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 <asm/cacheflush.h>
22 #include <asm/tlbflush.h>
24 #include <linux/memcontrol.h>
26 #define CREATE_TRACE_POINTS
27 #include <trace/events/kmem.h>
31 enum slab_state slab_state
;
32 LIST_HEAD(slab_caches
);
33 DEFINE_MUTEX(slab_mutex
);
34 struct kmem_cache
*kmem_cache
;
36 #ifdef CONFIG_HARDENED_USERCOPY
37 bool usercopy_fallback __ro_after_init
=
38 IS_ENABLED(CONFIG_HARDENED_USERCOPY_FALLBACK
);
39 module_param(usercopy_fallback
, bool, 0400);
40 MODULE_PARM_DESC(usercopy_fallback
,
41 "WARN instead of reject usercopy whitelist violations");
44 static LIST_HEAD(slab_caches_to_rcu_destroy
);
45 static void slab_caches_to_rcu_destroy_workfn(struct work_struct
*work
);
46 static DECLARE_WORK(slab_caches_to_rcu_destroy_work
,
47 slab_caches_to_rcu_destroy_workfn
);
50 * Set of flags that will prevent slab merging
52 #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
53 SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \
54 SLAB_FAILSLAB | SLAB_KASAN)
56 #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \
57 SLAB_CACHE_DMA32 | SLAB_ACCOUNT)
60 * Merge control. If this is set then no merging of slab caches will occur.
62 static bool slab_nomerge
= !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT
);
64 static int __init
setup_slab_nomerge(char *str
)
71 __setup_param("slub_nomerge", slub_nomerge
, setup_slab_nomerge
, 0);
74 __setup("slab_nomerge", setup_slab_nomerge
);
77 * Determine the size of a slab object
79 unsigned int kmem_cache_size(struct kmem_cache
*s
)
81 return s
->object_size
;
83 EXPORT_SYMBOL(kmem_cache_size
);
85 #ifdef CONFIG_DEBUG_VM
86 static int kmem_cache_sanity_check(const char *name
, unsigned int size
)
88 if (!name
|| in_interrupt() || size
< sizeof(void *) ||
89 size
> KMALLOC_MAX_SIZE
) {
90 pr_err("kmem_cache_create(%s) integrity check failed\n", name
);
94 WARN_ON(strchr(name
, ' ')); /* It confuses parsers */
98 static inline int kmem_cache_sanity_check(const char *name
, unsigned int size
)
104 void __kmem_cache_free_bulk(struct kmem_cache
*s
, size_t nr
, void **p
)
108 for (i
= 0; i
< nr
; i
++) {
110 kmem_cache_free(s
, p
[i
]);
116 int __kmem_cache_alloc_bulk(struct kmem_cache
*s
, gfp_t flags
, size_t nr
,
121 for (i
= 0; i
< nr
; i
++) {
122 void *x
= p
[i
] = kmem_cache_alloc(s
, flags
);
124 __kmem_cache_free_bulk(s
, i
, p
);
131 #ifdef CONFIG_MEMCG_KMEM
133 LIST_HEAD(slab_root_caches
);
134 static DEFINE_SPINLOCK(memcg_kmem_wq_lock
);
136 static void kmemcg_cache_shutdown(struct percpu_ref
*percpu_ref
);
138 void slab_init_memcg_params(struct kmem_cache
*s
)
140 s
->memcg_params
.root_cache
= NULL
;
141 RCU_INIT_POINTER(s
->memcg_params
.memcg_caches
, NULL
);
142 INIT_LIST_HEAD(&s
->memcg_params
.children
);
143 s
->memcg_params
.dying
= false;
146 static int init_memcg_params(struct kmem_cache
*s
,
147 struct kmem_cache
*root_cache
)
149 struct memcg_cache_array
*arr
;
152 int ret
= percpu_ref_init(&s
->memcg_params
.refcnt
,
153 kmemcg_cache_shutdown
,
158 s
->memcg_params
.root_cache
= root_cache
;
159 INIT_LIST_HEAD(&s
->memcg_params
.children_node
);
160 INIT_LIST_HEAD(&s
->memcg_params
.kmem_caches_node
);
164 slab_init_memcg_params(s
);
166 if (!memcg_nr_cache_ids
)
169 arr
= kvzalloc(sizeof(struct memcg_cache_array
) +
170 memcg_nr_cache_ids
* sizeof(void *),
175 RCU_INIT_POINTER(s
->memcg_params
.memcg_caches
, arr
);
179 static void destroy_memcg_params(struct kmem_cache
*s
)
181 if (is_root_cache(s
)) {
182 kvfree(rcu_access_pointer(s
->memcg_params
.memcg_caches
));
184 mem_cgroup_put(s
->memcg_params
.memcg
);
185 WRITE_ONCE(s
->memcg_params
.memcg
, NULL
);
186 percpu_ref_exit(&s
->memcg_params
.refcnt
);
190 static void free_memcg_params(struct rcu_head
*rcu
)
192 struct memcg_cache_array
*old
;
194 old
= container_of(rcu
, struct memcg_cache_array
, rcu
);
198 static int update_memcg_params(struct kmem_cache
*s
, int new_array_size
)
200 struct memcg_cache_array
*old
, *new;
202 new = kvzalloc(sizeof(struct memcg_cache_array
) +
203 new_array_size
* sizeof(void *), GFP_KERNEL
);
207 old
= rcu_dereference_protected(s
->memcg_params
.memcg_caches
,
208 lockdep_is_held(&slab_mutex
));
210 memcpy(new->entries
, old
->entries
,
211 memcg_nr_cache_ids
* sizeof(void *));
213 rcu_assign_pointer(s
->memcg_params
.memcg_caches
, new);
215 call_rcu(&old
->rcu
, free_memcg_params
);
219 int memcg_update_all_caches(int num_memcgs
)
221 struct kmem_cache
*s
;
224 mutex_lock(&slab_mutex
);
225 list_for_each_entry(s
, &slab_root_caches
, root_caches_node
) {
226 ret
= update_memcg_params(s
, num_memcgs
);
228 * Instead of freeing the memory, we'll just leave the caches
229 * up to this point in an updated state.
