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 <asm/cacheflush.h>
21 #include <asm/tlbflush.h>
23 #include <linux/memcontrol.h>
25 #define CREATE_TRACE_POINTS
26 #include <trace/events/kmem.h>
30 enum slab_state slab_state;
31 LIST_HEAD(slab_caches);
32 DEFINE_MUTEX(slab_mutex);
33 struct kmem_cache *kmem_cache;
35 #ifdef CONFIG_HARDENED_USERCOPY
36 bool usercopy_fallback __ro_after_init =
37 IS_ENABLED(CONFIG_HARDENED_USERCOPY_FALLBACK);
38 module_param(usercopy_fallback, bool, 0400);
39 MODULE_PARM_DESC(usercopy_fallback,
40 "WARN instead of reject usercopy whitelist violations");
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_KASAN)
55 #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \
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)
70 __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0);
73 __setup("slab_nomerge", setup_slab_nomerge);
76 * Determine the size of a slab object
78 unsigned int kmem_cache_size(struct kmem_cache *s)
80 return s->object_size;
82 EXPORT_SYMBOL(kmem_cache_size);
84 #ifdef CONFIG_DEBUG_VM
85 static int kmem_cache_sanity_check(const char *name, unsigned int size)
87 if (!name || in_interrupt() || size < sizeof(void *) ||
88 size > KMALLOC_MAX_SIZE) {
89 pr_err("kmem_cache_create(%s) integrity check failed\n", name);
93 WARN_ON(strchr(name, ' ')); /* It confuses parsers */
97 static inline int kmem_cache_sanity_check(const char *name, unsigned int size)
103 void __kmem_cache_free_bulk(struct kmem_cache *s, size_t nr, void **p)
107 for (i = 0; i < nr; i++) {
109 kmem_cache_free(s, p[i]);
115 int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t nr,
120 for (i = 0; i < nr; i++) {
121 void *x = p[i] = kmem_cache_alloc(s, flags);
123 __kmem_cache_free_bulk(s, i, p);
130 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
132 LIST_HEAD(slab_root_caches);
134 void slab_init_memcg_params(struct kmem_cache *s)
136 s->memcg_params.root_cache = NULL;
137 RCU_INIT_POINTER(s->memcg_params.memcg_caches, NULL);
138 INIT_LIST_HEAD(&s->memcg_params.children);
139 s->memcg_params.dying = false;
142 static int init_memcg_params(struct kmem_cache *s,
143 struct mem_cgroup *memcg, struct kmem_cache *root_cache)
145 struct memcg_cache_array *arr;
148 s->memcg_params.root_cache = root_cache;
149 s->memcg_params.memcg = memcg;
150 INIT_LIST_HEAD(&s->memcg_params.children_node);
151 INIT_LIST_HEAD(&s->memcg_params.kmem_caches_node);
155 slab_init_memcg_params(s);
157 if (!memcg_nr_cache_ids)
160 arr = kvzalloc(sizeof(struct memcg_cache_array) +
161 memcg_nr_cache_ids * sizeof(void *),
166 RCU_INIT_POINTER(s->memcg_params.memcg_caches, arr);
170 static void destroy_memcg_params(struct kmem_cache *s)
172 if (is_root_cache(s))
173 kvfree(rcu_access_pointer(s->memcg_params.memcg_caches));
176 static void free_memcg_params(struct rcu_head *rcu)
178 struct memcg_cache_array *old;
180 old = container_of(rcu, struct memcg_cache_array, rcu);
184 static int update_memcg_params(struct kmem_cache *s, int new_array_size)
186 struct memcg_cache_array *old, *new;
188 new = kvzalloc(sizeof(struct memcg_cache_array) +
189 new_array_size * sizeof(void *), GFP_KERNEL);
193 old = rcu_dereference_protected(s->memcg_params.memcg_caches,
194 lockdep_is_held(&slab_mutex));
196 memcpy(new->entries, old->entries,
197 memcg_nr_cache_ids * sizeof(void *));
199 rcu_assign_pointer(s->memcg_params.memcg_caches, new);
201 call_rcu(&old->rcu, free_memcg_params);
205 int memcg_update_all_caches(int num_memcgs)
207 struct kmem_cache *s;
210 mutex_lock(&slab_mutex);
211 list_for_each_entry(s, &slab_root_caches, root_caches_node) {
212 ret = update_memcg_params(s, num_memcgs);
214 * Instead of freeing the memory, we'll just leave the caches
215 * up to this point in an updated state.
