1 // SPDX-License-Identifier: GPL-2.0
3 * Slab allocator functions that are independent of the allocator strategy
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/kmemleak.h>
25 #include <linux/kasan.h>
26 #include <asm/cacheflush.h>
27 #include <asm/tlbflush.h>
29 #include <linux/memcontrol.h>
30 #include <linux/stackdepot.h>
35 #define CREATE_TRACE_POINTS
36 #include <trace/events/kmem.h>
38 enum slab_state slab_state;
39 LIST_HEAD(slab_caches);
40 DEFINE_MUTEX(slab_mutex);
41 struct kmem_cache *kmem_cache;
43 static LIST_HEAD(slab_caches_to_rcu_destroy);
44 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work);
45 static DECLARE_WORK(slab_caches_to_rcu_destroy_work,
46 slab_caches_to_rcu_destroy_workfn);
49 * Set of flags that will prevent slab merging
51 #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
52 SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \
53 SLAB_FAILSLAB | SLAB_NO_MERGE)
55 #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \
56 SLAB_CACHE_DMA32 | SLAB_ACCOUNT)
59 * Merge control. If this is set then no merging of slab caches will occur.
61 static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT);
63 static int __init setup_slab_nomerge(char *str)
69 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);
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)
109 * Figure out what the alignment of the objects will be given a set of
110 * flags, a user specified alignment and the size of the objects.
112 static unsigned int calculate_alignment(slab_flags_t flags,
113 unsigned int align, unsigned int size)
116 * If the user wants hardware cache aligned objects then follow that
117 * suggestion if the object is sufficiently large.
119 * The hardware cache alignment cannot override the specified
120 * alignment though. If that is greater then use it.
122 if (flags & SLAB_HWCACHE_ALIGN) {
125 ralign = cache_line_size();
126 while (size <= ralign / 2)
128 align = max(align, ralign);
131 align = max(align, arch_slab_minalign());
133 return ALIGN(align, sizeof(void *));
137 * Find a mergeable slab cache
139 int slab_unmergeable(struct kmem_cache *s)
141 if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
147 #ifdef CONFIG_HARDENED_USERCOPY
153 * We may have set a slab to be unmergeable during bootstrap.
161 struct kmem_cache *find_mergeable(unsigned int size, unsigned int align,
162 slab_flags_t flags, const char *name, void (*ctor)(void *))
164 struct kmem_cache *s;
172 size = ALIGN(size, sizeof(void *));
173 align = calculate_alignment(flags, align, size);
174 size = ALIGN(size, align);
175 flags = kmem_cache_flags(flags, name);
177 if (flags & SLAB_NEVER_MERGE)
180 list_for_each_entry_reverse(s, &slab_caches, list) {
181 if (slab_unmergeable(s))
187 if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
190 * Check if alignment is compatible.
191 * Courtesy of Adrian Drzewiecki
193 if ((s->size & ~(align - 1)) != s->size)
196 if (s->size - size >= sizeof(void *))
204 static struct kmem_cache *create_cache(const char *name,
205 unsigned int object_size, unsigned int align,
206 slab_flags_t flags, unsigned int useroffset,
207 unsigned int usersize, void (*ctor)(void *),
208 struct kmem_cache *root_cache)
210 struct kmem_cache *s;
213 if (WARN_ON(useroffset + usersize > object_size))
214 useroffset = usersize = 0;
217 s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
222 s->size = s->object_size = object_size;
225 #ifdef CONFIG_HARDENED_USERCOPY
226 s->useroffset = useroffset;
227 s->usersize = usersize;
230 err = __kmem_cache_create(s, flags);
235 list_add(&s->list, &slab_caches);
239 kmem_cache_free(kmem_cache, s);
245 * kmem_cache_create_usercopy - Create a cache with a region suitable
246 * for copying to userspace
247 * @name: A string which is used in /proc/slabinfo to identify this cache.
248 * @size: The size of objects to be created in this cache.
249 * @align: The required alignment for the objects.
251 * @useroffset: Usercopy region offset
252 * @usersize: Usercopy region size
253 * @ctor: A constructor for the objects.
255 * Cannot be called within a interrupt, but can be interrupted.
256 * The @ctor is run when new pages are allocated by the cache.
260 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
261 * to catch references to uninitialised memory.