234 mutex_unlock(&slab_mutex
);
238 void memcg_link_cache(struct kmem_cache
*s
, struct mem_cgroup
*memcg
)
240 if (is_root_cache(s
)) {
241 list_add(&s
->root_caches_node
, &slab_root_caches
);
243 css_get(&memcg
->css
);
244 s
->memcg_params
.memcg
= memcg
;
245 list_add(&s
->memcg_params
.children_node
,
246 &s
->memcg_params
.root_cache
->memcg_params
.children
);
247 list_add(&s
->memcg_params
.kmem_caches_node
,
248 &s
->memcg_params
.memcg
->kmem_caches
);
252 static void memcg_unlink_cache(struct kmem_cache
*s
)
254 if (is_root_cache(s
)) {
255 list_del(&s
->root_caches_node
);
257 list_del(&s
->memcg_params
.children_node
);
258 list_del(&s
->memcg_params
.kmem_caches_node
);
262 static inline int init_memcg_params(struct kmem_cache
*s
,
263 struct kmem_cache
*root_cache
)
268 static inline void destroy_memcg_params(struct kmem_cache
*s
)
272 static inline void memcg_unlink_cache(struct kmem_cache
*s
)
275 #endif /* CONFIG_MEMCG_KMEM */
278 * Figure out what the alignment of the objects will be given a set of
279 * flags, a user specified alignment and the size of the objects.
281 static unsigned int calculate_alignment(slab_flags_t flags
,
282 unsigned int align
, unsigned int size
)
285 * If the user wants hardware cache aligned objects then follow that
286 * suggestion if the object is sufficiently large.
288 * The hardware cache alignment cannot override the specified
289 * alignment though. If that is greater then use it.
291 if (flags
& SLAB_HWCACHE_ALIGN
) {
294 ralign
= cache_line_size();
295 while (size
<= ralign
/ 2)
297 align
= max(align
, ralign
);
300 if (align
< ARCH_SLAB_MINALIGN
)
301 align
= ARCH_SLAB_MINALIGN
;
303 return ALIGN(align
, sizeof(void *));
307 * Find a mergeable slab cache
309 int slab_unmergeable(struct kmem_cache
*s
)
311 if (slab_nomerge
|| (s
->flags
& SLAB_NEVER_MERGE
))
314 if (!is_root_cache(s
))
324 * We may have set a slab to be unmergeable during bootstrap.
329 #ifdef CONFIG_MEMCG_KMEM
331 * Skip the dying kmem_cache.
333 if (s
->memcg_params
.dying
)
340 struct kmem_cache
*find_mergeable(unsigned int size
, unsigned int align
,
341 slab_flags_t flags
, const char *name
, void (*ctor
)(void *))
343 struct kmem_cache
*s
;
351 size
= ALIGN(size
, sizeof(void *));
352 align
= calculate_alignment(flags
, align
, size
);
353 size
= ALIGN(size
, align
);
354 flags
= kmem_cache_flags(size
, flags
, name
, NULL
);
356 if (flags
& SLAB_NEVER_MERGE
)
359 list_for_each_entry_reverse(s
, &slab_root_caches
, root_caches_node
) {
360 if (slab_unmergeable(s
))
366 if ((flags
& SLAB_MERGE_SAME
) != (s
->flags
& SLAB_MERGE_SAME
))
369 * Check if alignment is compatible.
370 * Courtesy of Adrian Drzewiecki
372 if ((s
->size
& ~(align
- 1)) != s
->size
)
375 if (s
->size
- size
>= sizeof(void *))
378 if (IS_ENABLED(CONFIG_SLAB
) && align
&&
379 (align
> s
->align
|| s
->align
% align
))
387 static struct kmem_cache
*create_cache(const char *name
,
388 unsigned int object_size
, unsigned int align
,
389 slab_flags_t flags
, unsigned int useroffset
,
390 unsigned int usersize
, void (*ctor
)(void *),
391 struct mem_cgroup
*memcg
, struct kmem_cache
*root_cache
)
393 struct kmem_cache
*s
;
396 if (WARN_ON(useroffset
+ usersize
> object_size
))
397 useroffset
= usersize
= 0;
400 s
= kmem_cache_zalloc(kmem_cache
, GFP_KERNEL
);
405 s
->size
= s
->object_size
= object_size
;
408 s
->useroffset
= useroffset
;
409 s
->usersize
= usersize
;
411 err
= init_memcg_params(s
, root_cache
);
415 err
= __kmem_cache_create(s
, flags
);
420 list_add(&s
->list
, &slab_caches
);
421 memcg_link_cache(s
, memcg
);
428 destroy_memcg_params(s
);
429 kmem_cache_free(kmem_cache
, s
);
434 * kmem_cache_create_usercopy - Create a cache with a region suitable
435 * for copying to userspace
436 * @name: A string which is used in /proc/slabinfo to identify this cache.
437 * @size: The size of objects to be created in this cache.
438 * @align: The required alignment for the objects.
440 * @useroffset: Usercopy region offset
441 * @usersize: Usercopy region size
442 * @ctor: A constructor for the objects.
444 * Cannot be called within a interrupt, but can be interrupted.
445 * The @ctor is run when new pages are allocated by the cache.
449 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
450 * to catch references to uninitialised memory.
452 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
453 * for buffer overruns.
455 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
456 * cacheline. This can be beneficial if you're counting cycles as closely
459 * Return: a pointer to the cache on success, NULL on failure.
462 kmem_cache_create_usercopy(const char *name
,
463 unsigned int size
, unsigned int align
,
465 unsigned int useroffset
, unsigned int usersize
,
466 void (*ctor
)(void *))
468 struct kmem_cache
*s
= NULL
;
469 const char *cache_name
;
474 memcg_get_cache_ids();
476 mutex_lock(&slab_mutex
);
478 err
= kmem_cache_sanity_check(name
, size
);
483 /* Refuse requests with allocator specific flags */
484 if (flags
& ~SLAB_FLAGS_PERMITTED
) {
490 * Some allocators will constraint the set of valid flags to a subset
491 * of all flags. We expect them to define CACHE_CREATE_MASK in this
492 * case, and we'll just provide them with a sanitized version of the
495 flags
&= CACHE_CREATE_MASK
;
497 /* Fail closed on bad usersize of useroffset values. */
498 if (WARN_ON(!usersize
&& useroffset
) ||
499 WARN_ON(size
< usersize
|| size
- usersize
< useroffset
))
500 usersize
= useroffset
= 0;
503 s
= __kmem_cache_alias(name
, size
, align
, flags
, ctor
);
507 cache_name
= kstrdup_const(name
, GFP_KERNEL
);
513 s
= create_cache(cache_name
, size
,
514 calculate_alignment(flags
, align
, size
),
515 flags
, useroffset
, usersize
, ctor
, NULL
, NULL
);
518 kfree_const(cache_name
);
522 mutex_unlock(&slab_mutex
);
524 memcg_put_cache_ids();
529 if (flags
& SLAB_PANIC
)
530 panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n",
533 pr_warn("kmem_cache_create(%s) failed with error %d\n",
541 EXPORT_SYMBOL(kmem_cache_create_usercopy
);
544 * kmem_cache_create - Create a cache.
