1 // SPDX-License-Identifier: GPL-2.0
3 * Slab allocator functions that are independent of the allocator strategy
5 * (C) 2012 Christoph Lameter <cl@linux.com>
7 #include <linux/slab.h>
10 #include <linux/poison.h>
11 #include <linux/interrupt.h>
12 #include <linux/memory.h>
13 #include <linux/cache.h>
14 #include <linux/compiler.h>
15 #include <linux/kfence.h>
16 #include <linux/module.h>
17 #include <linux/cpu.h>
18 #include <linux/uaccess.h>
19 #include <linux/seq_file.h>
20 #include <linux/proc_fs.h>
21 #include <linux/debugfs.h>
22 #include <linux/kasan.h>
23 #include <asm/cacheflush.h>
24 #include <asm/tlbflush.h>
26 #include <linux/memcontrol.h>
27 #include <linux/stackdepot.h>
32 #define CREATE_TRACE_POINTS
33 #include <trace/events/kmem.h>
35 enum slab_state slab_state;
36 LIST_HEAD(slab_caches);
37 DEFINE_MUTEX(slab_mutex);
38 struct kmem_cache *kmem_cache;
40 static LIST_HEAD(slab_caches_to_rcu_destroy);
41 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work);
42 static DECLARE_WORK(slab_caches_to_rcu_destroy_work,
43 slab_caches_to_rcu_destroy_workfn);
46 * Set of flags that will prevent slab merging
48 #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
49 SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \
50 SLAB_FAILSLAB | kasan_never_merge())
52 #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \
53 SLAB_CACHE_DMA32 | SLAB_ACCOUNT)
56 * Merge control. If this is set then no merging of slab caches will occur.
58 static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT);
60 static int __init setup_slab_nomerge(char *str)
66 static int __init setup_slab_merge(char *str)
73 __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0);
74 __setup_param("slub_merge", slub_merge, setup_slab_merge, 0);
77 __setup("slab_nomerge", setup_slab_nomerge);
78 __setup("slab_merge", setup_slab_merge);
81 * Determine the size of a slab object
83 unsigned int kmem_cache_size(struct kmem_cache *s)
85 return s->object_size;
87 EXPORT_SYMBOL(kmem_cache_size);
89 #ifdef CONFIG_DEBUG_VM
90 static int kmem_cache_sanity_check(const char *name, unsigned int size)
92 if (!name || in_interrupt() || size > KMALLOC_MAX_SIZE) {
93 pr_err("kmem_cache_create(%s) integrity check failed\n", name);
97 WARN_ON(strchr(name, ' ')); /* It confuses parsers */
101 static inline int kmem_cache_sanity_check(const char *name, unsigned int size)
108 * Figure out what the alignment of the objects will be given a set of
109 * flags, a user specified alignment and the size of the objects.
111 static unsigned int calculate_alignment(slab_flags_t flags,
112 unsigned int align, unsigned int size)
115 * If the user wants hardware cache aligned objects then follow that
116 * suggestion if the object is sufficiently large.
118 * The hardware cache alignment cannot override the specified
119 * alignment though. If that is greater then use it.
121 if (flags & SLAB_HWCACHE_ALIGN) {
124 ralign = cache_line_size();
125 while (size <= ralign / 2)
127 align = max(align, ralign);
130 align = max(align, arch_slab_minalign());
132 return ALIGN(align, sizeof(void *));
136 * Find a mergeable slab cache
138 int slab_unmergeable(struct kmem_cache *s)
140 if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
150 * We may have set a slab to be unmergeable during bootstrap.
158 struct kmem_cache *find_mergeable(unsigned int size, unsigned int align,
159 slab_flags_t flags, const char *name, void (*ctor)(void *))
161 struct kmem_cache *s;
169 size = ALIGN(size, sizeof(void *));
170 align = calculate_alignment(flags, align, size);
171 size = ALIGN(size, align);
172 flags = kmem_cache_flags(size, flags, name);
174 if (flags & SLAB_NEVER_MERGE)
177 list_for_each_entry_reverse(s, &slab_caches, list) {
178 if (slab_unmergeable(s))
184 if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
187 * Check if alignment is compatible.
