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/module.h>
16 #include <linux/cpu.h>
17 #include <linux/uaccess.h>
18 #include <linux/seq_file.h>
19 #include <linux/proc_fs.h>
20 #include <linux/debugfs.h>
21 #include <linux/kasan.h>
22 #include <asm/cacheflush.h>
23 #include <asm/tlbflush.h>
25 #include <linux/memcontrol.h>
27 #define CREATE_TRACE_POINTS
28 #include <trace/events/kmem.h>
34 enum slab_state slab_state;
35 LIST_HEAD(slab_caches);
36 DEFINE_MUTEX(slab_mutex);
37 struct kmem_cache *kmem_cache;
39 #ifdef CONFIG_HARDENED_USERCOPY
40 bool usercopy_fallback __ro_after_init =
41 IS_ENABLED(CONFIG_HARDENED_USERCOPY_FALLBACK);
42 module_param(usercopy_fallback, bool, 0400);
43 MODULE_PARM_DESC(usercopy_fallback,
44 "WARN instead of reject usercopy whitelist violations");
47 static LIST_HEAD(slab_caches_to_rcu_destroy);
48 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work);
49 static DECLARE_WORK(slab_caches_to_rcu_destroy_work,
50 slab_caches_to_rcu_destroy_workfn);
53 * Set of flags that will prevent slab merging
55 #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
56 SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \
57 SLAB_FAILSLAB | kasan_never_merge())
59 #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \
60 SLAB_CACHE_DMA32 | SLAB_ACCOUNT)
63 * Merge control. If this is set then no merging of slab caches will occur.
65 static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT);
67 static int __init setup_slab_nomerge(char *str)
74 __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0);
77 __setup("slab_nomerge", setup_slab_nomerge);
80 * Determine the size of a slab object
82 unsigned int kmem_cache_size(struct kmem_cache *s)
84 return s->object_size;
86 EXPORT_SYMBOL(kmem_cache_size);
88 #ifdef CONFIG_DEBUG_VM
89 static int kmem_cache_sanity_check(const char *name, unsigned int size)
91 if (!name || in_interrupt() || size < sizeof(void *) ||
92 size > KMALLOC_MAX_SIZE) {
93 pr_err("kmem_cache_create(%s) integrity check failed\n", name);
97 WARN_ON(strchr(name, ' ')); /* It confuses parsers */
101 static inline int kmem_cache_sanity_check(const char *name, unsigned int size)
107 void __kmem_cache_free_bulk(struct kmem_cache *s, size_t nr, void **p)
111 for (i = 0; i < nr; i++) {
113 kmem_cache_free(s, p[i]);
119 int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t nr,
124 for (i = 0; i < nr; i++) {
125 void *x = p[i] = kmem_cache_alloc(s, flags);
127 __kmem_cache_free_bulk(s, i, p);
135 * Figure out what the alignment of the objects will be given a set of
136 * flags, a user specified alignment and the size of the objects.
138 static unsigned int calculate_alignment(slab_flags_t flags,
139 unsigned int align, unsigned int size)
142 * If the user wants hardware cache aligned objects then follow that
143 * suggestion if the object is sufficiently large.
145 * The hardware cache alignment cannot override the specified
146 * alignment though. If that is greater then use it.
148 if (flags & SLAB_HWCACHE_ALIGN) {
151 ralign = cache_line_size();
152 while (size <= ralign / 2)
154 align = max(align, ralign);
157 if (align < ARCH_SLAB_MINALIGN)
158 align = ARCH_SLAB_MINALIGN;
160 return ALIGN(align, sizeof(void *));
164 * Find a mergeable slab cache
166 int slab_unmergeable(struct kmem_cache *s)
168 if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
178 * We may have set a slab to be unmergeable during bootstrap.