263 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
264 * for buffer overruns.
266 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
267 * cacheline. This can be beneficial if you're counting cycles as closely
270 * Return: a pointer to the cache on success, NULL on failure.
273 kmem_cache_create_usercopy(const char *name,
274 unsigned int size, unsigned int align,
276 unsigned int useroffset, unsigned int usersize,
277 void (*ctor)(void *))
279 struct kmem_cache *s = NULL;
280 const char *cache_name;
283 #ifdef CONFIG_SLUB_DEBUG
285 * If no slab_debug was enabled globally, the static key is not yet
286 * enabled by setup_slub_debug(). Enable it if the cache is being
287 * created with any of the debugging flags passed explicitly.
288 * It's also possible that this is the first cache created with
289 * SLAB_STORE_USER and we should init stack_depot for it.
291 if (flags & SLAB_DEBUG_FLAGS)
292 static_branch_enable(&slub_debug_enabled);
293 if (flags & SLAB_STORE_USER)
297 mutex_lock(&slab_mutex);
299 err = kmem_cache_sanity_check(name, size);
304 /* Refuse requests with allocator specific flags */
305 if (flags & ~SLAB_FLAGS_PERMITTED) {
311 * Some allocators will constraint the set of valid flags to a subset
312 * of all flags. We expect them to define CACHE_CREATE_MASK in this
313 * case, and we'll just provide them with a sanitized version of the
316 flags &= CACHE_CREATE_MASK;
318 /* Fail closed on bad usersize of useroffset values. */
319 if (!IS_ENABLED(CONFIG_HARDENED_USERCOPY) ||
320 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 if (slab_state >= FULL) {
408 sysfs_slab_unlink(s);
409 sysfs_slab_release(s);
411 slab_kmem_cache_release(s);
415 static void kmem_cache_release(struct kmem_cache *s)
417 slab_kmem_cache_release(s);
421 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work)
423 LIST_HEAD(to_destroy);
424 struct kmem_cache *s, *s2;
427 * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the
428 * @slab_caches_to_rcu_destroy list. The slab pages are freed
429 * through RCU and the associated kmem_cache are dereferenced
430 * while freeing the pages, so the kmem_caches should be freed only
431 * after the pending RCU operations are finished. As rcu_barrier()
432 * is a pretty slow operation, we batch all pending destructions
435 mutex_lock(&slab_mutex);
436 list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy);
437 mutex_unlock(&slab_mutex);
439 if (list_empty(&to_destroy))
444 list_for_each_entry_safe(s, s2, &to_destroy, list) {
445 debugfs_slab_release(s);
446 kfence_shutdown_cache(s);
447 kmem_cache_release(s);
451 static int shutdown_cache(struct kmem_cache *s)
453 /* free asan quarantined objects */
454 kasan_cache_shutdown(s);
456 if (__kmem_cache_shutdown(s) != 0)
461 if (s->flags & SLAB_TYPESAFE_BY_RCU) {
462 list_add_tail(&s->list, &slab_caches_to_rcu_destroy);
463 schedule_work(&slab_caches_to_rcu_destroy_work);
465 kfence_shutdown_cache(s);
466 debugfs_slab_release(s);
472 void slab_kmem_cache_release(struct kmem_cache *s)
474 __kmem_cache_release(s);
475 kfree_const(s->name);
476 kmem_cache_free(kmem_cache, s);
479 void kmem_cache_destroy(struct kmem_cache *s)
484 if (unlikely(!s) || !kasan_check_byte(s))
488 mutex_lock(&slab_mutex);
490 rcu_set = s->flags & SLAB_TYPESAFE_BY_RCU;
496 err = shutdown_cache(s);
497 WARN(err, "%s %s: Slab cache still has objects when called from %pS",
498 __func__, s->name, (void *)_RET_IP_);
500 mutex_unlock(&slab_mutex);
502 if (!err && !rcu_set)
503 kmem_cache_release(s);
505 EXPORT_SYMBOL(kmem_cache_destroy);
508 * kmem_cache_shrink - Shrink a cache.
509 * @cachep: The cache to shrink.
511 * Releases as many slabs as possible for a cache.
512 * To help debugging, a zero exit status indicates all slabs were released.