188 * Courtesy of Adrian Drzewiecki
190 if ((s->size & ~(align - 1)) != s->size)
193 if (s->size - size >= sizeof(void *))
196 if (IS_ENABLED(CONFIG_SLAB) && align &&
197 (align > s->align || s->align % align))
205 static struct kmem_cache *create_cache(const char *name,
206 unsigned int object_size, unsigned int align,
207 slab_flags_t flags, unsigned int useroffset,
208 unsigned int usersize, void (*ctor)(void *),
209 struct kmem_cache *root_cache)
211 struct kmem_cache *s;
214 if (WARN_ON(useroffset + usersize > object_size))
215 useroffset = usersize = 0;
218 s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
223 s->size = s->object_size = object_size;
226 s->useroffset = useroffset;
227 s->usersize = usersize;
229 err = __kmem_cache_create(s, flags);
234 list_add(&s->list, &slab_caches);
241 kmem_cache_free(kmem_cache, s);
246 * kmem_cache_create_usercopy - Create a cache with a region suitable
247 * for copying to userspace
248 * @name: A string which is used in /proc/slabinfo to identify this cache.
249 * @size: The size of objects to be created in this cache.
250 * @align: The required alignment for the objects.
252 * @useroffset: Usercopy region offset
253 * @usersize: Usercopy region size
254 * @ctor: A constructor for the objects.
256 * Cannot be called within a interrupt, but can be interrupted.
257 * The @ctor is run when new pages are allocated by the cache.
261 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
262 * to catch references to uninitialised memory.
264 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
265 * for buffer overruns.
267 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
268 * cacheline. This can be beneficial if you're counting cycles as closely
271 * Return: a pointer to the cache on success, NULL on failure.
274 kmem_cache_create_usercopy(const char *name,
275 unsigned int size, unsigned int align,
277 unsigned int useroffset, unsigned int usersize,
278 void (*ctor)(void *))
280 struct kmem_cache *s = NULL;
281 const char *cache_name;
284 #ifdef CONFIG_SLUB_DEBUG
286 * If no slub_debug was enabled globally, the static key is not yet
287 * enabled by setup_slub_debug(). Enable it if the cache is being
288 * created with any of the debugging flags passed explicitly.
289 * It's also possible that this is the first cache created with
290 * SLAB_STORE_USER and we should init stack_depot for it.
292 if (flags & SLAB_DEBUG_FLAGS)
293 static_branch_enable(&slub_debug_enabled);
294 if (flags & SLAB_STORE_USER)
298 mutex_lock(&slab_mutex);
300 err = kmem_cache_sanity_check(name, size);
305 /* Refuse requests with allocator specific flags */
306 if (flags & ~SLAB_FLAGS_PERMITTED) {
312 * Some allocators will constraint the set of valid flags to a subset
313 * of all flags. We expect them to define CACHE_CREATE_MASK in this
314 * case, and we'll just provide them with a sanitized version of the
317 flags &= CACHE_CREATE_MASK;
319 /* Fail closed on bad usersize of useroffset values. */
320 if (WARN_ON(!usersize && useroffset) ||
321 WARN_ON(size < usersize || size - usersize < useroffset))
322 usersize = useroffset = 0;
325 s = __kmem_cache_alias(name, size, align, flags, ctor);
329 cache_name = kstrdup_const(name, GFP_KERNEL);
335 s = create_cache(cache_name, size,
336 calculate_alignment(flags, align, size),
337 flags, useroffset, usersize, ctor, NULL);
340 kfree_const(cache_name);
344 mutex_unlock(&slab_mutex);
347 if (flags & SLAB_PANIC)
348 panic("%s: Failed to create slab '%s'. Error %d\n",
349 __func__, name, err);
351 pr_warn("%s(%s) failed with error %d\n",
352 __func__, name, err);
359 EXPORT_SYMBOL(kmem_cache_create_usercopy);
362 * kmem_cache_create - Create a cache.
363 * @name: A string which is used in /proc/slabinfo to identify this cache.
364 * @size: The size of objects to be created in this cache.
365 * @align: The required alignment for the objects.
367 * @ctor: A constructor for the objects.