220 mutex_unlock(&slab_mutex);
224 void memcg_link_cache(struct kmem_cache *s)
226 if (is_root_cache(s)) {
227 list_add(&s->root_caches_node, &slab_root_caches);
229 list_add(&s->memcg_params.children_node,
230 &s->memcg_params.root_cache->memcg_params.children);
231 list_add(&s->memcg_params.kmem_caches_node,
232 &s->memcg_params.memcg->kmem_caches);
236 static void memcg_unlink_cache(struct kmem_cache *s)
238 if (is_root_cache(s)) {
239 list_del(&s->root_caches_node);
241 list_del(&s->memcg_params.children_node);
242 list_del(&s->memcg_params.kmem_caches_node);
246 static inline int init_memcg_params(struct kmem_cache *s,
247 struct mem_cgroup *memcg, struct kmem_cache *root_cache)
252 static inline void destroy_memcg_params(struct kmem_cache *s)
256 static inline void memcg_unlink_cache(struct kmem_cache *s)
259 #endif /* CONFIG_MEMCG && !CONFIG_SLOB */
262 * Figure out what the alignment of the objects will be given a set of
263 * flags, a user specified alignment and the size of the objects.
265 static unsigned int calculate_alignment(slab_flags_t flags,
266 unsigned int align, unsigned int size)
269 * If the user wants hardware cache aligned objects then follow that
270 * suggestion if the object is sufficiently large.
272 * The hardware cache alignment cannot override the specified
273 * alignment though. If that is greater then use it.
275 if (flags & SLAB_HWCACHE_ALIGN) {
278 ralign = cache_line_size();
279 while (size <= ralign / 2)
281 align = max(align, ralign);
284 if (align < ARCH_SLAB_MINALIGN)
285 align = ARCH_SLAB_MINALIGN;
287 return ALIGN(align, sizeof(void *));
291 * Find a mergeable slab cache
293 int slab_unmergeable(struct kmem_cache *s)
295 if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
298 if (!is_root_cache(s))
308 * We may have set a slab to be unmergeable during bootstrap.
316 struct kmem_cache *find_mergeable(unsigned int size, unsigned int align,
317 slab_flags_t flags, const char *name, void (*ctor)(void *))
319 struct kmem_cache *s;
327 size = ALIGN(size, sizeof(void *));
328 align = calculate_alignment(flags, align, size);
329 size = ALIGN(size, align);
330 flags = kmem_cache_flags(size, flags, name, NULL);
332 if (flags & SLAB_NEVER_MERGE)
335 list_for_each_entry_reverse(s, &slab_root_caches, root_caches_node) {
336 if (slab_unmergeable(s))
342 if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
345 * Check if alignment is compatible.
346 * Courtesy of Adrian Drzewiecki
348 if ((s->size & ~(align - 1)) != s->size)
351 if (s->size - size >= sizeof(void *))
354 if (IS_ENABLED(CONFIG_SLAB) && align &&
355 (align > s->align || s->align % align))
363 static struct kmem_cache *create_cache(const char *name,
364 unsigned int object_size, unsigned int align,
365 slab_flags_t flags, unsigned int useroffset,
366 unsigned int usersize, void (*ctor)(void *),
367 struct mem_cgroup *memcg, struct kmem_cache *root_cache)
369 struct kmem_cache *s;
372 if (WARN_ON(useroffset + usersize > object_size))
373 useroffset = usersize = 0;
376 s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
381 s->size = s->object_size = object_size;
384 s->useroffset = useroffset;
385 s->usersize = usersize;
387 err = init_memcg_params(s, memcg, root_cache);
391 err = __kmem_cache_create(s, flags);
396 list_add(&s->list, &slab_caches);
404 destroy_memcg_params(s);
405 kmem_cache_free(kmem_cache, s);
410 * kmem_cache_create_usercopy - Create a cache.
411 * @name: A string which is used in /proc/slabinfo to identify this cache.
412 * @size: The size of objects to be created in this cache.
413 * @align: The required alignment for the objects.
415 * @useroffset: Usercopy region offset
416 * @usersize: Usercopy region size
417 * @ctor: A constructor for the objects.
419 * Returns a ptr to the cache on success, NULL on failure.
420 * Cannot be called within a interrupt, but can be interrupted.
421 * The @ctor is run when new pages are allocated by the cache.
425 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
426 * to catch references to uninitialised memory.
428 * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check
429 * for buffer overruns.