186 struct kmem_cache *find_mergeable(unsigned int size, unsigned int align,
187 slab_flags_t flags, const char *name, void (*ctor)(void *))
189 struct kmem_cache *s;
197 size = ALIGN(size, sizeof(void *));
198 align = calculate_alignment(flags, align, size);
199 size = ALIGN(size, align);
200 flags = kmem_cache_flags(size, flags, name, NULL);
202 if (flags & SLAB_NEVER_MERGE)
205 list_for_each_entry_reverse(s, &slab_caches, list) {
206 if (slab_unmergeable(s))
212 if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
215 * Check if alignment is compatible.
216 * Courtesy of Adrian Drzewiecki
218 if ((s->size & ~(align - 1)) != s->size)
221 if (s->size - size >= sizeof(void *))
224 if (IS_ENABLED(CONFIG_SLAB) && align &&
225 (align > s->align || s->align % align))
233 static struct kmem_cache *create_cache(const char *name,
234 unsigned int object_size, unsigned int align,
235 slab_flags_t flags, unsigned int useroffset,
236 unsigned int usersize, void (*ctor)(void *),
237 struct kmem_cache *root_cache)
239 struct kmem_cache *s;
242 if (WARN_ON(useroffset + usersize > object_size))
243 useroffset = usersize = 0;
246 s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
251 s->size = s->object_size = object_size;
254 s->useroffset = useroffset;
255 s->usersize = usersize;
257 err = __kmem_cache_create(s, flags);
262 list_add(&s->list, &slab_caches);
269 kmem_cache_free(kmem_cache, s);
274 * kmem_cache_create_usercopy - Create a cache with a region suitable
275 * for copying to userspace
276 * @name: A string which is used in /proc/slabinfo to identify this cache.
277 * @size: The size of objects to be created in this cache.
278 * @align: The required alignment for the objects.
280 * @useroffset: Usercopy region offset
281 * @usersize: Usercopy region size
282 * @ctor: A constructor for the objects.
284 * Cannot be called within a interrupt, but can be interrupted.
285 * The @ctor is run when new pages are allocated by the cache.
289 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
290 * to catch references to uninitialised memory.
292 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
293 * for buffer overruns.
295 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
296 * cacheline. This can be beneficial if you're counting cycles as closely
299 * Return: a pointer to the cache on success, NULL on failure.
302 kmem_cache_create_usercopy(const char *name,
303 unsigned int size, unsigned int align,
305 unsigned int useroffset, unsigned int usersize,
306 void (*ctor)(void *))
308 struct kmem_cache *s = NULL;
309 const char *cache_name;
315 mutex_lock(&slab_mutex);
317 err = kmem_cache_sanity_check(name, size);
322 /* Refuse requests with allocator specific flags */
323 if (flags & ~SLAB_FLAGS_PERMITTED) {
329 * Some allocators will constraint the set of valid flags to a subset
330 * of all flags. We expect them to define CACHE_CREATE_MASK in this
331 * case, and we'll just provide them with a sanitized version of the
334 flags &= CACHE_CREATE_MASK;
336 /* Fail closed on bad usersize of useroffset values. */
337 if (WARN_ON(!usersize && useroffset) ||
338 WARN_ON(size < usersize || size - usersize < useroffset))
339 usersize = useroffset = 0;
342 s = __kmem_cache_alias(name, size, align, flags, ctor);
346 cache_name = kstrdup_const(name, GFP_KERNEL);
352 s = create_cache(cache_name, size,
353 calculate_alignment(flags, align, size),
354 flags, useroffset, usersize, ctor, NULL);
357 kfree_const(cache_name);
361 mutex_unlock(&slab_mutex);
367 if (flags & SLAB_PANIC)
368 panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n",
371 pr_warn("kmem_cache_create(%s) failed with error %d\n",
379 EXPORT_SYMBOL(kmem_cache_create_usercopy);
382 * kmem_cache_create - Create a cache.
383 * @name: A string which is used in /proc/slabinfo to identify this cache.
384 * @size: The size of objects to be created in this cache.
385 * @align: The required alignment for the objects.