369 * Cannot be called within a interrupt, but can be interrupted.
370 * The @ctor is run when new pages are allocated by the cache.
374 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
375 * to catch references to uninitialised memory.
377 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
378 * for buffer overruns.
380 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
381 * cacheline. This can be beneficial if you're counting cycles as closely
384 * Return: a pointer to the cache on success, NULL on failure.
387 kmem_cache_create(const char *name, unsigned int size, unsigned int align,
388 slab_flags_t flags, void (*ctor)(void *))
390 return kmem_cache_create_usercopy(name, size, align, flags, 0, 0,
393 EXPORT_SYMBOL(kmem_cache_create);
395 #ifdef SLAB_SUPPORTS_SYSFS
397 * For a given kmem_cache, kmem_cache_destroy() should only be called
398 * once or there will be a use-after-free problem. The actual deletion
399 * and release of the kobject does not need slab_mutex or cpu_hotplug_lock
400 * protection. So they are now done without holding those locks.
402 * Note that there will be a slight delay in the deletion of sysfs files
403 * if kmem_cache_release() is called indrectly from a work function.
405 static void kmem_cache_release(struct kmem_cache *s)
407 sysfs_slab_unlink(s);
408 sysfs_slab_release(s);
411 static void kmem_cache_release(struct kmem_cache *s)
413 slab_kmem_cache_release(s);
417 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work)
419 LIST_HEAD(to_destroy);
420 struct kmem_cache *s, *s2;
423 * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the
424 * @slab_caches_to_rcu_destroy list. The slab pages are freed
425 * through RCU and the associated kmem_cache are dereferenced
426 * while freeing the pages, so the kmem_caches should be freed only
427 * after the pending RCU operations are finished. As rcu_barrier()
428 * is a pretty slow operation, we batch all pending destructions
431 mutex_lock(&slab_mutex);
432 list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy);
433 mutex_unlock(&slab_mutex);
435 if (list_empty(&to_destroy))
440 list_for_each_entry_safe(s, s2, &to_destroy, list) {
441 debugfs_slab_release(s);
442 kfence_shutdown_cache(s);
443 kmem_cache_release(s);
447 static int shutdown_cache(struct kmem_cache *s)
449 /* free asan quarantined objects */
450 kasan_cache_shutdown(s);
452 if (__kmem_cache_shutdown(s) != 0)
457 if (s->flags & SLAB_TYPESAFE_BY_RCU) {
458 list_add_tail(&s->list, &slab_caches_to_rcu_destroy);
459 schedule_work(&slab_caches_to_rcu_destroy_work);
461 kfence_shutdown_cache(s);
462 debugfs_slab_release(s);
468 void slab_kmem_cache_release(struct kmem_cache *s)
470 __kmem_cache_release(s);
471 kfree_const(s->name);
472 kmem_cache_free(kmem_cache, s);
475 void kmem_cache_destroy(struct kmem_cache *s)
480 if (unlikely(!s) || !kasan_check_byte(s))
484 mutex_lock(&slab_mutex);
486 rcu_set = s->flags & SLAB_TYPESAFE_BY_RCU;
488 refcnt = --s->refcount;
492 WARN(shutdown_cache(s),
493 "%s %s: Slab cache still has objects when called from %pS",
494 __func__, s->name, (void *)_RET_IP_);
496 mutex_unlock(&slab_mutex);
498 if (!refcnt && !rcu_set)
499 kmem_cache_release(s);
501 EXPORT_SYMBOL(kmem_cache_destroy);
504 * kmem_cache_shrink - Shrink a cache.
505 * @cachep: The cache to shrink.
507 * Releases as many slabs as possible for a cache.
508 * To help debugging, a zero exit status indicates all slabs were released.
510 * Return: %0 if all slabs were released, non-zero otherwise
512 int kmem_cache_shrink(struct kmem_cache *cachep)
517 kasan_cache_shrink(cachep);
518 ret = __kmem_cache_shrink(cachep);
522 EXPORT_SYMBOL(kmem_cache_shrink);
524 bool slab_is_available(void)
526 return slab_state >= UP;
531 * kmem_valid_obj - does the pointer reference a valid slab object?