431 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
432 * cacheline. This can be beneficial if you're counting cycles as closely
436 kmem_cache_create_usercopy(const char *name,
437 unsigned int size, unsigned int align,
439 unsigned int useroffset, unsigned int usersize,
440 void (*ctor)(void *))
442 struct kmem_cache *s = NULL;
443 const char *cache_name;
448 memcg_get_cache_ids();
450 mutex_lock(&slab_mutex);
452 err = kmem_cache_sanity_check(name, size);
457 /* Refuse requests with allocator specific flags */
458 if (flags & ~SLAB_FLAGS_PERMITTED) {
464 * Some allocators will constraint the set of valid flags to a subset
465 * of all flags. We expect them to define CACHE_CREATE_MASK in this
466 * case, and we'll just provide them with a sanitized version of the
469 flags &= CACHE_CREATE_MASK;
471 /* Fail closed on bad usersize of useroffset values. */
472 if (WARN_ON(!usersize && useroffset) ||
473 WARN_ON(size < usersize || size - usersize < useroffset))
474 usersize = useroffset = 0;
477 s = __kmem_cache_alias(name, size, align, flags, ctor);
481 cache_name = kstrdup_const(name, GFP_KERNEL);
487 s = create_cache(cache_name, size,
488 calculate_alignment(flags, align, size),
489 flags, useroffset, usersize, ctor, NULL, NULL);
492 kfree_const(cache_name);
496 mutex_unlock(&slab_mutex);
498 memcg_put_cache_ids();
503 if (flags & SLAB_PANIC)
504 panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n",
507 pr_warn("kmem_cache_create(%s) failed with error %d\n",
515 EXPORT_SYMBOL(kmem_cache_create_usercopy);
518 kmem_cache_create(const char *name, unsigned int size, unsigned int align,
519 slab_flags_t flags, void (*ctor)(void *))
521 return kmem_cache_create_usercopy(name, size, align, flags, 0, 0,
524 EXPORT_SYMBOL(kmem_cache_create);
526 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work)
528 LIST_HEAD(to_destroy);
529 struct kmem_cache *s, *s2;
532 * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the
533 * @slab_caches_to_rcu_destroy list. The slab pages are freed
534 * through RCU and and the associated kmem_cache are dereferenced
535 * while freeing the pages, so the kmem_caches should be freed only
536 * after the pending RCU operations are finished. As rcu_barrier()
537 * is a pretty slow operation, we batch all pending destructions
540 mutex_lock(&slab_mutex);
541 list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy);
542 mutex_unlock(&slab_mutex);
544 if (list_empty(&to_destroy))
549 list_for_each_entry_safe(s, s2, &to_destroy, list) {
550 #ifdef SLAB_SUPPORTS_SYSFS
551 sysfs_slab_release(s);
553 slab_kmem_cache_release(s);
558 static int shutdown_cache(struct kmem_cache *s)
560 /* free asan quarantined objects */
561 kasan_cache_shutdown(s);
563 if (__kmem_cache_shutdown(s) != 0)
566 memcg_unlink_cache(s);
569 if (s->flags & SLAB_TYPESAFE_BY_RCU) {
570 list_add_tail(&s->list, &slab_caches_to_rcu_destroy);
571 schedule_work(&slab_caches_to_rcu_destroy_work);
573 #ifdef SLAB_SUPPORTS_SYSFS
574 sysfs_slab_release(s);
576 slab_kmem_cache_release(s);
583 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
585 * memcg_create_kmem_cache - Create a cache for a memory cgroup.
586 * @memcg: The memory cgroup the new cache is for.
587 * @root_cache: The parent of the new cache.
589 * This function attempts to create a kmem cache that will serve allocation
590 * requests going from @memcg to @root_cache. The new cache inherits properties
593 void memcg_create_kmem_cache(struct mem_cgroup *memcg,
594 struct kmem_cache *root_cache)
596 static char memcg_name_buf[NAME_MAX + 1]; /* protected by slab_mutex */
597 struct cgroup_subsys_state *css = &memcg->css;
598 struct memcg_cache_array *arr;
599 struct kmem_cache *s = NULL;
606 mutex_lock(&slab_mutex);
609 * The memory cgroup could have been offlined while the cache
610 * creation work was pending.
612 if (memcg->kmem_state != KMEM_ONLINE || root_cache->memcg_params.dying)
615 idx = memcg_cache_id(memcg);
616 arr = rcu_dereference_protected(root_cache->memcg_params.memcg_caches,
617 lockdep_is_held(&slab_mutex));
620 * Since per-memcg caches are created asynchronously on first
621 * allocation (see memcg_kmem_get_cache()), several threads can try to
622 * create the same cache, but only one of them may succeed.