545 * @name: A string which is used in /proc/slabinfo to identify this cache.
546 * @size: The size of objects to be created in this cache.
547 * @align: The required alignment for the objects.
549 * @ctor: A constructor for the objects.
551 * Cannot be called within a interrupt, but can be interrupted.
552 * The @ctor is run when new pages are allocated by the cache.
556 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
557 * to catch references to uninitialised memory.
559 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
560 * for buffer overruns.
562 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
563 * cacheline. This can be beneficial if you're counting cycles as closely
566 * Return: a pointer to the cache on success, NULL on failure.
569 kmem_cache_create(const char *name
, unsigned int size
, unsigned int align
,
570 slab_flags_t flags
, void (*ctor
)(void *))
572 return kmem_cache_create_usercopy(name
, size
, align
, flags
, 0, 0,
575 EXPORT_SYMBOL(kmem_cache_create
);
577 static void slab_caches_to_rcu_destroy_workfn(struct work_struct
*work
)
579 LIST_HEAD(to_destroy
);
580 struct kmem_cache
*s
, *s2
;
583 * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the
584 * @slab_caches_to_rcu_destroy list. The slab pages are freed
585 * through RCU and and the associated kmem_cache are dereferenced
586 * while freeing the pages, so the kmem_caches should be freed only
587 * after the pending RCU operations are finished. As rcu_barrier()
588 * is a pretty slow operation, we batch all pending destructions
591 mutex_lock(&slab_mutex
);
592 list_splice_init(&slab_caches_to_rcu_destroy
, &to_destroy
);
593 mutex_unlock(&slab_mutex
);
595 if (list_empty(&to_destroy
))
600 list_for_each_entry_safe(s
, s2
, &to_destroy
, list
) {
601 #ifdef SLAB_SUPPORTS_SYSFS
602 sysfs_slab_release(s
);
604 slab_kmem_cache_release(s
);
609 static int shutdown_cache(struct kmem_cache
*s
)
611 /* free asan quarantined objects */
612 kasan_cache_shutdown(s
);
614 if (__kmem_cache_shutdown(s
) != 0)
617 memcg_unlink_cache(s
);
620 if (s
->flags
& SLAB_TYPESAFE_BY_RCU
) {
621 #ifdef SLAB_SUPPORTS_SYSFS
622 sysfs_slab_unlink(s
);
624 list_add_tail(&s
->list
, &slab_caches_to_rcu_destroy
);
625 schedule_work(&slab_caches_to_rcu_destroy_work
);
627 #ifdef SLAB_SUPPORTS_SYSFS
628 sysfs_slab_unlink(s
);
629 sysfs_slab_release(s
);
631 slab_kmem_cache_release(s
);
638 #ifdef CONFIG_MEMCG_KMEM
640 * memcg_create_kmem_cache - Create a cache for a memory cgroup.
641 * @memcg: The memory cgroup the new cache is for.
642 * @root_cache: The parent of the new cache.
644 * This function attempts to create a kmem cache that will serve allocation
645 * requests going from @memcg to @root_cache. The new cache inherits properties
648 void memcg_create_kmem_cache(struct mem_cgroup
*memcg
,
649 struct kmem_cache
*root_cache
)
651 static char memcg_name_buf
[NAME_MAX
+ 1]; /* protected by slab_mutex */
652 struct cgroup_subsys_state
*css
= &memcg
->css
;
653 struct memcg_cache_array
*arr
;
654 struct kmem_cache
*s
= NULL
;
661 mutex_lock(&slab_mutex
);
664 * The memory cgroup could have been offlined while the cache
665 * creation work was pending.
667 if (memcg
->kmem_state
!= KMEM_ONLINE
)
670 idx
= memcg_cache_id(memcg
);
671 arr
= rcu_dereference_protected(root_cache
->memcg_params
.memcg_caches
,
672 lockdep_is_held(&slab_mutex
));
675 * Since per-memcg caches are created asynchronously on first
676 * allocation (see memcg_kmem_get_cache()), several threads can try to
677 * create the same cache, but only one of them may succeed.
679 if (arr
->entries
[idx
])
682 cgroup_name(css
->cgroup
, memcg_name_buf
, sizeof(memcg_name_buf
));
683 cache_name
= kasprintf(GFP_KERNEL
, "%s(%llu:%s)", root_cache
->name
,
684 css
->serial_nr
, memcg_name_buf
);
688 s
= create_cache(cache_name
, root_cache
->object_size
,
690 root_cache
->flags
& CACHE_CREATE_MASK
,
691 root_cache
->useroffset
, root_cache
->usersize
,
692 root_cache
->ctor
, memcg
, root_cache
);
694 * If we could not create a memcg cache, do not complain, because
695 * that's not critical at all as we can always proceed with the root
704 * Since readers won't lock (see memcg_kmem_get_cache()), we need a
705 * barrier here to ensure nobody will see the kmem_cache partially
709 arr
->entries
[idx
] = s
;
712 mutex_unlock(&slab_mutex
);
718 static void kmemcg_workfn(struct work_struct
*work
)
720 struct kmem_cache
*s
= container_of(work
, struct kmem_cache
,
726 mutex_lock(&slab_mutex
);
727 s
->memcg_params
.work_fn(s
);
728 mutex_unlock(&slab_mutex
);
734 static void kmemcg_rcufn(struct rcu_head
*head
)
736 struct kmem_cache
*s
= container_of(head
, struct kmem_cache
,
737 memcg_params
.rcu_head
);
740 * We need to grab blocking locks. Bounce to ->work. The
741 * work item shares the space with the RCU head and can't be
742 * initialized earlier.