532 * @object: pointer to query.
534 * Return: %true if the pointer is to a not-yet-freed object from
535 * kmalloc() or kmem_cache_alloc(), either %true or %false if the pointer
536 * is to an already-freed object, and %false otherwise.
538 bool kmem_valid_obj(void *object)
542 /* Some arches consider ZERO_SIZE_PTR to be a valid address. */
543 if (object < (void *)PAGE_SIZE || !virt_addr_valid(object))
545 folio = virt_to_folio(object);
546 return folio_test_slab(folio);
548 EXPORT_SYMBOL_GPL(kmem_valid_obj);
550 static void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab)
552 if (__kfence_obj_info(kpp, object, slab))
554 __kmem_obj_info(kpp, object, slab);
558 * kmem_dump_obj - Print available slab provenance information
559 * @object: slab object for which to find provenance information.
561 * This function uses pr_cont(), so that the caller is expected to have
562 * printed out whatever preamble is appropriate. The provenance information
563 * depends on the type of object and on how much debugging is enabled.
564 * For a slab-cache object, the fact that it is a slab object is printed,
565 * and, if available, the slab name, return address, and stack trace from
566 * the allocation and last free path of that object.
568 * This function will splat if passed a pointer to a non-slab object.
569 * If you are not sure what type of object you have, you should instead
570 * use mem_dump_obj().
572 void kmem_dump_obj(void *object)
574 char *cp = IS_ENABLED(CONFIG_MMU) ? "" : "/vmalloc";
577 unsigned long ptroffset;
578 struct kmem_obj_info kp = { };
580 if (WARN_ON_ONCE(!virt_addr_valid(object)))
582 slab = virt_to_slab(object);
583 if (WARN_ON_ONCE(!slab)) {
584 pr_cont(" non-slab memory.\n");
587 kmem_obj_info(&kp, object, slab);
588 if (kp.kp_slab_cache)
589 pr_cont(" slab%s %s", cp, kp.kp_slab_cache->name);
591 pr_cont(" slab%s", cp);
592 if (is_kfence_address(object))
593 pr_cont(" (kfence)");
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]);
614 if (kp.kp_free_stack[0])
615 pr_cont(" Free path:\n");
617 for (i = 0; i < ARRAY_SIZE(kp.kp_free_stack); i++) {
618 if (!kp.kp_free_stack[i])
620 pr_info(" %pS\n", kp.kp_free_stack[i]);
624 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 s->useroffset = useroffset;
648 s->usersize = usersize;
650 err = __kmem_cache_create(s, flags);
653 panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n",
656 s->refcount = -1; /* Exempt from merging for now */
659 struct kmem_cache *__init create_kmalloc_cache(const char *name,
660 unsigned int size, slab_flags_t flags,
661 unsigned int useroffset, unsigned int usersize)
663 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
666 panic("Out of memory when creating slab %s\n", name);
668 create_boot_cache(s, name, size, flags, useroffset, usersize);
669 kasan_cache_create_kmalloc(s);
670 list_add(&s->list, &slab_caches);
676 kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1] __ro_after_init =
677 { /* initialization for https://bugs.llvm.org/show_bug.cgi?id=42570 */ };
678 EXPORT_SYMBOL(kmalloc_caches);
681 * Conversion table for small slabs sizes / 8 to the index in the
682 * kmalloc array. This is necessary for slabs < 192 since we have non power
683 * of two cache sizes there. The size of larger slabs can be determined using
686 static u8 size_index[24] __ro_after_init = {
713 static inline unsigned int size_index_elem(unsigned int bytes)
715 return (bytes - 1) / 8;
719 * Find the kmem_cache structure that serves a given size of
722 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
728 return ZERO_SIZE_PTR;
730 index = size_index[size_index_elem(size)];