514 * Return: %0 if all slabs were released, non-zero otherwise
516 int kmem_cache_shrink(struct kmem_cache *cachep)
518 kasan_cache_shrink(cachep);
520 return __kmem_cache_shrink(cachep);
522 EXPORT_SYMBOL(kmem_cache_shrink);
524 bool slab_is_available(void)
526 return slab_state >= UP;
530 static void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab)
532 if (__kfence_obj_info(kpp, object, slab))
534 __kmem_obj_info(kpp, object, slab);
538 * kmem_dump_obj - Print available slab provenance information
539 * @object: slab object for which to find provenance information.
541 * This function uses pr_cont(), so that the caller is expected to have
542 * printed out whatever preamble is appropriate. The provenance information
543 * depends on the type of object and on how much debugging is enabled.
544 * For a slab-cache object, the fact that it is a slab object is printed,
545 * and, if available, the slab name, return address, and stack trace from
546 * the allocation and last free path of that object.
548 * Return: %true if the pointer is to a not-yet-freed object from
549 * kmalloc() or kmem_cache_alloc(), either %true or %false if the pointer
550 * is to an already-freed object, and %false otherwise.
552 bool kmem_dump_obj(void *object)
554 char *cp = IS_ENABLED(CONFIG_MMU) ? "" : "/vmalloc";
557 unsigned long ptroffset;
558 struct kmem_obj_info kp = { };
560 /* Some arches consider ZERO_SIZE_PTR to be a valid address. */
561 if (object < (void *)PAGE_SIZE || !virt_addr_valid(object))
563 slab = virt_to_slab(object);
567 kmem_obj_info(&kp, object, slab);
568 if (kp.kp_slab_cache)
569 pr_cont(" slab%s %s", cp, kp.kp_slab_cache->name);
571 pr_cont(" slab%s", cp);
572 if (is_kfence_address(object))
573 pr_cont(" (kfence)");
575 pr_cont(" start %px", kp.kp_objp);
576 if (kp.kp_data_offset)
577 pr_cont(" data offset %lu", kp.kp_data_offset);
579 ptroffset = ((char *)object - (char *)kp.kp_objp) - kp.kp_data_offset;
580 pr_cont(" pointer offset %lu", ptroffset);
582 if (kp.kp_slab_cache && kp.kp_slab_cache->object_size)
583 pr_cont(" size %u", kp.kp_slab_cache->object_size);
585 pr_cont(" allocated at %pS\n", kp.kp_ret);
588 for (i = 0; i < ARRAY_SIZE(kp.kp_stack); i++) {
591 pr_info(" %pS\n", kp.kp_stack[i]);
594 if (kp.kp_free_stack[0])
595 pr_cont(" Free path:\n");
597 for (i = 0; i < ARRAY_SIZE(kp.kp_free_stack); i++) {
598 if (!kp.kp_free_stack[i])
600 pr_info(" %pS\n", kp.kp_free_stack[i]);
605 EXPORT_SYMBOL_GPL(kmem_dump_obj);
608 /* Create a cache during boot when no slab services are available yet */
609 void __init create_boot_cache(struct kmem_cache *s, const char *name,
610 unsigned int size, slab_flags_t flags,
611 unsigned int useroffset, unsigned int usersize)
614 unsigned int align = ARCH_KMALLOC_MINALIGN;
617 s->size = s->object_size = size;
620 * For power of two sizes, guarantee natural alignment for kmalloc
621 * caches, regardless of SL*B debugging options.
623 if (is_power_of_2(size))
624 align = max(align, size);
625 s->align = calculate_alignment(flags, align, size);
627 #ifdef CONFIG_HARDENED_USERCOPY
628 s->useroffset = useroffset;
629 s->usersize = usersize;
632 err = __kmem_cache_create(s, flags);
635 panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n",
638 s->refcount = -1; /* Exempt from merging for now */
641 static struct kmem_cache *__init create_kmalloc_cache(const char *name,
645 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
648 panic("Out of memory when creating slab %s\n", name);
650 create_boot_cache(s, name, size, flags | SLAB_KMALLOC, 0, size);
651 list_add(&s->list, &slab_caches);
657 kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1] __ro_after_init =
658 { /* initialization for https://llvm.org/pr42570 */ };
659 EXPORT_SYMBOL(kmalloc_caches);
661 #ifdef CONFIG_RANDOM_KMALLOC_CACHES
662 unsigned long random_kmalloc_seed __ro_after_init;
663 EXPORT_SYMBOL(random_kmalloc_seed);
667 * Conversion table for small slabs sizes / 8 to the index in the
668 * kmalloc array. This is necessary for slabs < 192 since we have non power
669 * of two cache sizes there. The size of larger slabs can be determined using
672 u8 kmalloc_size_index[24] __ro_after_init = {
699 size_t kmalloc_size_roundup(size_t size)
701 if (size && size <= KMALLOC_MAX_CACHE_SIZE) {
703 * The flags don't matter since size_index is common to all.