744 INIT_WORK(&s
->memcg_params
.work
, kmemcg_workfn
);
745 queue_work(memcg_kmem_cache_wq
, &s
->memcg_params
.work
);
748 static void kmemcg_cache_shutdown_fn(struct kmem_cache
*s
)
750 WARN_ON(shutdown_cache(s
));
753 static void kmemcg_cache_shutdown(struct percpu_ref
*percpu_ref
)
755 struct kmem_cache
*s
= container_of(percpu_ref
, struct kmem_cache
,
756 memcg_params
.refcnt
);
759 spin_lock_irqsave(&memcg_kmem_wq_lock
, flags
);
760 if (s
->memcg_params
.root_cache
->memcg_params
.dying
)
763 s
->memcg_params
.work_fn
= kmemcg_cache_shutdown_fn
;
764 INIT_WORK(&s
->memcg_params
.work
, kmemcg_workfn
);
765 queue_work(memcg_kmem_cache_wq
, &s
->memcg_params
.work
);
768 spin_unlock_irqrestore(&memcg_kmem_wq_lock
, flags
);
771 static void kmemcg_cache_deactivate_after_rcu(struct kmem_cache
*s
)
773 __kmemcg_cache_deactivate_after_rcu(s
);
774 percpu_ref_kill(&s
->memcg_params
.refcnt
);
777 static void kmemcg_cache_deactivate(struct kmem_cache
*s
)
779 if (WARN_ON_ONCE(is_root_cache(s
)))
782 __kmemcg_cache_deactivate(s
);
783 s
->flags
|= SLAB_DEACTIVATED
;
786 * memcg_kmem_wq_lock is used to synchronize memcg_params.dying
787 * flag and make sure that no new kmem_cache deactivation tasks
788 * are queued (see flush_memcg_workqueue() ).
790 spin_lock_irq(&memcg_kmem_wq_lock
);
791 if (s
->memcg_params
.root_cache
->memcg_params
.dying
)
794 s
->memcg_params
.work_fn
= kmemcg_cache_deactivate_after_rcu
;
795 call_rcu(&s
->memcg_params
.rcu_head
, kmemcg_rcufn
);
797 spin_unlock_irq(&memcg_kmem_wq_lock
);
800 void memcg_deactivate_kmem_caches(struct mem_cgroup
*memcg
,
801 struct mem_cgroup
*parent
)
804 struct memcg_cache_array
*arr
;
805 struct kmem_cache
*s
, *c
;
806 unsigned int nr_reparented
;
808 idx
= memcg_cache_id(memcg
);
813 mutex_lock(&slab_mutex
);
814 list_for_each_entry(s
, &slab_root_caches
, root_caches_node
) {
815 arr
= rcu_dereference_protected(s
->memcg_params
.memcg_caches
,
816 lockdep_is_held(&slab_mutex
));
817 c
= arr
->entries
[idx
];
821 kmemcg_cache_deactivate(c
);
822 arr
->entries
[idx
] = NULL
;
825 list_for_each_entry(s
, &memcg
->kmem_caches
,
826 memcg_params
.kmem_caches_node
) {
827 WRITE_ONCE(s
->memcg_params
.memcg
, parent
);
828 css_put(&memcg
->css
);
832 list_splice_init(&memcg
->kmem_caches
,
833 &parent
->kmem_caches
);
834 css_get_many(&parent
->css
, nr_reparented
);
836 mutex_unlock(&slab_mutex
);
842 static int shutdown_memcg_caches(struct kmem_cache
*s
)
844 struct memcg_cache_array
*arr
;
845 struct kmem_cache
*c
, *c2
;
849 BUG_ON(!is_root_cache(s
));
852 * First, shutdown active caches, i.e. caches that belong to online
855 arr
= rcu_dereference_protected(s
->memcg_params
.memcg_caches
,
856 lockdep_is_held(&slab_mutex
));
857 for_each_memcg_cache_index(i
) {
861 if (shutdown_cache(c
))
863 * The cache still has objects. Move it to a temporary
864 * list so as not to try to destroy it for a second
865 * time while iterating over inactive caches below.
867 list_move(&c
->memcg_params
.children_node
, &busy
);
870 * The cache is empty and will be destroyed soon. Clear
871 * the pointer to it in the memcg_caches array so that
872 * it will never be accessed even if the root cache
875 arr
->entries
[i
] = NULL
;
879 * Second, shutdown all caches left from memory cgroups that are now
882 list_for_each_entry_safe(c
, c2
, &s
->memcg_params
.children
,
883 memcg_params
.children_node
)
886 list_splice(&busy
, &s
->memcg_params
.children
);
889 * A cache being destroyed must be empty. In particular, this means
890 * that all per memcg caches attached to it must be empty too.
892 if (!list_empty(&s
->memcg_params
.children
))
897 static void memcg_set_kmem_cache_dying(struct kmem_cache
*s
)
899 spin_lock_irq(&memcg_kmem_wq_lock
);
900 s
->memcg_params
.dying
= true;
901 spin_unlock_irq(&memcg_kmem_wq_lock
);
904 static void flush_memcg_workqueue(struct kmem_cache
*s
)
907 * SLAB and SLUB deactivate the kmem_caches through call_rcu. Make
908 * sure all registered rcu callbacks have been invoked.
913 * SLAB and SLUB create memcg kmem_caches through workqueue and SLUB
914 * deactivates the memcg kmem_caches through workqueue. Make sure all
915 * previous workitems on workqueue are processed.
917 if (likely(memcg_kmem_cache_wq
))
918 flush_workqueue(memcg_kmem_cache_wq
);
921 * If we're racing with children kmem_cache deactivation, it might
922 * take another rcu grace period to complete their destruction.
923 * At this moment the corresponding percpu_ref_kill() call should be
924 * done, but it might take another rcu grace period to complete
925 * switching to the atomic mode.
926 * Please, note that we check without grabbing the slab_mutex. It's safe
927 * because at this moment the children list can't grow.
929 if (!list_empty(&s
->memcg_params
.children
))
933 static inline int shutdown_memcg_caches(struct kmem_cache
*s
)
937 #endif /* CONFIG_MEMCG_KMEM */
939 void slab_kmem_cache_release(struct kmem_cache
*s
)
941 __kmem_cache_release(s
);
942 destroy_memcg_params(s
);
943 kfree_const(s
->name
);
944 kmem_cache_free(kmem_cache
, s
);
947 void kmem_cache_destroy(struct kmem_cache
*s
)
957 mutex_lock(&slab_mutex
);
963 #ifdef CONFIG_MEMCG_KMEM
964 memcg_set_kmem_cache_dying(s
);
966 mutex_unlock(&slab_mutex
);
971 flush_memcg_workqueue(s
);
976 mutex_lock(&slab_mutex
);
979 err
= shutdown_memcg_caches(s
);
981 err
= shutdown_cache(s
);
984 pr_err("kmem_cache_destroy %s: Slab cache still has objects\n",
989 mutex_unlock(&slab_mutex
);
994 EXPORT_SYMBOL(kmem_cache_destroy
);
997 * kmem_cache_shrink - Shrink a cache.