732 if (WARN_ON_ONCE(size > KMALLOC_MAX_CACHE_SIZE))
734 index = fls(size - 1);
737 return kmalloc_caches[kmalloc_type(flags)][index];
740 #ifdef CONFIG_ZONE_DMA
741 #define KMALLOC_DMA_NAME(sz) .name[KMALLOC_DMA] = "dma-kmalloc-" #sz,
743 #define KMALLOC_DMA_NAME(sz)
746 #ifdef CONFIG_MEMCG_KMEM
747 #define KMALLOC_CGROUP_NAME(sz) .name[KMALLOC_CGROUP] = "kmalloc-cg-" #sz,
749 #define KMALLOC_CGROUP_NAME(sz)
752 #define INIT_KMALLOC_INFO(__size, __short_size) \
754 .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \
755 .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #__short_size, \
756 KMALLOC_CGROUP_NAME(__short_size) \
757 KMALLOC_DMA_NAME(__short_size) \
762 * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
763 * kmalloc_index() supports up to 2^25=32MB, so the final entry of the table is
766 const struct kmalloc_info_struct kmalloc_info[] __initconst = {
767 INIT_KMALLOC_INFO(0, 0),
768 INIT_KMALLOC_INFO(96, 96),
769 INIT_KMALLOC_INFO(192, 192),
770 INIT_KMALLOC_INFO(8, 8),
771 INIT_KMALLOC_INFO(16, 16),
772 INIT_KMALLOC_INFO(32, 32),
773 INIT_KMALLOC_INFO(64, 64),
774 INIT_KMALLOC_INFO(128, 128),
775 INIT_KMALLOC_INFO(256, 256),
776 INIT_KMALLOC_INFO(512, 512),
777 INIT_KMALLOC_INFO(1024, 1k),
778 INIT_KMALLOC_INFO(2048, 2k),
779 INIT_KMALLOC_INFO(4096, 4k),
780 INIT_KMALLOC_INFO(8192, 8k),
781 INIT_KMALLOC_INFO(16384, 16k),
782 INIT_KMALLOC_INFO(32768, 32k),
783 INIT_KMALLOC_INFO(65536, 64k),
784 INIT_KMALLOC_INFO(131072, 128k),
785 INIT_KMALLOC_INFO(262144, 256k),
786 INIT_KMALLOC_INFO(524288, 512k),
787 INIT_KMALLOC_INFO(1048576, 1M),
788 INIT_KMALLOC_INFO(2097152, 2M),
789 INIT_KMALLOC_INFO(4194304, 4M),
790 INIT_KMALLOC_INFO(8388608, 8M),
791 INIT_KMALLOC_INFO(16777216, 16M),
792 INIT_KMALLOC_INFO(33554432, 32M)
796 * Patch up the size_index table if we have strange large alignment
797 * requirements for the kmalloc array. This is only the case for
798 * MIPS it seems. The standard arches will not generate any code here.
800 * Largest permitted alignment is 256 bytes due to the way we
801 * handle the index determination for the smaller caches.
803 * Make sure that nothing crazy happens if someone starts tinkering
804 * around with ARCH_KMALLOC_MINALIGN
806 void __init setup_kmalloc_cache_index_table(void)
810 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
811 !is_power_of_2(KMALLOC_MIN_SIZE));
813 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
814 unsigned int elem = size_index_elem(i);
816 if (elem >= ARRAY_SIZE(size_index))
818 size_index[elem] = KMALLOC_SHIFT_LOW;
821 if (KMALLOC_MIN_SIZE >= 64) {
823 * The 96 byte sized cache is not used if the alignment
826 for (i = 64 + 8; i <= 96; i += 8)
827 size_index[size_index_elem(i)] = 7;
831 if (KMALLOC_MIN_SIZE >= 128) {
833 * The 192 byte sized cache is not used if the alignment
834 * is 128 byte. Redirect kmalloc to use the 256 byte cache
837 for (i = 128 + 8; i <= 192; i += 8)
838 size_index[size_index_elem(i)] = 8;
843 new_kmalloc_cache(int idx, enum kmalloc_cache_type type, slab_flags_t flags)
845 if (type == KMALLOC_RECLAIM) {
846 flags |= SLAB_RECLAIM_ACCOUNT;
847 } else if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_CGROUP)) {
848 if (mem_cgroup_kmem_disabled()) {
849 kmalloc_caches[type][idx] = kmalloc_caches[KMALLOC_NORMAL][idx];
852 flags |= SLAB_ACCOUNT;
853 } else if (IS_ENABLED(CONFIG_ZONE_DMA) && (type == KMALLOC_DMA)) {
854 flags |= SLAB_CACHE_DMA;
857 kmalloc_caches[type][idx] = create_kmalloc_cache(
858 kmalloc_info[idx].name[type],
859 kmalloc_info[idx].size, flags, 0,
860 kmalloc_info[idx].size);
863 * If CONFIG_MEMCG_KMEM is enabled, disable cache merging for
864 * KMALLOC_NORMAL caches.