387 * @ctor: A constructor for the objects.
389 * Cannot be called within a interrupt, but can be interrupted.
390 * The @ctor is run when new pages are allocated by the cache.
394 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
395 * to catch references to uninitialised memory.
397 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
398 * for buffer overruns.
400 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
401 * cacheline. This can be beneficial if you're counting cycles as closely
404 * Return: a pointer to the cache on success, NULL on failure.
407 kmem_cache_create(const char *name, unsigned int size, unsigned int align,
408 slab_flags_t flags, void (*ctor)(void *))
410 return kmem_cache_create_usercopy(name, size, align, flags, 0, 0,
413 EXPORT_SYMBOL(kmem_cache_create);
415 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work)
417 LIST_HEAD(to_destroy);
418 struct kmem_cache *s, *s2;
421 * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the
422 * @slab_caches_to_rcu_destroy list. The slab pages are freed
423 * through RCU and the associated kmem_cache are dereferenced
424 * while freeing the pages, so the kmem_caches should be freed only
425 * after the pending RCU operations are finished. As rcu_barrier()
426 * is a pretty slow operation, we batch all pending destructions
429 mutex_lock(&slab_mutex);
430 list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy);
431 mutex_unlock(&slab_mutex);
433 if (list_empty(&to_destroy))
438 list_for_each_entry_safe(s, s2, &to_destroy, list) {
439 #ifdef SLAB_SUPPORTS_SYSFS
440 sysfs_slab_release(s);
442 slab_kmem_cache_release(s);
447 static int shutdown_cache(struct kmem_cache *s)
449 /* free asan quarantined objects */
450 kasan_cache_shutdown(s);
452 if (__kmem_cache_shutdown(s) != 0)
457 if (s->flags & SLAB_TYPESAFE_BY_RCU) {
458 #ifdef SLAB_SUPPORTS_SYSFS
459 sysfs_slab_unlink(s);
461 list_add_tail(&s->list, &slab_caches_to_rcu_destroy);
462 schedule_work(&slab_caches_to_rcu_destroy_work);
464 #ifdef SLAB_SUPPORTS_SYSFS
465 sysfs_slab_unlink(s);
466 sysfs_slab_release(s);
468 slab_kmem_cache_release(s);
475 void slab_kmem_cache_release(struct kmem_cache *s)
477 __kmem_cache_release(s);
478 kfree_const(s->name);
479 kmem_cache_free(kmem_cache, s);
482 void kmem_cache_destroy(struct kmem_cache *s)
492 mutex_lock(&slab_mutex);
498 err = shutdown_cache(s);
500 pr_err("kmem_cache_destroy %s: Slab cache still has objects\n",
505 mutex_unlock(&slab_mutex);
510 EXPORT_SYMBOL(kmem_cache_destroy);
513 * kmem_cache_shrink - Shrink a cache.
514 * @cachep: The cache to shrink.
516 * Releases as many slabs as possible for a cache.
517 * To help debugging, a zero exit status indicates all slabs were released.
519 * Return: %0 if all slabs were released, non-zero otherwise
521 int kmem_cache_shrink(struct kmem_cache *cachep)
527 kasan_cache_shrink(cachep);
528 ret = __kmem_cache_shrink(cachep);
533 EXPORT_SYMBOL(kmem_cache_shrink);
535 bool slab_is_available(void)
537 return slab_state >= UP;
541 * kmem_valid_obj - does the pointer reference a valid slab object?
542 * @object: pointer to query.
544 * Return: %true if the pointer is to a not-yet-freed object from
545 * kmalloc() or kmem_cache_alloc(), either %true or %false if the pointer
546 * is to an already-freed object, and %false otherwise.
548 bool kmem_valid_obj(void *object)
552 /* Some arches consider ZERO_SIZE_PTR to be a valid address. */
553 if (object < (void *)PAGE_SIZE || !virt_addr_valid(object))
555 page = virt_to_head_page(object);
556 return PageSlab(page);
560 * kmem_dump_obj - Print available slab provenance information
561 * @object: slab object for which to find provenance information.