998 * @cachep: The cache to shrink.
1000 * Releases as many slabs as possible for a cache.
1001 * To help debugging, a zero exit status indicates all slabs were released.
1003 * Return: %0 if all slabs were released, non-zero otherwise
1005 int kmem_cache_shrink(struct kmem_cache
*cachep
)
1011 kasan_cache_shrink(cachep
);
1012 ret
= __kmem_cache_shrink(cachep
);
1017 EXPORT_SYMBOL(kmem_cache_shrink
);
1020 * kmem_cache_shrink_all - shrink a cache and all memcg caches for root cache
1021 * @s: The cache pointer
1023 void kmem_cache_shrink_all(struct kmem_cache
*s
)
1025 struct kmem_cache
*c
;
1027 if (!IS_ENABLED(CONFIG_MEMCG_KMEM
) || !is_root_cache(s
)) {
1028 kmem_cache_shrink(s
);
1034 kasan_cache_shrink(s
);
1035 __kmem_cache_shrink(s
);
1038 * We have to take the slab_mutex to protect from the memcg list
1041 mutex_lock(&slab_mutex
);
1042 for_each_memcg_cache(c
, s
) {
1044 * Don't need to shrink deactivated memcg caches.
1046 if (s
->flags
& SLAB_DEACTIVATED
)
1048 kasan_cache_shrink(c
);
1049 __kmem_cache_shrink(c
);
1051 mutex_unlock(&slab_mutex
);
1056 bool slab_is_available(void)
1058 return slab_state
>= UP
;
1062 /* Create a cache during boot when no slab services are available yet */
1063 void __init
create_boot_cache(struct kmem_cache
*s
, const char *name
,
1064 unsigned int size
, slab_flags_t flags
,
1065 unsigned int useroffset
, unsigned int usersize
)
1068 unsigned int align
= ARCH_KMALLOC_MINALIGN
;
1071 s
->size
= s
->object_size
= size
;
1074 * For power of two sizes, guarantee natural alignment for kmalloc
1075 * caches, regardless of SL*B debugging options.
1077 if (is_power_of_2(size
))
1078 align
= max(align
, size
);
1079 s
->align
= calculate_alignment(flags
, align
, size
);
1081 s
->useroffset
= useroffset
;
1082 s
->usersize
= usersize
;
1084 slab_init_memcg_params(s
);
1086 err
= __kmem_cache_create(s
, flags
);
1089 panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n",
1092 s
->refcount
= -1; /* Exempt from merging for now */
1095 struct kmem_cache
*__init
create_kmalloc_cache(const char *name
,
1096 unsigned int size
, slab_flags_t flags
,
1097 unsigned int useroffset
, unsigned int usersize
)
1099 struct kmem_cache
*s
= kmem_cache_zalloc(kmem_cache
, GFP_NOWAIT
);
1102 panic("Out of memory when creating slab %s\n", name
);
1104 create_boot_cache(s
, name
, size
, flags
, useroffset
, usersize
);
1105 list_add(&s
->list
, &slab_caches
);
1106 memcg_link_cache(s
, NULL
);
1112 kmalloc_caches
[NR_KMALLOC_TYPES
][KMALLOC_SHIFT_HIGH
+ 1] __ro_after_init
=
1113 { /* initialization for https://bugs.llvm.org/show_bug.cgi?id=42570 */ };
1114 EXPORT_SYMBOL(kmalloc_caches
);
1117 * Conversion table for small slabs sizes / 8 to the index in the
1118 * kmalloc array. This is necessary for slabs < 192 since we have non power
1119 * of two cache sizes there. The size of larger slabs can be determined using
1122 static u8 size_index
[24] __ro_after_init
= {
1149 static inline unsigned int size_index_elem(unsigned int bytes
)
1151 return (bytes
- 1) / 8;
1155 * Find the kmem_cache structure that serves a given size of
1158 struct kmem_cache
*kmalloc_slab(size_t size
, gfp_t flags
)
1164 return ZERO_SIZE_PTR
;
1166 index
= size_index
[size_index_elem(size
)];
1168 if (WARN_ON_ONCE(size
> KMALLOC_MAX_CACHE_SIZE
))
1170 index
= fls(size
- 1);
1173 return kmalloc_caches
[kmalloc_type(flags
)][index
];
1176 #ifdef CONFIG_ZONE_DMA
1177 #define INIT_KMALLOC_INFO(__size, __short_size) \
1179 .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \
1180 .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #__short_size, \
1181 .name[KMALLOC_DMA] = "dma-kmalloc-" #__short_size, \
1185 #define INIT_KMALLOC_INFO(__size, __short_size) \
1187 .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \
1188 .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #__short_size, \
1194 * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
1195 * kmalloc_index() supports up to 2^26=64MB, so the final entry of the table is
1198 const struct kmalloc_info_struct kmalloc_info
[] __initconst
= {
1199 INIT_KMALLOC_INFO(0, 0),
1200 INIT_KMALLOC_INFO(96, 96),
1201 INIT_KMALLOC_INFO(192, 192),
1202 INIT_KMALLOC_INFO(8, 8),
1203 INIT_KMALLOC_INFO(16, 16),
1204 INIT_KMALLOC_INFO(32, 32),
1205 INIT_KMALLOC_INFO(64, 64),
1206 INIT_KMALLOC_INFO(128, 128),
1207 INIT_KMALLOC_INFO(256, 256),
1208 INIT_KMALLOC_INFO(512, 512),
1209 INIT_KMALLOC_INFO(1024, 1k
),
1210 INIT_KMALLOC_INFO(2048, 2k
),
1211 INIT_KMALLOC_INFO(4096, 4k
),
1212 INIT_KMALLOC_INFO(8192, 8k
),
1213 INIT_KMALLOC_INFO(16384, 16k
),
1214 INIT_KMALLOC_INFO(32768, 32k
),
1215 INIT_KMALLOC_INFO(65536, 64k
),
1216 INIT_KMALLOC_INFO(131072, 128k
),
1217 INIT_KMALLOC_INFO(262144, 256k
),
1218 INIT_KMALLOC_INFO(524288, 512k
),
1219 INIT_KMALLOC_INFO(1048576, 1M
),
1220 INIT_KMALLOC_INFO(2097152, 2M
),
1221 INIT_KMALLOC_INFO(4194304, 4M
),
1222 INIT_KMALLOC_INFO(8388608, 8M
),
1223 INIT_KMALLOC_INFO(16777216, 16M
),
1224 INIT_KMALLOC_INFO(33554432, 32M
),
1225 INIT_KMALLOC_INFO(67108864, 64M
)
1229 * Patch up the size_index table if we have strange large alignment
1230 * requirements for the kmalloc array. This is only the case for
1231 * MIPS it seems. The standard arches will not generate any code here.