866 if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_NORMAL))
867 kmalloc_caches[type][idx]->refcount = -1;
871 * Create the kmalloc array. Some of the regular kmalloc arrays
872 * may already have been created because they were needed to
873 * enable allocations for slab creation.
875 void __init create_kmalloc_caches(slab_flags_t flags)
878 enum kmalloc_cache_type type;
881 * Including KMALLOC_CGROUP if CONFIG_MEMCG_KMEM defined
883 for (type = KMALLOC_NORMAL; type < NR_KMALLOC_TYPES; type++) {
884 for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
885 if (!kmalloc_caches[type][i])
886 new_kmalloc_cache(i, type, flags);
889 * Caches that are not of the two-to-the-power-of size.
890 * These have to be created immediately after the
891 * earlier power of two caches
893 if (KMALLOC_MIN_SIZE <= 32 && i == 6 &&
894 !kmalloc_caches[type][1])
895 new_kmalloc_cache(1, type, flags);
896 if (KMALLOC_MIN_SIZE <= 64 && i == 7 &&
897 !kmalloc_caches[type][2])
898 new_kmalloc_cache(2, type, flags);
902 /* Kmalloc array is now usable */
905 #endif /* !CONFIG_SLOB */
907 gfp_t kmalloc_fix_flags(gfp_t flags)
909 gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
911 flags &= ~GFP_SLAB_BUG_MASK;
912 pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
913 invalid_mask, &invalid_mask, flags, &flags);
920 * To avoid unnecessary overhead, we pass through large allocation requests
921 * directly to the page allocator. We use __GFP_COMP, because we will need to
922 * know the allocation order to free the pages properly in kfree.
924 void *kmalloc_order(size_t size, gfp_t flags, unsigned int order)
929 if (unlikely(flags & GFP_SLAB_BUG_MASK))
930 flags = kmalloc_fix_flags(flags);
933 page = alloc_pages(flags, order);
935 ret = page_address(page);
936 mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B,
939 ret = kasan_kmalloc_large(ret, size, flags);
940 /* As ret might get tagged, call kmemleak hook after KASAN. */
941 kmemleak_alloc(ret, size, 1, flags);
944 EXPORT_SYMBOL(kmalloc_order);
946 #ifdef CONFIG_TRACING
947 void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
949 void *ret = kmalloc_order(size, flags, order);
950 trace_kmalloc(_RET_IP_, ret, NULL, size, PAGE_SIZE << order, flags);
953 EXPORT_SYMBOL(kmalloc_order_trace);
956 #ifdef CONFIG_SLAB_FREELIST_RANDOM
957 /* Randomize a generic freelist */
958 static void freelist_randomize(struct rnd_state *state, unsigned int *list,
964 for (i = 0; i < count; i++)
967 /* Fisher-Yates shuffle */
968 for (i = count - 1; i > 0; i--) {
969 rand = prandom_u32_state(state);
971 swap(list[i], list[rand]);
975 /* Create a random sequence per cache */
976 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
979 struct rnd_state state;
981 if (count < 2 || cachep->random_seq)
984 cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp);
985 if (!cachep->random_seq)
988 /* Get best entropy at this stage of boot */
989 prandom_seed_state(&state, get_random_long());
991 freelist_randomize(&state, cachep->random_seq, count);
995 /* Destroy the per-cache random freelist sequence */
996 void cache_random_seq_destroy(struct kmem_cache *cachep)
998 kfree(cachep->random_seq);
999 cachep->random_seq = NULL;
1001 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
1003 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
1005 #define SLABINFO_RIGHTS (0600)
1007 #define SLABINFO_RIGHTS (0400)
1010 static void print_slabinfo_header(struct seq_file *m)
1013 * Output format version, so at least we can change it
1014 * without _too_ many complaints.