624 if (arr->entries[idx])
627 cgroup_name(css->cgroup, memcg_name_buf, sizeof(memcg_name_buf));
628 cache_name = kasprintf(GFP_KERNEL, "%s(%llu:%s)", root_cache->name,
629 css->serial_nr, memcg_name_buf);
633 s = create_cache(cache_name, root_cache->object_size,
635 root_cache->flags & CACHE_CREATE_MASK,
636 root_cache->useroffset, root_cache->usersize,
637 root_cache->ctor, memcg, root_cache);
639 * If we could not create a memcg cache, do not complain, because
640 * that's not critical at all as we can always proceed with the root
649 * Since readers won't lock (see cache_from_memcg_idx()), we need a
650 * barrier here to ensure nobody will see the kmem_cache partially
654 arr->entries[idx] = s;
657 mutex_unlock(&slab_mutex);
663 static void kmemcg_deactivate_workfn(struct work_struct *work)
665 struct kmem_cache *s = container_of(work, struct kmem_cache,
666 memcg_params.deact_work);
671 mutex_lock(&slab_mutex);
673 s->memcg_params.deact_fn(s);
675 mutex_unlock(&slab_mutex);
680 /* done, put the ref from slab_deactivate_memcg_cache_rcu_sched() */
681 css_put(&s->memcg_params.memcg->css);
684 static void kmemcg_deactivate_rcufn(struct rcu_head *head)
686 struct kmem_cache *s = container_of(head, struct kmem_cache,
687 memcg_params.deact_rcu_head);
690 * We need to grab blocking locks. Bounce to ->deact_work. The
691 * work item shares the space with the RCU head and can't be
692 * initialized eariler.
694 INIT_WORK(&s->memcg_params.deact_work, kmemcg_deactivate_workfn);
695 queue_work(memcg_kmem_cache_wq, &s->memcg_params.deact_work);
699 * slab_deactivate_memcg_cache_rcu_sched - schedule deactivation after a
700 * sched RCU grace period
701 * @s: target kmem_cache
702 * @deact_fn: deactivation function to call
704 * Schedule @deact_fn to be invoked with online cpus, mems and slab_mutex
705 * held after a sched RCU grace period. The slab is guaranteed to stay
706 * alive until @deact_fn is finished. This is to be used from
707 * __kmemcg_cache_deactivate().
709 void slab_deactivate_memcg_cache_rcu_sched(struct kmem_cache *s,
710 void (*deact_fn)(struct kmem_cache *))
712 if (WARN_ON_ONCE(is_root_cache(s)) ||
713 WARN_ON_ONCE(s->memcg_params.deact_fn))
716 if (s->memcg_params.root_cache->memcg_params.dying)
719 /* pin memcg so that @s doesn't get destroyed in the middle */
720 css_get(&s->memcg_params.memcg->css);
722 s->memcg_params.deact_fn = deact_fn;
723 call_rcu_sched(&s->memcg_params.deact_rcu_head, kmemcg_deactivate_rcufn);
726 void memcg_deactivate_kmem_caches(struct mem_cgroup *memcg)
729 struct memcg_cache_array *arr;
730 struct kmem_cache *s, *c;
732 idx = memcg_cache_id(memcg);
737 mutex_lock(&slab_mutex);
738 list_for_each_entry(s, &slab_root_caches, root_caches_node) {
739 arr = rcu_dereference_protected(s->memcg_params.memcg_caches,
740 lockdep_is_held(&slab_mutex));
741 c = arr->entries[idx];
745 __kmemcg_cache_deactivate(c);
746 arr->entries[idx] = NULL;
748 mutex_unlock(&slab_mutex);
754 void memcg_destroy_kmem_caches(struct mem_cgroup *memcg)
756 struct kmem_cache *s, *s2;
761 mutex_lock(&slab_mutex);
762 list_for_each_entry_safe(s, s2, &memcg->kmem_caches,
763 memcg_params.kmem_caches_node) {
765 * The cgroup is about to be freed and therefore has no charges
766 * left. Hence, all its caches must be empty by now.
768 BUG_ON(shutdown_cache(s));
770 mutex_unlock(&slab_mutex);
776 static int shutdown_memcg_caches(struct kmem_cache *s)
778 struct memcg_cache_array *arr;
779 struct kmem_cache *c, *c2;
783 BUG_ON(!is_root_cache(s));
786 * First, shutdown active caches, i.e. caches that belong to online
789 arr = rcu_dereference_protected(s->memcg_params.memcg_caches,
790 lockdep_is_held(&slab_mutex));
791 for_each_memcg_cache_index(i) {
795 if (shutdown_cache(c))
797 * The cache still has objects. Move it to a temporary
798 * list so as not to try to destroy it for a second
799 * time while iterating over inactive caches below.
801 list_move(&c->memcg_params.children_node, &busy);
804 * The cache is empty and will be destroyed soon. Clear
805 * the pointer to it in the memcg_caches array so that
806 * it will never be accessed even if the root cache
809 arr->entries[i] = NULL;
813 * Second, shutdown all caches left from memory cgroups that are now
816 list_for_each_entry_safe(c, c2, &s->memcg_params.children,
817 memcg_params.children_node)
820 list_splice(&busy, &s->memcg_params.children);
823 * A cache being destroyed must be empty. In particular, this means
824 * that all per memcg caches attached to it must be empty too.