1233 * Largest permitted alignment is 256 bytes due to the way we
1234 * handle the index determination for the smaller caches.
1236 * Make sure that nothing crazy happens if someone starts tinkering
1237 * around with ARCH_KMALLOC_MINALIGN
1239 void __init
setup_kmalloc_cache_index_table(void)
1243 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 256 ||
1244 (KMALLOC_MIN_SIZE
& (KMALLOC_MIN_SIZE
- 1)));
1246 for (i
= 8; i
< KMALLOC_MIN_SIZE
; i
+= 8) {
1247 unsigned int elem
= size_index_elem(i
);
1249 if (elem
>= ARRAY_SIZE(size_index
))
1251 size_index
[elem
] = KMALLOC_SHIFT_LOW
;
1254 if (KMALLOC_MIN_SIZE
>= 64) {
1256 * The 96 byte size cache is not used if the alignment
1259 for (i
= 64 + 8; i
<= 96; i
+= 8)
1260 size_index
[size_index_elem(i
)] = 7;
1264 if (KMALLOC_MIN_SIZE
>= 128) {
1266 * The 192 byte sized cache is not used if the alignment
1267 * is 128 byte. Redirect kmalloc to use the 256 byte cache
1270 for (i
= 128 + 8; i
<= 192; i
+= 8)
1271 size_index
[size_index_elem(i
)] = 8;
1276 new_kmalloc_cache(int idx
, enum kmalloc_cache_type type
, slab_flags_t flags
)
1278 if (type
== KMALLOC_RECLAIM
)
1279 flags
|= SLAB_RECLAIM_ACCOUNT
;
1281 kmalloc_caches
[type
][idx
] = create_kmalloc_cache(
1282 kmalloc_info
[idx
].name
[type
],
1283 kmalloc_info
[idx
].size
, flags
, 0,
1284 kmalloc_info
[idx
].size
);
1288 * Create the kmalloc array. Some of the regular kmalloc arrays
1289 * may already have been created because they were needed to
1290 * enable allocations for slab creation.
1292 void __init
create_kmalloc_caches(slab_flags_t flags
)
1295 enum kmalloc_cache_type type
;
1297 for (type
= KMALLOC_NORMAL
; type
<= KMALLOC_RECLAIM
; type
++) {
1298 for (i
= KMALLOC_SHIFT_LOW
; i
<= KMALLOC_SHIFT_HIGH
; i
++) {
1299 if (!kmalloc_caches
[type
][i
])
1300 new_kmalloc_cache(i
, type
, flags
);
1303 * Caches that are not of the two-to-the-power-of size.
1304 * These have to be created immediately after the
1305 * earlier power of two caches
1307 if (KMALLOC_MIN_SIZE
<= 32 && i
== 6 &&
1308 !kmalloc_caches
[type
][1])
1309 new_kmalloc_cache(1, type
, flags
);
1310 if (KMALLOC_MIN_SIZE
<= 64 && i
== 7 &&
1311 !kmalloc_caches
[type
][2])
1312 new_kmalloc_cache(2, type
, flags
);
1316 /* Kmalloc array is now usable */
1319 #ifdef CONFIG_ZONE_DMA
1320 for (i
= 0; i
<= KMALLOC_SHIFT_HIGH
; i
++) {
1321 struct kmem_cache
*s
= kmalloc_caches
[KMALLOC_NORMAL
][i
];
1324 kmalloc_caches
[KMALLOC_DMA
][i
] = create_kmalloc_cache(
1325 kmalloc_info
[i
].name
[KMALLOC_DMA
],
1326 kmalloc_info
[i
].size
,
1327 SLAB_CACHE_DMA
| flags
, 0,
1328 kmalloc_info
[i
].size
);
1333 #endif /* !CONFIG_SLOB */
1336 * To avoid unnecessary overhead, we pass through large allocation requests
1337 * directly to the page allocator. We use __GFP_COMP, because we will need to
1338 * know the allocation order to free the pages properly in kfree.
1340 void *kmalloc_order(size_t size
, gfp_t flags
, unsigned int order
)
1345 flags
|= __GFP_COMP
;
1346 page
= alloc_pages(flags
, order
);
1348 ret
= page_address(page
);
1349 mod_node_page_state(page_pgdat(page
), NR_SLAB_UNRECLAIMABLE
,
1352 ret
= kasan_kmalloc_large(ret
, size
, flags
);
1353 /* As ret might get tagged, call kmemleak hook after KASAN. */
1354 kmemleak_alloc(ret
, size
, 1, flags
);
1357 EXPORT_SYMBOL(kmalloc_order
);
1359 #ifdef CONFIG_TRACING
1360 void *kmalloc_order_trace(size_t size
, gfp_t flags
, unsigned int order
)
1362 void *ret
= kmalloc_order(size
, flags
, order
);
1363 trace_kmalloc(_RET_IP_
, ret
, size
, PAGE_SIZE
<< order
, flags
);
1366 EXPORT_SYMBOL(kmalloc_order_trace
);
1369 #ifdef CONFIG_SLAB_FREELIST_RANDOM
1370 /* Randomize a generic freelist */
1371 static void freelist_randomize(struct rnd_state
*state
, unsigned int *list
,
1377 for (i
= 0; i
< count
; i
++)
1380 /* Fisher-Yates shuffle */
1381 for (i
= count
- 1; i
> 0; i
--) {
1382 rand
= prandom_u32_state(state
);
1384 swap(list
[i
], list
[rand
]);
1388 /* Create a random sequence per cache */
1389 int cache_random_seq_create(struct kmem_cache
*cachep
, unsigned int count
,
1392 struct rnd_state state
;
1394 if (count
< 2 || cachep
->random_seq
)
1397 cachep
->random_seq
= kcalloc(count
, sizeof(unsigned int), gfp
);
1398 if (!cachep
->random_seq
)
1401 /* Get best entropy at this stage of boot */
1402 prandom_seed_state(&state
, get_random_long());
1404 freelist_randomize(&state
, cachep
->random_seq
, count
);
1408 /* Destroy the per-cache random freelist sequence */
1409 void cache_random_seq_destroy(struct kmem_cache
*cachep
)
1411 kfree(cachep
->random_seq
);
1412 cachep
->random_seq
= NULL
;
1414 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
1416 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
1418 #define SLABINFO_RIGHTS (0600)
1420 #define SLABINFO_RIGHTS (0400)
1423 static void print_slabinfo_header(struct seq_file
*m
)
1426 * Output format version, so at least we can change it
1427 * without _too_ many complaints.