1016 #ifdef CONFIG_DEBUG_SLAB
1017 seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
1019 seq_puts(m, "slabinfo - version: 2.1\n");
1021 seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
1022 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
1023 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
1024 #ifdef CONFIG_DEBUG_SLAB
1025 seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
1026 seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
1031 static void *slab_start(struct seq_file *m, loff_t *pos)
1033 mutex_lock(&slab_mutex);
1034 return seq_list_start(&slab_caches, *pos);
1037 static void *slab_next(struct seq_file *m, void *p, loff_t *pos)
1039 return seq_list_next(p, &slab_caches, pos);
1042 static void slab_stop(struct seq_file *m, void *p)
1044 mutex_unlock(&slab_mutex);
1047 static void cache_show(struct kmem_cache *s, struct seq_file *m)
1049 struct slabinfo sinfo;
1051 memset(&sinfo, 0, sizeof(sinfo));
1052 get_slabinfo(s, &sinfo);
1054 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
1055 s->name, sinfo.active_objs, sinfo.num_objs, s->size,
1056 sinfo.objects_per_slab, (1 << sinfo.cache_order));
1058 seq_printf(m, " : tunables %4u %4u %4u",
1059 sinfo.limit, sinfo.batchcount, sinfo.shared);
1060 seq_printf(m, " : slabdata %6lu %6lu %6lu",
1061 sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
1062 slabinfo_show_stats(m, s);
1066 static int slab_show(struct seq_file *m, void *p)
1068 struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
1070 if (p == slab_caches.next)
1071 print_slabinfo_header(m);
1076 void dump_unreclaimable_slab(void)
1078 struct kmem_cache *s;
1079 struct slabinfo sinfo;
1082 * Here acquiring slab_mutex is risky since we don't prefer to get
1083 * sleep in oom path. But, without mutex hold, it may introduce a
1085 * Use mutex_trylock to protect the list traverse, dump nothing
1086 * without acquiring the mutex.
1088 if (!mutex_trylock(&slab_mutex)) {
1089 pr_warn("excessive unreclaimable slab but cannot dump stats\n");
1093 pr_info("Unreclaimable slab info:\n");
1094 pr_info("Name Used Total\n");
1096 list_for_each_entry(s, &slab_caches, list) {
1097 if (s->flags & SLAB_RECLAIM_ACCOUNT)
1100 get_slabinfo(s, &sinfo);
1102 if (sinfo.num_objs > 0)
1103 pr_info("%-17s %10luKB %10luKB\n", s->name,
1104 (sinfo.active_objs * s->size) / 1024,
1105 (sinfo.num_objs * s->size) / 1024);
1107 mutex_unlock(&slab_mutex);
1111 * slabinfo_op - iterator that generates /proc/slabinfo
1120 * num-pages-per-slab
1121 * + further values on SMP and with statistics enabled
1123 static const struct seq_operations slabinfo_op = {
1124 .start = slab_start,
1130 static int slabinfo_open(struct inode *inode, struct file *file)
1132 return seq_open(file, &slabinfo_op);
1135 static const struct proc_ops slabinfo_proc_ops = {
1136 .proc_flags = PROC_ENTRY_PERMANENT,
1137 .proc_open = slabinfo_open,
1138 .proc_read = seq_read,
1139 .proc_write = slabinfo_write,
1140 .proc_lseek = seq_lseek,
1141 .proc_release = seq_release,
1144 static int __init slab_proc_init(void)
1146 proc_create("slabinfo", SLABINFO_RIGHTS, NULL, &slabinfo_proc_ops);
1149 module_init(slab_proc_init);
1151 #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */
1153 static __always_inline void *__do_krealloc(const void *p, size_t new_size,
1159 /* Don't use instrumented ksize to allow precise KASAN poisoning. */
1160 if (likely(!ZERO_OR_NULL_PTR(p))) {
1161 if (!kasan_check_byte(p))
1163 ks = kfence_ksize(p) ?: __ksize(p);
1167 /* If the object still fits, repoison it precisely. */
1168 if (ks >= new_size) {
1169 p = kasan_krealloc((void *)p, new_size, flags);
1173 ret = kmalloc_track_caller(new_size, flags);
1175 /* Disable KASAN checks as the object's redzone is accessed. */
1176 kasan_disable_current();
1177 memcpy(ret, kasan_reset_tag(p), ks);
1178 kasan_enable_current();
1185 * krealloc - reallocate memory. The contents will remain unchanged.