563 * This function uses pr_cont(), so that the caller is expected to have
564 * printed out whatever preamble is appropriate. The provenance information
565 * depends on the type of object and on how much debugging is enabled.
566 * For a slab-cache object, the fact that it is a slab object is printed,
567 * and, if available, the slab name, return address, and stack trace from
568 * the allocation of that object.
570 * This function will splat if passed a pointer to a non-slab object.
571 * If you are not sure what type of object you have, you should instead
572 * use mem_dump_obj().
574 void kmem_dump_obj(void *object)
576 char *cp = IS_ENABLED(CONFIG_MMU) ? "" : "/vmalloc";
579 unsigned long ptroffset;
580 struct kmem_obj_info kp = { };
582 if (WARN_ON_ONCE(!virt_addr_valid(object)))
584 page = virt_to_head_page(object);
585 if (WARN_ON_ONCE(!PageSlab(page))) {
586 pr_cont(" non-slab memory.\n");
589 kmem_obj_info(&kp, object, page);
590 if (kp.kp_slab_cache)
591 pr_cont(" slab%s %s", cp, kp.kp_slab_cache->name);
593 pr_cont(" slab%s", cp);
595 pr_cont(" start %px", kp.kp_objp);
596 if (kp.kp_data_offset)
597 pr_cont(" data offset %lu", kp.kp_data_offset);
599 ptroffset = ((char *)object - (char *)kp.kp_objp) - kp.kp_data_offset;
600 pr_cont(" pointer offset %lu", ptroffset);
602 if (kp.kp_slab_cache && kp.kp_slab_cache->usersize)
603 pr_cont(" size %u", kp.kp_slab_cache->usersize);
605 pr_cont(" allocated at %pS\n", kp.kp_ret);
608 for (i = 0; i < ARRAY_SIZE(kp.kp_stack); i++) {
611 pr_info(" %pS\n", kp.kp_stack[i]);
616 /* Create a cache during boot when no slab services are available yet */
617 void __init create_boot_cache(struct kmem_cache *s, const char *name,
618 unsigned int size, slab_flags_t flags,
619 unsigned int useroffset, unsigned int usersize)
622 unsigned int align = ARCH_KMALLOC_MINALIGN;
625 s->size = s->object_size = size;
628 * For power of two sizes, guarantee natural alignment for kmalloc
629 * caches, regardless of SL*B debugging options.
631 if (is_power_of_2(size))
632 align = max(align, size);
633 s->align = calculate_alignment(flags, align, size);
635 s->useroffset = useroffset;
636 s->usersize = usersize;
638 err = __kmem_cache_create(s, flags);
641 panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n",
644 s->refcount = -1; /* Exempt from merging for now */
647 struct kmem_cache *__init create_kmalloc_cache(const char *name,
648 unsigned int size, slab_flags_t flags,
649 unsigned int useroffset, unsigned int usersize)
651 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
654 panic("Out of memory when creating slab %s\n", name);
656 create_boot_cache(s, name, size, flags, useroffset, usersize);
657 list_add(&s->list, &slab_caches);
663 kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1] __ro_after_init =
664 { /* initialization for https://bugs.llvm.org/show_bug.cgi?id=42570 */ };
665 EXPORT_SYMBOL(kmalloc_caches);
668 * Conversion table for small slabs sizes / 8 to the index in the
669 * kmalloc array. This is necessary for slabs < 192 since we have non power
670 * of two cache sizes there. The size of larger slabs can be determined using
673 static u8 size_index[24] __ro_after_init = {
700 static inline unsigned int size_index_elem(unsigned int bytes)
702 return (bytes - 1) / 8;
706 * Find the kmem_cache structure that serves a given size of
709 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
715 return ZERO_SIZE_PTR;