704 * Neither does the caller for just getting ->object_size.
706 return kmalloc_slab(size, GFP_KERNEL, 0)->object_size;
709 /* Above the smaller buckets, size is a multiple of page size. */
710 if (size && size <= KMALLOC_MAX_SIZE)
711 return PAGE_SIZE << get_order(size);
714 * Return 'size' for 0 - kmalloc() returns ZERO_SIZE_PTR
715 * and very large size - kmalloc() may fail.
720 EXPORT_SYMBOL(kmalloc_size_roundup);
722 #ifdef CONFIG_ZONE_DMA
723 #define KMALLOC_DMA_NAME(sz) .name[KMALLOC_DMA] = "dma-kmalloc-" #sz,
725 #define KMALLOC_DMA_NAME(sz)
728 #ifdef CONFIG_MEMCG_KMEM
729 #define KMALLOC_CGROUP_NAME(sz) .name[KMALLOC_CGROUP] = "kmalloc-cg-" #sz,
731 #define KMALLOC_CGROUP_NAME(sz)
734 #ifndef CONFIG_SLUB_TINY
735 #define KMALLOC_RCL_NAME(sz) .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #sz,
737 #define KMALLOC_RCL_NAME(sz)
740 #ifdef CONFIG_RANDOM_KMALLOC_CACHES
741 #define __KMALLOC_RANDOM_CONCAT(a, b) a ## b
742 #define KMALLOC_RANDOM_NAME(N, sz) __KMALLOC_RANDOM_CONCAT(KMA_RAND_, N)(sz)
743 #define KMA_RAND_1(sz) .name[KMALLOC_RANDOM_START + 1] = "kmalloc-rnd-01-" #sz,
744 #define KMA_RAND_2(sz) KMA_RAND_1(sz) .name[KMALLOC_RANDOM_START + 2] = "kmalloc-rnd-02-" #sz,
745 #define KMA_RAND_3(sz) KMA_RAND_2(sz) .name[KMALLOC_RANDOM_START + 3] = "kmalloc-rnd-03-" #sz,
746 #define KMA_RAND_4(sz) KMA_RAND_3(sz) .name[KMALLOC_RANDOM_START + 4] = "kmalloc-rnd-04-" #sz,
747 #define KMA_RAND_5(sz) KMA_RAND_4(sz) .name[KMALLOC_RANDOM_START + 5] = "kmalloc-rnd-05-" #sz,
748 #define KMA_RAND_6(sz) KMA_RAND_5(sz) .name[KMALLOC_RANDOM_START + 6] = "kmalloc-rnd-06-" #sz,
749 #define KMA_RAND_7(sz) KMA_RAND_6(sz) .name[KMALLOC_RANDOM_START + 7] = "kmalloc-rnd-07-" #sz,
750 #define KMA_RAND_8(sz) KMA_RAND_7(sz) .name[KMALLOC_RANDOM_START + 8] = "kmalloc-rnd-08-" #sz,
751 #define KMA_RAND_9(sz) KMA_RAND_8(sz) .name[KMALLOC_RANDOM_START + 9] = "kmalloc-rnd-09-" #sz,
752 #define KMA_RAND_10(sz) KMA_RAND_9(sz) .name[KMALLOC_RANDOM_START + 10] = "kmalloc-rnd-10-" #sz,
753 #define KMA_RAND_11(sz) KMA_RAND_10(sz) .name[KMALLOC_RANDOM_START + 11] = "kmalloc-rnd-11-" #sz,
754 #define KMA_RAND_12(sz) KMA_RAND_11(sz) .name[KMALLOC_RANDOM_START + 12] = "kmalloc-rnd-12-" #sz,
755 #define KMA_RAND_13(sz) KMA_RAND_12(sz) .name[KMALLOC_RANDOM_START + 13] = "kmalloc-rnd-13-" #sz,
756 #define KMA_RAND_14(sz) KMA_RAND_13(sz) .name[KMALLOC_RANDOM_START + 14] = "kmalloc-rnd-14-" #sz,
757 #define KMA_RAND_15(sz) KMA_RAND_14(sz) .name[KMALLOC_RANDOM_START + 15] = "kmalloc-rnd-15-" #sz,
758 #else // CONFIG_RANDOM_KMALLOC_CACHES
759 #define KMALLOC_RANDOM_NAME(N, sz)
762 #define INIT_KMALLOC_INFO(__size, __short_size) \
764 .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \
765 KMALLOC_RCL_NAME(__short_size) \
766 KMALLOC_CGROUP_NAME(__short_size) \
767 KMALLOC_DMA_NAME(__short_size) \
768 KMALLOC_RANDOM_NAME(RANDOM_KMALLOC_CACHES_NR, __short_size) \
773 * kmalloc_info[] is to make slab_debug=,kmalloc-xx option work at boot time.