826 if (!list_empty(&s->memcg_params.children))
831 static void flush_memcg_workqueue(struct kmem_cache *s)
833 mutex_lock(&slab_mutex);
834 s->memcg_params.dying = true;
835 mutex_unlock(&slab_mutex);
838 * SLUB deactivates the kmem_caches through call_rcu_sched. Make
839 * sure all registered rcu callbacks have been invoked.
841 if (IS_ENABLED(CONFIG_SLUB))
845 * SLAB and SLUB create memcg kmem_caches through workqueue and SLUB
846 * deactivates the memcg kmem_caches through workqueue. Make sure all
847 * previous workitems on workqueue are processed.
849 flush_workqueue(memcg_kmem_cache_wq);
852 static inline int shutdown_memcg_caches(struct kmem_cache *s)
857 static inline void flush_memcg_workqueue(struct kmem_cache *s)
860 #endif /* CONFIG_MEMCG && !CONFIG_SLOB */
862 void slab_kmem_cache_release(struct kmem_cache *s)
864 __kmem_cache_release(s);
865 destroy_memcg_params(s);
866 kfree_const(s->name);
867 kmem_cache_free(kmem_cache, s);
870 void kmem_cache_destroy(struct kmem_cache *s)
877 flush_memcg_workqueue(s);
882 mutex_lock(&slab_mutex);
888 err = shutdown_memcg_caches(s);
890 err = shutdown_cache(s);
893 pr_err("kmem_cache_destroy %s: Slab cache still has objects\n",
898 mutex_unlock(&slab_mutex);
903 EXPORT_SYMBOL(kmem_cache_destroy);
906 * kmem_cache_shrink - Shrink a cache.
907 * @cachep: The cache to shrink.
909 * Releases as many slabs as possible for a cache.
910 * To help debugging, a zero exit status indicates all slabs were released.
912 int kmem_cache_shrink(struct kmem_cache *cachep)
918 kasan_cache_shrink(cachep);
919 ret = __kmem_cache_shrink(cachep);
924 EXPORT_SYMBOL(kmem_cache_shrink);
926 bool slab_is_available(void)
928 return slab_state >= UP;
932 /* Create a cache during boot when no slab services are available yet */
933 void __init create_boot_cache(struct kmem_cache *s, const char *name,
934 unsigned int size, slab_flags_t flags,
935 unsigned int useroffset, unsigned int usersize)
940 s->size = s->object_size = size;
941 s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size);
942 s->useroffset = useroffset;
943 s->usersize = usersize;
945 slab_init_memcg_params(s);
947 err = __kmem_cache_create(s, flags);
950 panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n",
953 s->refcount = -1; /* Exempt from merging for now */
956 struct kmem_cache *__init create_kmalloc_cache(const char *name,
957 unsigned int size, slab_flags_t flags,
958 unsigned int useroffset, unsigned int usersize)
960 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
963 panic("Out of memory when creating slab %s\n", name);
965 create_boot_cache(s, name, size, flags, useroffset, usersize);
966 list_add(&s->list, &slab_caches);
972 struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1] __ro_after_init;
973 EXPORT_SYMBOL(kmalloc_caches);
975 #ifdef CONFIG_ZONE_DMA
976 struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1] __ro_after_init;
977 EXPORT_SYMBOL(kmalloc_dma_caches);
981 * Conversion table for small slabs sizes / 8 to the index in the
982 * kmalloc array. This is necessary for slabs < 192 since we have non power
983 * of two cache sizes there. The size of larger slabs can be determined using
986 static u8 size_index[24] __ro_after_init = {
1013 static inline unsigned int size_index_elem(unsigned int bytes)
1015 return (bytes - 1) / 8;
1019 * Find the kmem_cache structure that serves a given size of
1022 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
1026 if (unlikely(size > KMALLOC_MAX_SIZE)) {
1027 WARN_ON_ONCE(!(flags & __GFP_NOWARN));
1033 return ZERO_SIZE_PTR;
1035 index = size_index[size_index_elem(size)];
1037 index = fls(size - 1);
1039 #ifdef CONFIG_ZONE_DMA
1040 if (unlikely((flags & GFP_DMA)))
1041 return kmalloc_dma_caches[index];
1044 return kmalloc_caches[index];
1048 * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
1049 * kmalloc_index() supports up to 2^26=64MB, so the final entry of the table is
1052 const struct kmalloc_info_struct kmalloc_info[] __initconst = {
1053 {NULL, 0}, {"kmalloc-96", 96},
1054 {"kmalloc-192", 192}, {"kmalloc-8", 8},
1055 {"kmalloc-16", 16}, {"kmalloc-32", 32},
1056 {"kmalloc-64", 64}, {"kmalloc-128", 128},
1057 {"kmalloc-256", 256}, {"kmalloc-512", 512},
1058 {"kmalloc-1024", 1024}, {"kmalloc-2048", 2048},
1059 {"kmalloc-4096", 4096}, {"kmalloc-8192", 8192},
1060 {"kmalloc-16384", 16384}, {"kmalloc-32768", 32768},
1061 {"kmalloc-65536", 65536}, {"kmalloc-131072", 131072},
1062 {"kmalloc-262144", 262144}, {"kmalloc-524288", 524288},
1063 {"kmalloc-1048576", 1048576}, {"kmalloc-2097152", 2097152},
1064 {"kmalloc-4194304", 4194304}, {"kmalloc-8388608", 8388608},
1065 {"kmalloc-16777216", 16777216}, {"kmalloc-33554432", 33554432},
1066 {"kmalloc-67108864", 67108864}
1070 * Patch up the size_index table if we have strange large alignment
1071 * requirements for the kmalloc array. This is only the case for
1072 * MIPS it seems. The standard arches will not generate any code here.