1429 #ifdef CONFIG_DEBUG_SLAB
1430 seq_puts(m
, "slabinfo - version: 2.1 (statistics)\n");
1432 seq_puts(m
, "slabinfo - version: 2.1\n");
1434 seq_puts(m
, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
1435 seq_puts(m
, " : tunables <limit> <batchcount> <sharedfactor>");
1436 seq_puts(m
, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
1437 #ifdef CONFIG_DEBUG_SLAB
1438 seq_puts(m
, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
1439 seq_puts(m
, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
1444 void *slab_start(struct seq_file
*m
, loff_t
*pos
)
1446 mutex_lock(&slab_mutex
);
1447 return seq_list_start(&slab_root_caches
, *pos
);
1450 void *slab_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
1452 return seq_list_next(p
, &slab_root_caches
, pos
);
1455 void slab_stop(struct seq_file
*m
, void *p
)
1457 mutex_unlock(&slab_mutex
);
1461 memcg_accumulate_slabinfo(struct kmem_cache
*s
, struct slabinfo
*info
)
1463 struct kmem_cache
*c
;
1464 struct slabinfo sinfo
;
1466 if (!is_root_cache(s
))
1469 for_each_memcg_cache(c
, s
) {
1470 memset(&sinfo
, 0, sizeof(sinfo
));
1471 get_slabinfo(c
, &sinfo
);
1473 info
->active_slabs
+= sinfo
.active_slabs
;
1474 info
->num_slabs
+= sinfo
.num_slabs
;
1475 info
->shared_avail
+= sinfo
.shared_avail
;
1476 info
->active_objs
+= sinfo
.active_objs
;
1477 info
->num_objs
+= sinfo
.num_objs
;
1481 static void cache_show(struct kmem_cache
*s
, struct seq_file
*m
)
1483 struct slabinfo sinfo
;
1485 memset(&sinfo
, 0, sizeof(sinfo
));
1486 get_slabinfo(s
, &sinfo
);
1488 memcg_accumulate_slabinfo(s
, &sinfo
);
1490 seq_printf(m
, "%-17s %6lu %6lu %6u %4u %4d",
1491 cache_name(s
), sinfo
.active_objs
, sinfo
.num_objs
, s
->size
,
1492 sinfo
.objects_per_slab
, (1 << sinfo
.cache_order
));
1494 seq_printf(m
, " : tunables %4u %4u %4u",
1495 sinfo
.limit
, sinfo
.batchcount
, sinfo
.shared
);
1496 seq_printf(m
, " : slabdata %6lu %6lu %6lu",
1497 sinfo
.active_slabs
, sinfo
.num_slabs
, sinfo
.shared_avail
);
1498 slabinfo_show_stats(m
, s
);
1502 static int slab_show(struct seq_file
*m
, void *p
)
1504 struct kmem_cache
*s
= list_entry(p
, struct kmem_cache
, root_caches_node
);
1506 if (p
== slab_root_caches
.next
)
1507 print_slabinfo_header(m
);
1512 void dump_unreclaimable_slab(void)
1514 struct kmem_cache
*s
, *s2
;
1515 struct slabinfo sinfo
;
1518 * Here acquiring slab_mutex is risky since we don't prefer to get
1519 * sleep in oom path. But, without mutex hold, it may introduce a
1521 * Use mutex_trylock to protect the list traverse, dump nothing
1522 * without acquiring the mutex.
1524 if (!mutex_trylock(&slab_mutex
)) {
1525 pr_warn("excessive unreclaimable slab but cannot dump stats\n");
1529 pr_info("Unreclaimable slab info:\n");
1530 pr_info("Name Used Total\n");
1532 list_for_each_entry_safe(s
, s2
, &slab_caches
, list
) {
1533 if (!is_root_cache(s
) || (s
->flags
& SLAB_RECLAIM_ACCOUNT
))
1536 get_slabinfo(s
, &sinfo
);
1538 if (sinfo
.num_objs
> 0)
1539 pr_info("%-17s %10luKB %10luKB\n", cache_name(s
),
1540 (sinfo
.active_objs
* s
->size
) / 1024,
1541 (sinfo
.num_objs
* s
->size
) / 1024);
1543 mutex_unlock(&slab_mutex
);
1546 #if defined(CONFIG_MEMCG_KMEM)
1547 void *memcg_slab_start(struct seq_file
*m
, loff_t
*pos
)
1549 struct mem_cgroup
*memcg
= mem_cgroup_from_seq(m
);
1551 mutex_lock(&slab_mutex
);
1552 return seq_list_start(&memcg
->kmem_caches
, *pos
);
1555 void *memcg_slab_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
1557 struct mem_cgroup
*memcg
= mem_cgroup_from_seq(m
);
1559 return seq_list_next(p
, &memcg
->kmem_caches
, pos
);
1562 void memcg_slab_stop(struct seq_file
*m
, void *p
)
1564 mutex_unlock(&slab_mutex
);
1567 int memcg_slab_show(struct seq_file
*m
, void *p
)
1569 struct kmem_cache
*s
= list_entry(p
, struct kmem_cache
,
1570 memcg_params
.kmem_caches_node
);
1571 struct mem_cgroup
*memcg
= mem_cgroup_from_seq(m
);
1573 if (p
== memcg
->kmem_caches
.next
)
1574 print_slabinfo_header(m
);
1581 * slabinfo_op - iterator that generates /proc/slabinfo
1590 * num-pages-per-slab
1591 * + further values on SMP and with statistics enabled
1593 static const struct seq_operations slabinfo_op
= {
1594 .start
= slab_start
,
1600 static int slabinfo_open(struct inode
*inode
, struct file
*file
)
1602 return seq_open(file
, &slabinfo_op
);
1605 static const struct proc_ops slabinfo_proc_ops
= {
1606 .proc_flags
= PROC_ENTRY_PERMANENT
,
1607 .proc_open
= slabinfo_open
,
1608 .proc_read
= seq_read
,
1609 .proc_write
= slabinfo_write
,
1610 .proc_lseek
= seq_lseek
,
1611 .proc_release
= seq_release
,
1614 static int __init
slab_proc_init(void)
1616 proc_create("slabinfo", SLABINFO_RIGHTS
, NULL
, &slabinfo_proc_ops
);
1619 module_init(slab_proc_init
);
1621 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_MEMCG_KMEM)
1623 * Display information about kmem caches that have child memcg caches.