717 index = size_index[size_index_elem(size)];
719 if (WARN_ON_ONCE(size > KMALLOC_MAX_CACHE_SIZE))
721 index = fls(size - 1);
724 return kmalloc_caches[kmalloc_type(flags)][index];
727 #ifdef CONFIG_ZONE_DMA
728 #define INIT_KMALLOC_INFO(__size, __short_size) \
730 .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \
731 .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #__short_size, \
732 .name[KMALLOC_DMA] = "dma-kmalloc-" #__short_size, \
736 #define INIT_KMALLOC_INFO(__size, __short_size) \
738 .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \
739 .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #__short_size, \
745 * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
746 * kmalloc_index() supports up to 2^26=64MB, so the final entry of the table is
749 const struct kmalloc_info_struct kmalloc_info[] __initconst = {
750 INIT_KMALLOC_INFO(0, 0),
751 INIT_KMALLOC_INFO(96, 96),
752 INIT_KMALLOC_INFO(192, 192),
753 INIT_KMALLOC_INFO(8, 8),
754 INIT_KMALLOC_INFO(16, 16),
755 INIT_KMALLOC_INFO(32, 32),
756 INIT_KMALLOC_INFO(64, 64),
757 INIT_KMALLOC_INFO(128, 128),
758 INIT_KMALLOC_INFO(256, 256),
759 INIT_KMALLOC_INFO(512, 512),
760 INIT_KMALLOC_INFO(1024, 1k),
761 INIT_KMALLOC_INFO(2048, 2k),
762 INIT_KMALLOC_INFO(4096, 4k),
763 INIT_KMALLOC_INFO(8192, 8k),
764 INIT_KMALLOC_INFO(16384, 16k),
765 INIT_KMALLOC_INFO(32768, 32k),
766 INIT_KMALLOC_INFO(65536, 64k),
767 INIT_KMALLOC_INFO(131072, 128k),
768 INIT_KMALLOC_INFO(262144, 256k),
769 INIT_KMALLOC_INFO(524288, 512k),
770 INIT_KMALLOC_INFO(1048576, 1M),
771 INIT_KMALLOC_INFO(2097152, 2M),
772 INIT_KMALLOC_INFO(4194304, 4M),
773 INIT_KMALLOC_INFO(8388608, 8M),
774 INIT_KMALLOC_INFO(16777216, 16M),
775 INIT_KMALLOC_INFO(33554432, 32M),
776 INIT_KMALLOC_INFO(67108864, 64M)
780 * Patch up the size_index table if we have strange large alignment
781 * requirements for the kmalloc array. This is only the case for
782 * MIPS it seems. The standard arches will not generate any code here.
784 * Largest permitted alignment is 256 bytes due to the way we
785 * handle the index determination for the smaller caches.
787 * Make sure that nothing crazy happens if someone starts tinkering
788 * around with ARCH_KMALLOC_MINALIGN
790 void __init setup_kmalloc_cache_index_table(void)
794 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
795 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
797 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
798 unsigned int elem = size_index_elem(i);
800 if (elem >= ARRAY_SIZE(size_index))
802 size_index[elem] = KMALLOC_SHIFT_LOW;
805 if (KMALLOC_MIN_SIZE >= 64) {
807 * The 96 byte size cache is not used if the alignment
810 for (i = 64 + 8; i <= 96; i += 8)
811 size_index[size_index_elem(i)] = 7;
815 if (KMALLOC_MIN_SIZE >= 128) {
817 * The 192 byte sized cache is not used if the alignment
818 * is 128 byte. Redirect kmalloc to use the 256 byte cache
821 for (i = 128 + 8; i <= 192; i += 8)
822 size_index[size_index_elem(i)] = 8;
827 new_kmalloc_cache(int idx, enum kmalloc_cache_type type, slab_flags_t flags)
829 if (type == KMALLOC_RECLAIM)
830 flags |= SLAB_RECLAIM_ACCOUNT;
832 kmalloc_caches[type][idx] = create_kmalloc_cache(
833 kmalloc_info[idx].name[type],
834 kmalloc_info[idx].size, flags, 0,
835 kmalloc_info[idx].size);
839 * Create the kmalloc array. Some of the regular kmalloc arrays
840 * may already have been created because they were needed to
841 * enable allocations for slab creation.