1074 * Largest permitted alignment is 256 bytes due to the way we
1075 * handle the index determination for the smaller caches.
1077 * Make sure that nothing crazy happens if someone starts tinkering
1078 * around with ARCH_KMALLOC_MINALIGN
1080 void __init setup_kmalloc_cache_index_table(void)
1084 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
1085 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
1087 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
1088 unsigned int elem = size_index_elem(i);
1090 if (elem >= ARRAY_SIZE(size_index))
1092 size_index[elem] = KMALLOC_SHIFT_LOW;
1095 if (KMALLOC_MIN_SIZE >= 64) {
1097 * The 96 byte size cache is not used if the alignment
1100 for (i = 64 + 8; i <= 96; i += 8)
1101 size_index[size_index_elem(i)] = 7;
1105 if (KMALLOC_MIN_SIZE >= 128) {
1107 * The 192 byte sized cache is not used if the alignment
1108 * is 128 byte. Redirect kmalloc to use the 256 byte cache
1111 for (i = 128 + 8; i <= 192; i += 8)
1112 size_index[size_index_elem(i)] = 8;
1116 static void __init new_kmalloc_cache(int idx, slab_flags_t flags)
1118 kmalloc_caches[idx] = create_kmalloc_cache(kmalloc_info[idx].name,
1119 kmalloc_info[idx].size, flags, 0,
1120 kmalloc_info[idx].size);
1124 * Create the kmalloc array. Some of the regular kmalloc arrays
1125 * may already have been created because they were needed to
1126 * enable allocations for slab creation.
1128 void __init create_kmalloc_caches(slab_flags_t flags)
1132 for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
1133 if (!kmalloc_caches[i])
1134 new_kmalloc_cache(i, flags);
1137 * Caches that are not of the two-to-the-power-of size.
1138 * These have to be created immediately after the
1139 * earlier power of two caches
1141 if (KMALLOC_MIN_SIZE <= 32 && !kmalloc_caches[1] && i == 6)
1142 new_kmalloc_cache(1, flags);
1143 if (KMALLOC_MIN_SIZE <= 64 && !kmalloc_caches[2] && i == 7)
1144 new_kmalloc_cache(2, flags);
1147 /* Kmalloc array is now usable */
1150 #ifdef CONFIG_ZONE_DMA
1151 for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
1152 struct kmem_cache *s = kmalloc_caches[i];
1155 unsigned int size = kmalloc_size(i);
1156 char *n = kasprintf(GFP_NOWAIT,
1157 "dma-kmalloc-%u", size);
1160 kmalloc_dma_caches[i] = create_kmalloc_cache(n,
1161 size, SLAB_CACHE_DMA | flags, 0, 0);
1166 #endif /* !CONFIG_SLOB */
1169 * To avoid unnecessary overhead, we pass through large allocation requests
1170 * directly to the page allocator. We use __GFP_COMP, because we will need to
1171 * know the allocation order to free the pages properly in kfree.