774 * kmalloc_index() supports up to 2^21=2MB, so the final entry of the table is
777 const struct kmalloc_info_struct kmalloc_info[] __initconst = {
778 INIT_KMALLOC_INFO(0, 0),
779 INIT_KMALLOC_INFO(96, 96),
780 INIT_KMALLOC_INFO(192, 192),
781 INIT_KMALLOC_INFO(8, 8),
782 INIT_KMALLOC_INFO(16, 16),
783 INIT_KMALLOC_INFO(32, 32),
784 INIT_KMALLOC_INFO(64, 64),
785 INIT_KMALLOC_INFO(128, 128),
786 INIT_KMALLOC_INFO(256, 256),
787 INIT_KMALLOC_INFO(512, 512),
788 INIT_KMALLOC_INFO(1024, 1k),
789 INIT_KMALLOC_INFO(2048, 2k),
790 INIT_KMALLOC_INFO(4096, 4k),
791 INIT_KMALLOC_INFO(8192, 8k),
792 INIT_KMALLOC_INFO(16384, 16k),
793 INIT_KMALLOC_INFO(32768, 32k),
794 INIT_KMALLOC_INFO(65536, 64k),
795 INIT_KMALLOC_INFO(131072, 128k),
796 INIT_KMALLOC_INFO(262144, 256k),
797 INIT_KMALLOC_INFO(524288, 512k),
798 INIT_KMALLOC_INFO(1048576, 1M),
799 INIT_KMALLOC_INFO(2097152, 2M)
803 * Patch up the size_index table if we have strange large alignment
804 * requirements for the kmalloc array. This is only the case for
805 * MIPS it seems. The standard arches will not generate any code here.
807 * Largest permitted alignment is 256 bytes due to the way we
808 * handle the index determination for the smaller caches.