843 void __init create_kmalloc_caches(slab_flags_t flags)
846 enum kmalloc_cache_type type;
848 for (type = KMALLOC_NORMAL; type <= KMALLOC_RECLAIM; type++) {
849 for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
850 if (!kmalloc_caches[type][i])
851 new_kmalloc_cache(i, type, flags);
854 * Caches that are not of the two-to-the-power-of size.
855 * These have to be created immediately after the
856 * earlier power of two caches
858 if (KMALLOC_MIN_SIZE <= 32 && i == 6 &&
859 !kmalloc_caches[type][1])
860 new_kmalloc_cache(1, type, flags);
861 if (KMALLOC_MIN_SIZE <= 64 && i == 7 &&
862 !kmalloc_caches[type][2])
863 new_kmalloc_cache(2, type, flags);
867 /* Kmalloc array is now usable */
870 #ifdef CONFIG_ZONE_DMA
871 for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
872 struct kmem_cache *s = kmalloc_caches[KMALLOC_NORMAL][i];
875 kmalloc_caches[KMALLOC_DMA][i] = create_kmalloc_cache(
876 kmalloc_info[i].name[KMALLOC_DMA],
877 kmalloc_info[i].size,
878 SLAB_CACHE_DMA | flags, 0,
879 kmalloc_info[i].size);
884 #endif /* !CONFIG_SLOB */
886 gfp_t kmalloc_fix_flags(gfp_t flags)
888 gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
890 flags &= ~GFP_SLAB_BUG_MASK;
891 pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
892 invalid_mask, &invalid_mask, flags, &flags);
899 * To avoid unnecessary overhead, we pass through large allocation requests
900 * directly to the page allocator. We use __GFP_COMP, because we will need to
901 * know the allocation order to free the pages properly in kfree.
903 void *kmalloc_order(size_t size, gfp_t flags, unsigned int order)
908 if (unlikely(flags & GFP_SLAB_BUG_MASK))
909 flags = kmalloc_fix_flags(flags);
912 page = alloc_pages(flags, order);
914 ret = page_address(page);
915 mod_node_page_state(page_pgdat(page), NR_SLAB_UNRECLAIMABLE_B,
918 ret = kasan_kmalloc_large(ret, size, flags);
919 /* As ret might get tagged, call kmemleak hook after KASAN. */
920 kmemleak_alloc(ret, size, 1, flags);
923 EXPORT_SYMBOL(kmalloc_order);
925 #ifdef CONFIG_TRACING
926 void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
928 void *ret = kmalloc_order(size, flags, order);
929 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
932 EXPORT_SYMBOL(kmalloc_order_trace);
935 #ifdef CONFIG_SLAB_FREELIST_RANDOM
936 /* Randomize a generic freelist */
937 static void freelist_randomize(struct rnd_state *state, unsigned int *list,
943 for (i = 0; i < count; i++)
946 /* Fisher-Yates shuffle */
947 for (i = count - 1; i > 0; i--) {
948 rand = prandom_u32_state(state);
950 swap(list[i], list[rand]);
954 /* Create a random sequence per cache */
955 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
958 struct rnd_state state;
960 if (count < 2 || cachep->random_seq)
963 cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp);
964 if (!cachep->random_seq)
967 /* Get best entropy at this stage of boot */
968 prandom_seed_state(&state, get_random_long());
970 freelist_randomize(&state, cachep->random_seq, count);
974 /* Destroy the per-cache random freelist sequence */
975 void cache_random_seq_destroy(struct kmem_cache *cachep)
977 kfree(cachep->random_seq);
978 cachep->random_seq = NULL;
980 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
982 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
984 #define SLABINFO_RIGHTS (0600)
986 #define SLABINFO_RIGHTS (0400)
989 static void print_slabinfo_header(struct seq_file *m)
992 * Output format version, so at least we can change it
993 * without _too_ many complaints.
