring slab_mutex is risky since we don't prefer to get
1321 * sleep in oom path. But, without mutex hold, it may introduce a
1323 * Use mutex_trylock to protect the list traverse, dump nothing
1324 * without acquiring the mutex.
1326 if (!mutex_trylock(&slab_mutex)) {
1327 pr_warn("excessive unreclaimable slab but cannot dump stats\n");
1331 pr_info("Unreclaimable slab info:\n");
1332 pr_info("Name Used Total\n");
1334 list_for_each_entry(s, &slab_caches, list) {
1335 if (s->flags & SLAB_RECLAIM_ACCOUNT)
1338 get_slabinfo(s, &sinfo);
1340 if (sinfo.num_objs > 0)
1341 pr_info("%-17s %10luKB %10luKB\n", s->name,
1342 (sinfo.active_objs * s->size) / 1024,
1343 (sinfo.num_objs * s->size) / 1024);
1345 mutex_unlock(&slab_mutex);
1349 * slabinfo_op - iterator that generates /proc/slabinfo
1358 * num-pages-per-slab
1359 * + further values on SMP and with statistics enabled
1361 static const struct seq_operations slabinfo_op = {
1362 .start = slab_start,
1368 static int slabinfo_open(struct inode *inode, struct file *file)
1370 return seq_open(file, &slabinfo_op);
1373 static const struct proc_ops slabinfo_proc_ops = {
1374 .proc_flags = PROC_ENTRY_PERMANENT,
1375 .proc_open = slabinfo_open,
1376 .proc_read = seq_read,
1377 .proc_write = slabinfo_write,
1378 .proc_lseek = seq_lseek,
1379 .proc_release = seq_release,
1382 static int __init slab_proc_init(void)
1384 proc_create("slabinfo", SLABINFO_RIGHTS, NULL, &slabinfo_proc_ops);
1387 module_init(slab_proc_init);
1389 #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */
1391 static __always_inline __realloc_size(2) void *
1392 __do_krealloc(const void *p, size_t new_size, gfp_t flags)
1397 /* Check for double-free before calling ksize. */
1398 if (likely(!ZERO_OR_NULL_PTR(p))) {
1399 if (!kasan_check_byte(p))
1405 /* If the object still fits, repoison it precisely. */
1406 if (ks >= new_size) {
1407 p = kasan_krealloc((void *)p, new_size, flags);
1411 ret = kmalloc_track_caller(new_size, flags);
1413 /* Disable KASAN checks as the object's redzone is accessed. */
1414 kasan_disable_current();
1415 memcpy(ret, kasan_reset_tag(p), ks);
1416 kasan_enable_current();
1423 * krealloc - reallocate memory. The contents will remain unchanged.
1424 * @p: object to reallocate memory for.
1425 * @new_size: how many bytes of memory are required.
1426 * @flags: the type of memory to allocate.
1428 * The contents of the object pointed to are preserved up to the
1429 * lesser of the new and old sizes (__GFP_ZERO flag is effectively ignored).
1430 * If @p is %NULL, krealloc() behaves exactly like kmalloc(). If @new_size
1431 * is 0 and @p is not a %NULL pointer, the object pointed to is freed.
1433 * Return: pointer to the allocated memory or %NULL in case of error
1435 void *krealloc(const void *p, size_t new_size, gfp_t flags)
1439 if (unlikely(!new_size)) {
1441 return ZERO_SIZE_PTR;
1444 ret = __do_krealloc(p, new_size, flags);
1445 if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret))
1450 EXPORT_SYMBOL(krealloc);
1453 * kfree_sensitive - Clear sensitive information in memory before freeing
1454 * @p: object to free memory of
1456 * The memory of the object @p points to is zeroed before freed.
1457 * If @p is %NULL, kfree_sensitive() does nothing.
1459 * Note: this function zeroes the whole allocated buffer which can be a good
1460 * deal bigger than the requested buffer size passed to kmalloc(). So be
1461 * careful when using this function in performance sensitive code.
1463 void kfree_sensitive(const void *p)
1466 void *mem = (void *)p;
1470 kasan_unpoison_range(mem, ks);
1471 memzero_explicit(mem, ks);
1475 EXPORT_SYMBOL(kfree_sensitive);
1477 size_t ksize(const void *objp)
1480 * We need to first check that the pointer to the object is valid.
1481 * The KASAN report printed from ksize() is more useful, then when
1482 * it's printed later when the behaviour could be undefined due to
1483 * a potential use-after-free or double-free.
1485 * We use kasan_check_byte(), which is supported for the hardware
1486 * tag-based KASAN mode, unlike kasan_check_read/write().
1488 * If the pointed to memory is invalid, we return 0 to avoid users of
1489 * ksize() writing to and potentially corrupting the memory region.
1491 * We want to perform the check before __ksize(), to avoid potentially
1492 * crashing in __ksize() due to accessing invalid metadata.
1494 if (unlikely(ZERO_OR_NULL_PTR(objp)) || !kasan_check_byte(objp))
1497 return kfence_ksize(objp) ?: __ksize(objp);
1499 EXPORT_SYMBOL(ksize);
1501 /* Tracepoints definitions. */
1502 EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1503 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1504 EXPORT_TRACEPOINT_SYMBOL(kfree);
1505 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);
1507 int should_failslab(struct kmem_cache *s, gfp_t gfpflags)
1509 if (__should_failslab(s, gfpflags))
1513 ALLOW_ERROR_INJECTION(should_failslab, ERRNO);