810 * Make sure that nothing crazy happens if someone starts tinkering
811 * around with ARCH_KMALLOC_MINALIGN
813 void __init setup_kmalloc_cache_index_table(void)
817 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
818 !is_power_of_2(KMALLOC_MIN_SIZE));
820 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
821 unsigned int elem = size_index_elem(i);
823 if (elem >= ARRAY_SIZE(kmalloc_size_index))
825 kmalloc_size_index[elem] = KMALLOC_SHIFT_LOW;
828 if (KMALLOC_MIN_SIZE >= 64) {
830 * The 96 byte sized cache is not used if the alignment
833 for (i = 64 + 8; i <= 96; i += 8)
834 kmalloc_size_index[size_index_elem(i)] = 7;
838 if (KMALLOC_MIN_SIZE >= 128) {
840 * The 192 byte sized cache is not used if the alignment
841 * is 128 byte. Redirect kmalloc to use the 256 byte cache
844 for (i = 128 + 8; i <= 192; i += 8)
845 kmalloc_size_index[size_index_elem(i)] = 8;
849 static unsigned int __kmalloc_minalign(void)
851 unsigned int minalign = dma_get_cache_alignment();
853 if (IS_ENABLED(CONFIG_DMA_BOUNCE_UNALIGNED_KMALLOC) &&
854 is_swiotlb_allocated())
855 minalign = ARCH_KMALLOC_MINALIGN;
857 return max(minalign, arch_slab_minalign());
861 new_kmalloc_cache(int idx, enum kmalloc_cache_type type)
863 slab_flags_t flags = 0;
864 unsigned int minalign = __kmalloc_minalign();
865 unsigned int aligned_size = kmalloc_info[idx].size;
866 int aligned_idx = idx;
868 if ((KMALLOC_RECLAIM != KMALLOC_NORMAL) && (type == KMALLOC_RECLAIM)) {
869 flags |= SLAB_RECLAIM_ACCOUNT;
870 } else if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_CGROUP)) {
871 if (mem_cgroup_kmem_disabled()) {
872 kmalloc_caches[type][idx] = kmalloc_caches[KMALLOC_NORMAL][idx];
875 flags |= SLAB_ACCOUNT;
876 } else if (IS_ENABLED(CONFIG_ZONE_DMA) && (type == KMALLOC_DMA)) {
877 flags |= SLAB_CACHE_DMA;
880 #ifdef CONFIG_RANDOM_KMALLOC_CACHES
881 if (type >= KMALLOC_RANDOM_START && type <= KMALLOC_RANDOM_END)
882 flags |= SLAB_NO_MERGE;
886 * If CONFIG_MEMCG_KMEM is enabled, disable cache merging for
887 * KMALLOC_NORMAL caches.
889 if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_NORMAL))
890 flags |= SLAB_NO_MERGE;
892 if (minalign > ARCH_KMALLOC_MINALIGN) {
893 aligned_size = ALIGN(aligned_size, minalign);
894 aligned_idx = __kmalloc_index(aligned_size, false);
897 if (!kmalloc_caches[type][aligned_idx])
898 kmalloc_caches[type][aligned_idx] = create_kmalloc_cache(
899 kmalloc_info[aligned_idx].name[type],
900 aligned_size, flags);
901 if (idx != aligned_idx)
902 kmalloc_caches[type][idx] = kmalloc_caches[type][aligned_idx];
906 * Create the kmalloc array. Some of the regular kmalloc arrays
907 * may already have been created because they were needed to
908 * enable allocations for slab creation.
910 void __init create_kmalloc_caches(void)
913 enum kmalloc_cache_type type;
916 * Including KMALLOC_CGROUP if CONFIG_MEMCG_KMEM defined
918 for (type = KMALLOC_NORMAL; type < NR_KMALLOC_TYPES; type++) {
919 for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
920 if (!kmalloc_caches[type][i])
921 new_kmalloc_cache(i, type);
924 * Caches that are not of the two-to-the-power-of size.
925 * These have to be created immediately after the
926 * earlier power of two caches
928 if (KMALLOC_MIN_SIZE <= 32 && i == 6 &&
929 !kmalloc_caches[type][1])
930 new_kmalloc_cache(1, type);
931 if (KMALLOC_MIN_SIZE <= 64 && i == 7 &&
932 !kmalloc_caches[type][2])
933 new_kmalloc_cache(2, type);
936 #ifdef CONFIG_RANDOM_KMALLOC_CACHES
937 random_kmalloc_seed = get_random_u64();
940 /* Kmalloc array is now usable */
945 * __ksize -- Report full size of underlying allocation
946 * @object: pointer to the object
948 * This should only be used internally to query the true size of allocations.
949 * It is not meant to be a way to discover the usable size of an allocation
950 * after the fact. Instead, use kmalloc_size_roundup(). Using memory beyond
951 * the originally requested allocation size may trigger KASAN, UBSAN_BOUNDS,
952 * and/or FORTIFY_SOURCE.