1173 void *kmalloc_order(size_t size, gfp_t flags, unsigned int order)
1178 flags |= __GFP_COMP;
1179 page = alloc_pages(flags, order);
1180 ret = page ? page_address(page) : NULL;
1181 kmemleak_alloc(ret, size, 1, flags);
1182 kasan_kmalloc_large(ret, size, flags);
1185 EXPORT_SYMBOL(kmalloc_order);
1187 #ifdef CONFIG_TRACING
1188 void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
1190 void *ret = kmalloc_order(size, flags, order);
1191 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
1194 EXPORT_SYMBOL(kmalloc_order_trace);
1197 #ifdef CONFIG_SLAB_FREELIST_RANDOM
1198 /* Randomize a generic freelist */
1199 static void freelist_randomize(struct rnd_state *state, unsigned int *list,
1205 for (i = 0; i < count; i++)
1208 /* Fisher-Yates shuffle */
1209 for (i = count - 1; i > 0; i--) {
1210 rand = prandom_u32_state(state);
1212 swap(list[i], list[rand]);
1216 /* Create a random sequence per cache */
1217 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
1220 struct rnd_state state;
1222 if (count < 2 || cachep->random_seq)
1225 cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp);
1226 if (!cachep->random_seq)
1229 /* Get best entropy at this stage of boot */
1230 prandom_seed_state(&state, get_random_long());
1232 freelist_randomize(&state, cachep->random_seq, count);
1236 /* Destroy the per-cache random freelist sequence */
1237 void cache_random_seq_destroy(struct kmem_cache *cachep)
1239 kfree(cachep->random_seq);
1240 cachep->random_seq = NULL;
1242 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
1244 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
1246 #define SLABINFO_RIGHTS (0600)
1248 #define SLABINFO_RIGHTS (0400)
1251 static void print_slabinfo_header(struct seq_file *m)
1254 * Output format version, so at least we can change it
1255 * without _too_ many complaints.
1257 #ifdef CONFIG_DEBUG_SLAB
1258 seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
1260 seq_puts(m, "slabinfo - version: 2.1\n");
1262 seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
1263 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
1264 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
1265 #ifdef CONFIG_DEBUG_SLAB
1266 seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
1267 seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
1272 void *slab_start(struct seq_file *m, loff_t *pos)
1274 mutex_lock(&slab_mutex);
1275 return seq_list_start(&slab_root_caches, *pos);
1278 void *slab_next(struct seq_file *m, void *p, loff_t *pos)
1280 return seq_list_next(p, &slab_root_caches, pos);
1283 void slab_stop(struct seq_file *m, void *p)
1285 mutex_unlock(&slab_mutex);
1289 memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info)
1291 struct kmem_cache *c;
1292 struct slabinfo sinfo;
1294 if (!is_root_cache(s))
1297 for_each_memcg_cache(c, s) {
1298 memset(&sinfo, 0, sizeof(sinfo));
1299 get_slabinfo(c, &sinfo);
1301 info->active_slabs += sinfo.active_slabs;
1302 info->num_slabs += sinfo.num_slabs;
1303 info->shared_avail += sinfo.shared_avail;
1304 info->active_objs += sinfo.active_objs;
1305 info->num_objs += sinfo.num_objs;
1309 static void cache_show(struct kmem_cache *s, struct seq_file *m)
1311 struct slabinfo sinfo;
1313 memset(&sinfo, 0, sizeof(sinfo));
1314 get_slabinfo(s, &sinfo);
1316 memcg_accumulate_slabinfo(s, &sinfo);
1318 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
1319 cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size,
1320 sinfo.objects_per_slab, (1 << sinfo.cache_order));
1322 seq_printf(m, " : tunables %4u %4u %4u",
1323 sinfo.limit, sinfo.batchcount, sinfo.shared);
1324 seq_printf(m, " : slabdata %6lu %6lu %6lu",
1325 sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
1326 slabinfo_show_stats(m, s);
1330 static int slab_show(struct seq_file *m, void *p)
1332 struct kmem_cache *s = list_entry(p, struct kmem_cache, root_caches_node);
1334 if (p == slab_root_caches.next)
1335 print_slabinfo_header(m);
1340 void dump_unreclaimable_slab(void)
1342 struct kmem_cache *s, *s2;
1343 struct slabinfo sinfo;
1346 * Here acquiring slab_mutex is risky since we don't prefer to get
1347 * sleep in oom path. But, without mutex hold, it may introduce a
1349 * Use mutex_trylock to protect the list traverse, dump nothing
1350 * without acquiring the mutex.