1625 static int memcg_slabinfo_show(struct seq_file
*m
, void *unused
)
1627 struct kmem_cache
*s
, *c
;
1628 struct slabinfo sinfo
;
1630 mutex_lock(&slab_mutex
);
1631 seq_puts(m
, "# <name> <css_id[:dead|deact]> <active_objs> <num_objs>");
1632 seq_puts(m
, " <active_slabs> <num_slabs>\n");
1633 list_for_each_entry(s
, &slab_root_caches
, root_caches_node
) {
1635 * Skip kmem caches that don't have any memcg children.
1637 if (list_empty(&s
->memcg_params
.children
))
1640 memset(&sinfo
, 0, sizeof(sinfo
));
1641 get_slabinfo(s
, &sinfo
);
1642 seq_printf(m
, "%-17s root %6lu %6lu %6lu %6lu\n",
1643 cache_name(s
), sinfo
.active_objs
, sinfo
.num_objs
,
1644 sinfo
.active_slabs
, sinfo
.num_slabs
);
1646 for_each_memcg_cache(c
, s
) {
1647 struct cgroup_subsys_state
*css
;
1650 css
= &c
->memcg_params
.memcg
->css
;
1651 if (!(css
->flags
& CSS_ONLINE
))
1653 else if (c
->flags
& SLAB_DEACTIVATED
)
1656 memset(&sinfo
, 0, sizeof(sinfo
));
1657 get_slabinfo(c
, &sinfo
);
1658 seq_printf(m
, "%-17s %4d%-6s %6lu %6lu %6lu %6lu\n",
1659 cache_name(c
), css
->id
, status
,
1660 sinfo
.active_objs
, sinfo
.num_objs
,
1661 sinfo
.active_slabs
, sinfo
.num_slabs
);
1664 mutex_unlock(&slab_mutex
);
1667 DEFINE_SHOW_ATTRIBUTE(memcg_slabinfo
);
1669 static int __init
memcg_slabinfo_init(void)
1671 debugfs_create_file("memcg_slabinfo", S_IFREG
| S_IRUGO
,
1672 NULL
, NULL
, &memcg_slabinfo_fops
);
1676 late_initcall(memcg_slabinfo_init
);
1677 #endif /* CONFIG_DEBUG_FS && CONFIG_MEMCG_KMEM */
1678 #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */
1680 static __always_inline
void *__do_krealloc(const void *p
, size_t new_size
,
1689 if (ks
>= new_size
) {
1690 p
= kasan_krealloc((void *)p
, new_size
, flags
);
1694 ret
= kmalloc_track_caller(new_size
, flags
);
1702 * krealloc - reallocate memory. The contents will remain unchanged.
1703 * @p: object to reallocate memory for.
1704 * @new_size: how many bytes of memory are required.
1705 * @flags: the type of memory to allocate.
1707 * The contents of the object pointed to are preserved up to the
1708 * lesser of the new and old sizes. If @p is %NULL, krealloc()
1709 * behaves exactly like kmalloc(). If @new_size is 0 and @p is not a
1710 * %NULL pointer, the object pointed to is freed.
1712 * Return: pointer to the allocated memory or %NULL in case of error
1714 void *krealloc(const void *p
, size_t new_size
, gfp_t flags
)
1718 if (unlikely(!new_size
)) {
1720 return ZERO_SIZE_PTR
;
1723 ret
= __do_krealloc(p
, new_size
, flags
);
1724 if (ret
&& kasan_reset_tag(p
) != kasan_reset_tag(ret
))
1729 EXPORT_SYMBOL(krealloc
);
1732 * kzfree - like kfree but zero memory
1733 * @p: object to free memory of
1735 * The memory of the object @p points to is zeroed before freed.
1736 * If @p is %NULL, kzfree() does nothing.
1738 * Note: this function zeroes the whole allocated buffer which can be a good
1739 * deal bigger than the requested buffer size passed to kmalloc(). So be
1740 * careful when using this function in performance sensitive code.
1742 void kzfree(const void *p
)
1745 void *mem
= (void *)p
;
1747 if (unlikely(ZERO_OR_NULL_PTR(mem
)))
1750 memzero_explicit(mem
, ks
);
1753 EXPORT_SYMBOL(kzfree
);
1756 * ksize - get the actual amount of memory allocated for a given object
1757 * @objp: Pointer to the object
1759 * kmalloc may internally round up allocations and return more memory
1760 * than requested. ksize() can be used to determine the actual amount of
1761 * memory allocated. The caller may use this additional memory, even though
1762 * a smaller amount of memory was initially specified with the kmalloc call.
1763 * The caller must guarantee that objp points to a valid object previously
1764 * allocated with either kmalloc() or kmem_cache_alloc(). The object
1765 * must not be freed during the duration of the call.
1767 * Return: size of the actual memory used by @objp in bytes
1769 size_t ksize(const void *objp
)
1773 if (WARN_ON_ONCE(!objp
))
1776 * We need to check that the pointed to object is valid, and only then
1777 * unpoison the shadow memory below. We use __kasan_check_read(), to
1778 * generate a more useful report at the time ksize() is called (rather
1779 * than later where behaviour is undefined due to potential
1780 * use-after-free or double-free).
1782 * If the pointed to memory is invalid we return 0, to avoid users of
1783 * ksize() writing to and potentially corrupting the memory region.
1785 * We want to perform the check before __ksize(), to avoid potentially
1786 * crashing in __ksize() due to accessing invalid metadata.
1788 if (unlikely(objp
== ZERO_SIZE_PTR
) || !__kasan_check_read(objp
, 1))
1791 size
= __ksize(objp
);
1793 * We assume that ksize callers could use whole allocated area,
1794 * so we need to unpoison this area.
1796 kasan_unpoison_shadow(objp
, size
);
1799 EXPORT_SYMBOL(ksize
);
1801 /* Tracepoints definitions. */
1802 EXPORT_TRACEPOINT_SYMBOL(kmalloc
);
1803 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc
);
1804 EXPORT_TRACEPOINT_SYMBOL(kmalloc_node
);
1805 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node
);
1806 EXPORT_TRACEPOINT_SYMBOL(kfree
);
1807 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free
);
1809 int should_failslab(struct kmem_cache
*s
, gfp_t gfpflags
)
1811 if (__should_failslab(s
, gfpflags
))
1815 ALLOW_ERROR_INJECTION(should_failslab
, ERRNO
);