954 * Return: size of the actual memory used by @object in bytes
956 size_t __ksize(const void *object)
960 if (unlikely(object == ZERO_SIZE_PTR))
963 folio = virt_to_folio(object);
965 if (unlikely(!folio_test_slab(folio))) {
966 if (WARN_ON(folio_size(folio) <= KMALLOC_MAX_CACHE_SIZE))
968 if (WARN_ON(object != folio_address(folio)))
970 return folio_size(folio);
973 #ifdef CONFIG_SLUB_DEBUG
974 skip_orig_size_check(folio_slab(folio)->slab_cache, object);
977 return slab_ksize(folio_slab(folio)->slab_cache);
980 gfp_t kmalloc_fix_flags(gfp_t flags)
982 gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
984 flags &= ~GFP_SLAB_BUG_MASK;
985 pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
986 invalid_mask, &invalid_mask, flags, &flags);
992 #ifdef CONFIG_SLAB_FREELIST_RANDOM
993 /* Randomize a generic freelist */
994 static void freelist_randomize(unsigned int *list,
1000 for (i = 0; i < count; i++)
1003 /* Fisher-Yates shuffle */
1004 for (i = count - 1; i > 0; i--) {
1005 rand = get_random_u32_below(i + 1);
1006 swap(list[i], list[rand]);
1010 /* Create a random sequence per cache */
1011 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
1015 if (count < 2 || cachep->random_seq)
1018 cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp);
1019 if (!cachep->random_seq)
1022 freelist_randomize(cachep->random_seq, count);
1026 /* Destroy the per-cache random freelist sequence */
1027 void cache_random_seq_destroy(struct kmem_cache *cachep)
1029 kfree(cachep->random_seq);
1030 cachep->random_seq = NULL;
1032 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
1034 #ifdef CONFIG_SLUB_DEBUG
1035 #define SLABINFO_RIGHTS (0400)
1037 static void print_slabinfo_header(struct seq_file *m)
1040 * Output format version, so at least we can change it
1041 * without _too_ many complaints.
1043 seq_puts(m, "slabinfo - version: 2.1\n");
1044 seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
1045 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
1046 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
1050 static void *slab_start(struct seq_file *m, loff_t *pos)
1052 mutex_lock(&slab_mutex);
1053 return seq_list_start(&slab_caches, *pos);
1056 static void *slab_next(struct seq_file *m, void *p, loff_t *pos)
1058 return seq_list_next(p, &slab_caches, pos);
1061 static void slab_stop(struct seq_file *m, void *p)
1063 mutex_unlock(&slab_mutex);
1066 static void cache_show(struct kmem_cache *s, struct seq_file *m)
1068 struct slabinfo sinfo;
1070 memset(&sinfo, 0, sizeof(sinfo));
1071 get_slabinfo(s, &sinfo);
1073 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
1074 s->name, sinfo.active_objs, sinfo.num_objs, s->size,
1075 sinfo.objects_per_slab, (1 << sinfo.cache_order));
1077 seq_printf(m, " : tunables %4u %4u %4u",
1078 sinfo.limit, sinfo.batchcount, sinfo.shared);
1079 seq_printf(m, " : slabdata %6lu %6lu %6lu",
1080 sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
1081 slabinfo_show_stats(m, s);
1085 static int slab_show(struct seq_file *m, void *p)
1087 struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
1089 if (p == slab_caches.next)
1090 print_slabinfo_header(m);
1095 void dump_unreclaimable_slab(void)
1097 struct kmem_cache *s;
1098 struct slabinfo sinfo;
1101 * Here acquiring slab_mutex is risky since we don't prefer to get
1102 * sleep in oom path. But, without mutex hold, it may introduce a
1104 * Use mutex_trylock to protect the list traverse, dump nothing
1105 * without acquiring the mutex.
