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>
33 enum slab_state slab_state
;
34 LIST_HEAD(slab_caches
);
35 DEFINE_MUTEX(slab_mutex
);
36 struct kmem_cache
*kmem_cache
;
38 #ifdef CONFIG_HARDENED_USERCOPY
39 bool usercopy_fallback __ro_after_init
=
40 IS_ENABLED(CONFIG_HARDENED_USERCOPY_FALLBACK
);
41 module_param(usercopy_fallback
, bool, 0400);
42 MODULE_PARM_DESC(usercopy_fallback
,
43 "WARN instead of reject usercopy whitelist violations");
46 static LIST_HEAD(slab_caches_to_rcu_destroy
);
47 static void slab_caches_to_rcu_destroy_workfn(struct work_struct
*work
);
48 static DECLARE_WORK(slab_caches_to_rcu_destroy_work
,
49 slab_caches_to_rcu_destroy_workfn
);
52 * Set of flags that will prevent slab merging
54 #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
55 SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \
56 SLAB_FAILSLAB | SLAB_KASAN)
58 #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \
59 SLAB_CACHE_DMA32 | SLAB_ACCOUNT)
62 * Merge control. If this is set then no merging of slab caches will occur.
64 static bool slab_nomerge
= !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT
);
66 static int __init
setup_slab_nomerge(char *str
)
73 __setup_param("slub_nomerge", slub_nomerge
, setup_slab_nomerge
, 0);
76 __setup("slab_nomerge", setup_slab_nomerge
);
79 * Determine the size of a slab object
81 unsigned int kmem_cache_size(struct kmem_cache
*s
)
83 return s
->object_size
;
85 EXPORT_SYMBOL(kmem_cache_size
);
87 #ifdef CONFIG_DEBUG_VM
88 static int kmem_cache_sanity_check(const char *name
, unsigned int size
)
90 if (!name
|| in_interrupt() || size
< sizeof(void *) ||
91 size
> KMALLOC_MAX_SIZE
) {
92 pr_err("kmem_cache_create(%s) integrity check failed\n", name
);
96 WARN_ON(strchr(name
, ' ')); /* It confuses parsers */
100 static inline int kmem_cache_sanity_check(const char *name
, unsigned int size
)
106 void __kmem_cache_free_bulk(struct kmem_cache
*s
, size_t nr
, void **p
)
110 for (i
= 0; i
< nr
; i
++) {
112 kmem_cache_free(s
, p
[i
]);
118 int __kmem_cache_alloc_bulk(struct kmem_cache
*s
, gfp_t flags
, size_t nr
,
123 for (i
= 0; i
< nr
; i
++) {
124 void *x
= p
[i
] = kmem_cache_alloc(s
, flags
);
126 __kmem_cache_free_bulk(s
, i
, p
);
133 #ifdef CONFIG_MEMCG_KMEM
135 LIST_HEAD(slab_root_caches
);
136 static DEFINE_SPINLOCK(memcg_kmem_wq_lock
);
138 static void kmemcg_cache_shutdown(struct percpu_ref
*percpu_ref
);
140 void slab_init_memcg_params(struct kmem_cache
*s
)
142 s
->memcg_params
.root_cache
= NULL
;
143 RCU_INIT_POINTER(s
->memcg_params
.memcg_caches
, NULL
);
144 INIT_LIST_HEAD(&s
->memcg_params
.children
);
145 s
->memcg_params
.dying
= false;
148 static int init_memcg_params(struct kmem_cache
*s
,
149 struct kmem_cache
*root_cache
)
151 struct memcg_cache_array
*arr
;
154 int ret
= percpu_ref_init(&s
->memcg_params
.refcnt
,
155 kmemcg_cache_shutdown
,
160 s
->memcg_params
.root_cache
= root_cache
;
161 INIT_LIST_HEAD(&s
->memcg_params
.children_node
);
162 INIT_LIST_HEAD(&s
->memcg_params
.kmem_caches_node
);
166 slab_init_memcg_params(s
);
168 if (!memcg_nr_cache_ids
)
171 arr
= kvzalloc(sizeof(struct memcg_cache_array
) +
172 memcg_nr_cache_ids
* sizeof(void *),
177 RCU_INIT_POINTER(s
->memcg_params
.memcg_caches
, arr
);
181 static void destroy_memcg_params(struct kmem_cache
*s
)
183 if (is_root_cache(s
)) {
184 kvfree(rcu_access_pointer(s
->memcg_params
.memcg_caches
));
186 mem_cgroup_put(s
->memcg_params
.memcg
);
187 WRITE_ONCE(s
->memcg_params
.memcg
, NULL
);
188 percpu_ref_exit(&s
->memcg_params
.refcnt
);
192 static void free_memcg_params(struct rcu_head
*rcu
)
194 struct memcg_cache_array
*old
;
196 old
= container_of(rcu
, struct memcg_cache_array
, rcu
);
200 static int update_memcg_params(struct kmem_cache
*s
, int new_array_size
)
202 struct memcg_cache_array
*old
, *new;
204 new = kvzalloc(sizeof(struct memcg_cache_array
) +
205 new_array_size
* sizeof(void *), GFP_KERNEL
);
209 old
= rcu_dereference_protected(s
->memcg_params
.memcg_caches
,
210 lockdep_is_held(&slab_mutex
));
212 memcpy(new->entries
, old
->entries
,
213 memcg_nr_cache_ids
* sizeof(void *));
215 rcu_assign_pointer(s
->memcg_params
.memcg_caches
, new);
217 call_rcu(&old
->rcu
, free_memcg_params
);
221 int memcg_update_all_caches(int num_memcgs
)
223 struct kmem_cache
*s
;
226 mutex_lock(&slab_mutex
);
227 list_for_each_entry(s
, &slab_root_caches
, root_caches_node
) {
228 ret
= update_memcg_params(s
, num_memcgs
);
230 * Instead of freeing the memory, we'll just leave the caches
231 * up to this point in an updated state.