1186 * @p: object to reallocate memory for.
1187 * @new_size: how many bytes of memory are required.
1188 * @flags: the type of memory to allocate.
1190 * The contents of the object pointed to are preserved up to the
1191 * lesser of the new and old sizes (__GFP_ZERO flag is effectively ignored).
1192 * If @p is %NULL, krealloc() behaves exactly like kmalloc(). If @new_size
1193 * is 0 and @p is not a %NULL pointer, the object pointed to is freed.
1195 * Return: pointer to the allocated memory or %NULL in case of error
1197 void *krealloc(const void *p, size_t new_size, gfp_t flags)
1201 if (unlikely(!new_size)) {
1203 return ZERO_SIZE_PTR;
1206 ret = __do_krealloc(p, new_size, flags);
1207 if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret))
1212 EXPORT_SYMBOL(krealloc);
1215 * kfree_sensitive - Clear sensitive information in memory before freeing
1216 * @p: object to free memory of
1218 * The memory of the object @p points to is zeroed before freed.
1219 * If @p is %NULL, kfree_sensitive() does nothing.
1221 * Note: this function zeroes the whole allocated buffer which can be a good
1222 * deal bigger than the requested buffer size passed to kmalloc(). So be
1223 * careful when using this function in performance sensitive code.
1225 void kfree_sensitive(const void *p)
1228 void *mem = (void *)p;
1232 memzero_explicit(mem, ks);
1235 EXPORT_SYMBOL(kfree_sensitive);
1238 * ksize - get the actual amount of memory allocated for a given object
1239 * @objp: Pointer to the object
1241 * kmalloc may internally round up allocations and return more memory
1242 * than requested. ksize() can be used to determine the actual amount of
1243 * memory allocated. The caller may use this additional memory, even though
1244 * a smaller amount of memory was initially specified with the kmalloc call.
1245 * The caller must guarantee that objp points to a valid object previously
1246 * allocated with either kmalloc() or kmem_cache_alloc(). The object
1247 * must not be freed during the duration of the call.
1249 * Return: size of the actual memory used by @objp in bytes
1251 size_t ksize(const void *objp)
1256 * We need to first check that the pointer to the object is valid, and
1257 * only then unpoison the memory. The report printed from ksize() is
1258 * more useful, then when it's printed later when the behaviour could
1259 * be undefined due to a potential use-after-free or double-free.
1261 * We use kasan_check_byte(), which is supported for the hardware
1262 * tag-based KASAN mode, unlike kasan_check_read/write().
1264 * If the pointed to memory is invalid, we return 0 to avoid users of
1265 * ksize() writing to and potentially corrupting the memory region.
1267 * We want to perform the check before __ksize(), to avoid potentially
1268 * crashing in __ksize() due to accessing invalid metadata.
1270 if (unlikely(ZERO_OR_NULL_PTR(objp)) || !kasan_check_byte(objp))
1273 size = kfence_ksize(objp) ?: __ksize(objp);
1275 * We assume that ksize callers could use whole allocated area,
1276 * so we need to unpoison this area.
1278 kasan_unpoison_range(objp, size);
1281 EXPORT_SYMBOL(ksize);
1283 /* Tracepoints definitions. */
1284 EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1285 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1286 EXPORT_TRACEPOINT_SYMBOL(kmalloc_node);
1287 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node);
1288 EXPORT_TRACEPOINT_SYMBOL(kfree);
1289 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);
1291 int should_failslab(struct kmem_cache *s, gfp_t gfpflags)
1293 if (__should_failslab(s, gfpflags))
1297 ALLOW_ERROR_INJECTION(should_failslab, ERRNO);