1107 if (!mutex_trylock(&slab_mutex)) {
1108 pr_warn("excessive unreclaimable slab but cannot dump stats\n");
1112 pr_info("Unreclaimable slab info:\n");
1113 pr_info("Name Used Total\n");
1115 list_for_each_entry(s, &slab_caches, list) {
1116 if (s->flags & SLAB_RECLAIM_ACCOUNT)
1119 get_slabinfo(s, &sinfo);
1121 if (sinfo.num_objs > 0)
1122 pr_info("%-17s %10luKB %10luKB\n", s->name,
1123 (sinfo.active_objs * s->size) / 1024,
1124 (sinfo.num_objs * s->size) / 1024);
1126 mutex_unlock(&slab_mutex);
1130 * slabinfo_op - iterator that generates /proc/slabinfo
1139 * num-pages-per-slab
1140 * + further values on SMP and with statistics enabled
1142 static const struct seq_operations slabinfo_op = {
1143 .start = slab_start,
1149 static int slabinfo_open(struct inode *inode, struct file *file)
1151 return seq_open(file, &slabinfo_op);
1154 static const struct proc_ops slabinfo_proc_ops = {
1155 .proc_flags = PROC_ENTRY_PERMANENT,
1156 .proc_open = slabinfo_open,
1157 .proc_read = seq_read,
1158 .proc_write = slabinfo_write,
1159 .proc_lseek = seq_lseek,
1160 .proc_release = seq_release,
1163 static int __init slab_proc_init(void)
1165 proc_create("slabinfo", SLABINFO_RIGHTS, NULL, &slabinfo_proc_ops);
1168 module_init(slab_proc_init);
1170 #endif /* CONFIG_SLUB_DEBUG */
1172 static __always_inline __realloc_size(2) void *
1173 __do_krealloc(const void *p, size_t new_size, gfp_t flags)
1178 /* Check for double-free before calling ksize. */
1179 if (likely(!ZERO_OR_NULL_PTR(p))) {
1180 if (!kasan_check_byte(p))
1186 /* If the object still fits, repoison it precisely. */
1187 if (ks >= new_size) {
1188 p = kasan_krealloc((void *)p, new_size, flags);
1192 ret = kmalloc_node_track_caller_noprof(new_size, flags, NUMA_NO_NODE, _RET_IP_);
1194 /* Disable KASAN checks as the object's redzone is accessed. */
1195 kasan_disable_current();
1196 memcpy(ret, kasan_reset_tag(p), ks);
1197 kasan_enable_current();
1204 * krealloc - reallocate memory. The contents will remain unchanged.
1205 * @p: object to reallocate memory for.
1206 * @new_size: how many bytes of memory are required.
1207 * @flags: the type of memory to allocate.
1209 * The contents of the object pointed to are preserved up to the
1210 * lesser of the new and old sizes (__GFP_ZERO flag is effectively ignored).
1211 * If @p is %NULL, krealloc() behaves exactly like kmalloc(). If @new_size
1212 * is 0 and @p is not a %NULL pointer, the object pointed to is freed.
1214 * Return: pointer to the allocated memory or %NULL in case of error
1216 void *krealloc_noprof(const void *p, size_t new_size, gfp_t flags)
1220 if (unlikely(!new_size)) {
1222 return ZERO_SIZE_PTR;
1225 ret = __do_krealloc(p, new_size, flags);
1226 if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret))
1231 EXPORT_SYMBOL(krealloc_noprof);
1234 * kfree_sensitive - Clear sensitive information in memory before freeing
1235 * @p: object to free memory of
1237 * The memory of the object @p points to is zeroed before freed.
1238 * If @p is %NULL, kfree_sensitive() does nothing.
1240 * Note: this function zeroes the whole allocated buffer which can be a good
1241 * deal bigger than the requested buffer size passed to kmalloc(). So be
1242 * careful when using this function in performance sensitive code.
1244 void kfree_sensitive(const void *p)
1247 void *mem = (void *)p;
1251 kasan_unpoison_range(mem, ks);
1252 memzero_explicit(mem, ks);
1256 EXPORT_SYMBOL(kfree_sensitive);
1258 size_t ksize(const void *objp)
1261 * We need to first check that the pointer to the object is valid.
1262 * The KASAN report printed from ksize() is more useful, then when
1263 * it's printed later when the behaviour could be undefined due to
1264 * a potential use-after-free or double-free.
1266 * We use kasan_check_byte(), which is supported for the hardware
1267 * tag-based KASAN mode, unlike kasan_check_read/write().
1269 * If the pointed to memory is invalid, we return 0 to avoid users of
1270 * ksize() writing to and potentially corrupting the memory region.
1272 * We want to perform the check before __ksize(), to avoid potentially
1273 * crashing in __ksize() due to accessing invalid metadata.
1275 if (unlikely(ZERO_OR_NULL_PTR(objp)) || !kasan_check_byte(objp))
1278 return kfence_ksize(objp) ?: __ksize(objp);
1280 EXPORT_SYMBOL(ksize);
1282 /* Tracepoints definitions. */
1283 EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1284 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1285 EXPORT_TRACEPOINT_SYMBOL(kfree);
1286 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);