995 #ifdef CONFIG_DEBUG_SLAB
996 seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
998 seq_puts(m, "slabinfo - version: 2.1\n");
1000 seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
1001 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
1002 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
1003 #ifdef CONFIG_DEBUG_SLAB
1004 seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
1005 seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
1010 void *slab_start(struct seq_file *m, loff_t *pos)
1012 mutex_lock(&slab_mutex);
1013 return seq_list_start(&slab_caches, *pos);
1016 void *slab_next(struct seq_file *m, void *p, loff_t *pos)
1018 return seq_list_next(p, &slab_caches, pos);
1021 void slab_stop(struct seq_file *m, void *p)
1023 mutex_unlock(&slab_mutex);
1026 static void cache_show(struct kmem_cache *s, struct seq_file *m)
1028 struct slabinfo sinfo;
1030 memset(&sinfo, 0, sizeof(sinfo));
1031 get_slabinfo(s, &sinfo);
1033 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
1034 s->name, sinfo.active_objs, sinfo.num_objs, s->size,
1035 sinfo.objects_per_slab, (1 << sinfo.cache_order));
1037 seq_printf(m, " : tunables %4u %4u %4u",
1038 sinfo.limit, sinfo.batchcount, sinfo.shared);
1039 seq_printf(m, " : slabdata %6lu %6lu %6lu",
1040 sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
1041 slabinfo_show_stats(m, s);
1045 static int slab_show(struct seq_file *m, void *p)
1047 struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
1049 if (p == slab_caches.next)
1050 print_slabinfo_header(m);
1055 void dump_unreclaimable_slab(void)
1057 struct kmem_cache *s;
1058 struct slabinfo sinfo;
1061 * Here acquiring slab_mutex is risky since we don't prefer to get
1062 * sleep in oom path. But, without mutex hold, it may introduce a
1064 * Use mutex_trylock to protect the list traverse, dump nothing
1065 * without acquiring the mutex.
1067 if (!mutex_trylock(&slab_mutex)) {
1068 pr_warn("excessive unreclaimable slab but cannot dump stats\n");
1072 pr_info("Unreclaimable slab info:\n");
1073 pr_info("Name Used Total\n");
1075 list_for_each_entry(s, &slab_caches, list) {
1076 if (s->flags & SLAB_RECLAIM_ACCOUNT)
1079 get_slabinfo(s, &sinfo);
1081 if (sinfo.num_objs > 0)
1082 pr_info("%-17s %10luKB %10luKB\n", s->name,
1083 (sinfo.active_objs * s->size) / 1024,
1084 (sinfo.num_objs * s->size) / 1024);
1086 mutex_unlock(&slab_mutex);
1089 #if defined(CONFIG_MEMCG_KMEM)
1090 int memcg_slab_show(struct seq_file *m, void *p)
1094 * Please, take a look at tools/cgroup/slabinfo.py .
1101 * slabinfo_op - iterator that generates /proc/slabinfo
1110 * num-pages-per-slab
1111 * + further values on SMP and with statistics enabled
1113 static const struct seq_operations slabinfo_op = {
1114 .start = slab_start,
1120 static int slabinfo_open(struct inode *inode, struct file *file)
1122 return seq_open(file, &slabinfo_op);
1125 static const struct proc_ops slabinfo_proc_ops = {
1126 .proc_flags = PROC_ENTRY_PERMANENT,
1127 .proc_open = slabinfo_open,
1128 .proc_read = seq_read,
1129 .proc_write = slabinfo_write,
1130 .proc_lseek = seq_lseek,
1131 .proc_release = seq_release,
1134 static int __init slab_proc_init(void)
1136 proc_create("slabinfo", SLABINFO_RIGHTS, NULL, &slabinfo_proc_ops);
1139 module_init(slab_proc_init);
1141 #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */
1143 static __always_inline void *__do_krealloc(const void *p, size_t new_size,
1151 if (ks >= new_size) {
1152 p = kasan_krealloc((void *)p, new_size, flags);
1156 ret = kmalloc_track_caller(new_size, flags);
1164 * krealloc - reallocate memory. The contents will remain unchanged.