1352 if (!mutex_trylock(&slab_mutex)) {
1353 pr_warn("excessive unreclaimable slab but cannot dump stats\n");
1357 pr_info("Unreclaimable slab info:\n");
1358 pr_info("Name Used Total\n");
1360 list_for_each_entry_safe(s, s2, &slab_caches, list) {
1361 if (!is_root_cache(s) || (s->flags & SLAB_RECLAIM_ACCOUNT))
1364 get_slabinfo(s, &sinfo);
1366 if (sinfo.num_objs > 0)
1367 pr_info("%-17s %10luKB %10luKB\n", cache_name(s),
1368 (sinfo.active_objs * s->size) / 1024,
1369 (sinfo.num_objs * s->size) / 1024);
1371 mutex_unlock(&slab_mutex);
1374 #if defined(CONFIG_MEMCG)
1375 void *memcg_slab_start(struct seq_file *m, loff_t *pos)
1377 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
1379 mutex_lock(&slab_mutex);
1380 return seq_list_start(&memcg->kmem_caches, *pos);
1383 void *memcg_slab_next(struct seq_file *m, void *p, loff_t *pos)
1385 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
1387 return seq_list_next(p, &memcg->kmem_caches, pos);
1390 void memcg_slab_stop(struct seq_file *m, void *p)
1392 mutex_unlock(&slab_mutex);
1395 int memcg_slab_show(struct seq_file *m, void *p)
1397 struct kmem_cache *s = list_entry(p, struct kmem_cache,
1398 memcg_params.kmem_caches_node);
1399 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
1401 if (p == memcg->kmem_caches.next)
1402 print_slabinfo_header(m);
1409 * slabinfo_op - iterator that generates /proc/slabinfo
1418 * num-pages-per-slab
1419 * + further values on SMP and with statistics enabled
1421 static const struct seq_operations slabinfo_op = {
1422 .start = slab_start,
1428 static int slabinfo_open(struct inode *inode, struct file *file)
1430 return seq_open(file, &slabinfo_op);
1433 static const struct file_operations proc_slabinfo_operations = {
1434 .open = slabinfo_open,
1436 .write = slabinfo_write,
1437 .llseek = seq_lseek,
1438 .release = seq_release,
1441 static int __init slab_proc_init(void)
1443 proc_create("slabinfo", SLABINFO_RIGHTS, NULL,
1444 &proc_slabinfo_operations);
1447 module_init(slab_proc_init);
1448 #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */
1450 static __always_inline void *__do_krealloc(const void *p, size_t new_size,
1459 if (ks >= new_size) {
1460 kasan_krealloc((void *)p, new_size, flags);
1464 ret = kmalloc_track_caller(new_size, flags);
1472 * __krealloc - like krealloc() but don't free @p.
1473 * @p: object to reallocate memory for.
1474 * @new_size: how many bytes of memory are required.
1475 * @flags: the type of memory to allocate.
1477 * This function is like krealloc() except it never frees the originally
1478 * allocated buffer. Use this if you don't want to free the buffer immediately
1479 * like, for example, with RCU.
1481 void *__krealloc(const void *p, size_t new_size, gfp_t flags)
1483 if (unlikely(!new_size))
1484 return ZERO_SIZE_PTR;
1486 return __do_krealloc(p, new_size, flags);
1489 EXPORT_SYMBOL(__krealloc);
1492 * krealloc - reallocate memory. The contents will remain unchanged.
1493 * @p: object to reallocate memory for.
1494 * @new_size: how many bytes of memory are required.
1495 * @flags: the type of memory to allocate.
1497 * The contents of the object pointed to are preserved up to the
1498 * lesser of the new and old sizes. If @p is %NULL, krealloc()
1499 * behaves exactly like kmalloc(). If @new_size is 0 and @p is not a
1500 * %NULL pointer, the object pointed to is freed.
1502 void *krealloc(const void *p, size_t new_size, gfp_t flags)
1506 if (unlikely(!new_size)) {
1508 return ZERO_SIZE_PTR;
1511 ret = __do_krealloc(p, new_size, flags);
1512 if (ret && p != ret)
1517 EXPORT_SYMBOL(krealloc);
1520 * kzfree - like kfree but zero memory
1521 * @p: object to free memory of
1523 * The memory of the object @p points to is zeroed before freed.
1524 * If @p is %NULL, kzfree() does nothing.
1526 * Note: this function zeroes the whole allocated buffer which can be a good
1527 * deal bigger than the requested buffer size passed to kmalloc(). So be
1528 * careful when using this function in performance sensitive code.
1530 void kzfree(const void *p)
1533 void *mem = (void *)p;
1535 if (unlikely(ZERO_OR_NULL_PTR(mem)))
1541 EXPORT_SYMBOL(kzfree);
1543 /* Tracepoints definitions. */
1544 EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1545 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1546 EXPORT_TRACEPOINT_SYMBOL(kmalloc_node);
1547 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node);
1548 EXPORT_TRACEPOINT_SYMBOL(kfree);
1549 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);
1551 int should_failslab(struct kmem_cache *s, gfp_t gfpflags)
1553 if (__should_failslab(s, gfpflags))
1557 ALLOW_ERROR_INJECTION(should_failslab, ERRNO);