1165 * @p: object to reallocate memory for.
1166 * @new_size: how many bytes of memory are required.
1167 * @flags: the type of memory to allocate.
1169 * The contents of the object pointed to are preserved up to the
1170 * lesser of the new and old sizes (__GFP_ZERO flag is effectively ignored).
1171 * If @p is %NULL, krealloc() behaves exactly like kmalloc(). If @new_size
1172 * is 0 and @p is not a %NULL pointer, the object pointed to is freed.
1174 * Return: pointer to the allocated memory or %NULL in case of error
1176 void *krealloc(const void *p, size_t new_size, gfp_t flags)
1180 if (unlikely(!new_size)) {
1182 return ZERO_SIZE_PTR;
1185 ret = __do_krealloc(p, new_size, flags);
1186 if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret))
1191 EXPORT_SYMBOL(krealloc);
1194 * kfree_sensitive - Clear sensitive information in memory before freeing
1195 * @p: object to free memory of
1197 * The memory of the object @p points to is zeroed before freed.
1198 * If @p is %NULL, kfree_sensitive() does nothing.
1200 * Note: this function zeroes the whole allocated buffer which can be a good
1201 * deal bigger than the requested buffer size passed to kmalloc(). So be
1202 * careful when using this function in performance sensitive code.
1204 void kfree_sensitive(const void *p)
1207 void *mem = (void *)p;
1211 memzero_explicit(mem, ks);
1214 EXPORT_SYMBOL(kfree_sensitive);
1217 * ksize - get the actual amount of memory allocated for a given object
1218 * @objp: Pointer to the object
1220 * kmalloc may internally round up allocations and return more memory
1221 * than requested. ksize() can be used to determine the actual amount of
1222 * memory allocated. The caller may use this additional memory, even though
1223 * a smaller amount of memory was initially specified with the kmalloc call.
1224 * The caller must guarantee that objp points to a valid object previously
1225 * allocated with either kmalloc() or kmem_cache_alloc(). The object
1226 * must not be freed during the duration of the call.
1228 * Return: size of the actual memory used by @objp in bytes
1230 size_t ksize(const void *objp)
1235 * We need to check that the pointed to object is valid, and only then
1236 * unpoison the shadow memory below. We use __kasan_check_read(), to
1237 * generate a more useful report at the time ksize() is called (rather
1238 * than later where behaviour is undefined due to potential
1239 * use-after-free or double-free).
1241 * If the pointed to memory is invalid we return 0, to avoid users of
1242 * ksize() writing to and potentially corrupting the memory region.
1244 * We want to perform the check before __ksize(), to avoid potentially
1245 * crashing in __ksize() due to accessing invalid metadata.
1247 if (unlikely(ZERO_OR_NULL_PTR(objp)) || !__kasan_check_read(objp, 1))
1250 size = __ksize(objp);
1252 * We assume that ksize callers could use whole allocated area,
1253 * so we need to unpoison this area.
1255 kasan_unpoison_range(objp, size);
1258 EXPORT_SYMBOL(ksize);
1260 /* Tracepoints definitions. */
1261 EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1262 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1263 EXPORT_TRACEPOINT_SYMBOL(kmalloc_node);
1264 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node);
1265 EXPORT_TRACEPOINT_SYMBOL(kfree);
1266 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);
1268 int should_failslab(struct kmem_cache *s, gfp_t gfpflags)
1270 if (__should_failslab(s, gfpflags))
1274 ALLOW_ERROR_INJECTION(should_failslab, ERRNO);