236 mutex_unlock(&slab_mutex
);
240 void memcg_link_cache(struct kmem_cache
*s
, struct mem_cgroup
*memcg
)
242 if (is_root_cache(s
)) {
243 list_add(&s
->root_caches_node
, &slab_root_caches
);
245 css_get(&memcg
->css
);
246 s
->memcg_params
.memcg
= memcg
;
247 list_add(&s
->memcg_params
.children_node
,
248 &s
->memcg_params
.root_cache
->memcg_params
.children
);
249 list_add(&s
->memcg_params
.kmem_caches_node
,
250 &s
->memcg_params
.memcg
->kmem_caches
);
254 static void memcg_unlink_cache(struct kmem_cache
*s
)
256 if (is_root_cache(s
)) {
257 list_del(&s
->root_caches_node
);
259 list_del(&s
->memcg_params
.children_node
);
260 list_del(&s
->memcg_params
.kmem_caches_node
);
264 static inline int init_memcg_params(struct kmem_cache
*s
,
265 struct kmem_cache
*root_cache
)
270 static inline void destroy_memcg_params(struct kmem_cache
*s
)
274 static inline void memcg_unlink_cache(struct kmem_cache
*s
)
277 #endif /* CONFIG_MEMCG_KMEM */
280 * Figure out what the alignment of the objects will be given a set of
281 * flags, a user specified alignment and the size of the objects.
283 static unsigned int calculate_alignment(slab_flags_t flags
,
284 unsigned int align
, unsigned int size
)
287 * If the user wants hardware cache aligned objects then follow that
288 * suggestion if the object is sufficiently large.
290 * The hardware cache alignment cannot override the specified
291 * alignment though. If that is greater then use it.
293 if (flags
& SLAB_HWCACHE_ALIGN
) {
296 ralign
= cache_line_size();
297 while (size
<= ralign
/ 2)
299 align
= max(align
, ralign
);
302 if (align
< ARCH_SLAB_MINALIGN
)
303 align
= ARCH_SLAB_MINALIGN
;
305 return ALIGN(align
, sizeof(void *));
309 * Find a mergeable slab cache
311 int slab_unmergeable(struct kmem_cache
*s
)
313 if (slab_nomerge
|| (s
->flags
& SLAB_NEVER_MERGE
))
316 if (!is_root_cache(s
))
326 * We may have set a slab to be unmergeable during bootstrap.
331 #ifdef CONFIG_MEMCG_KMEM
333 * Skip the dying kmem_cache.
335 if (s
->memcg_params
.dying
)
342 struct kmem_cache
*find_mergeable(unsigned int size
, unsigned int align
,
343 slab_flags_t flags
, const char *name
, void (*ctor
)(void *))
345 struct kmem_cache
*s
;
353 size
= ALIGN(size
, sizeof(void *));
354 align
= calculate_alignment(flags
, align
, size
);
355 size
= ALIGN(size
, align
);
356 flags
= kmem_cache_flags(size
, flags
, name
, NULL
);
358 if (flags
& SLAB_NEVER_MERGE
)
361 list_for_each_entry_reverse(s
, &slab_root_caches
, root_caches_node
) {
362 if (slab_unmergeable(s
))
368 if ((flags
& SLAB_MERGE_SAME
) != (s
->flags
& SLAB_MERGE_SAME
))
371 * Check if alignment is compatible.
372 * Courtesy of Adrian Drzewiecki
374 if ((s
->size
& ~(align
- 1)) != s
->size
)
377 if (s
->size
- size
>= sizeof(void *))
380 if (IS_ENABLED(CONFIG_SLAB
) && align
&&
381 (align
> s
->align
|| s
->align
% align
))
389 static struct kmem_cache
*create_cache(const char *name
,
390 unsigned int object_size
, unsigned int align
,
391 slab_flags_t flags
, unsigned int useroffset
,
392 unsigned int usersize
, void (*ctor
)(void *),
393 struct mem_cgroup
*memcg
, struct kmem_cache
*root_cache
)
395 struct kmem_cache
*s
;
398 if (WARN_ON(useroffset
+ usersize
> object_size
))
399 useroffset
= usersize
= 0;
402 s
= kmem_cache_zalloc(kmem_cache
, GFP_KERNEL
);
407 s
->size
= s
->object_size
= object_size
;
410 s
->useroffset
= useroffset
;
411 s
->usersize
= usersize
;
413 err
= init_memcg_params(s
, root_cache
);
417 err
= __kmem_cache_create(s
, flags
);
422 list_add(&s
->list
, &slab_caches
);
423 memcg_link_cache(s
, memcg
);
430 destroy_memcg_params(s
);
431 kmem_cache_free(kmem_cache
, s
);
436 * kmem_cache_create_usercopy - Create a cache with a region suitable
437 * for copying to userspace
438 * @name: A string which is used in /proc/slabinfo to identify this cache.
439 * @size: The size of objects to be created in this cache.
440 * @align: The required alignment for the objects.
442 * @useroffset: Usercopy region offset
443 * @usersize: Usercopy region size
444 * @ctor: A constructor for the objects.
446 * Cannot be called within a interrupt, but can be interrupted.
447 * The @ctor is run when new pages are allocated by the cache.
451 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
452 * to catch references to uninitialised memory.
454 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
455 * for buffer overruns.
457 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
458 * cacheline. This can be beneficial if you're counting cycles as closely
461 * Return: a pointer to the cache on success, NULL on failure.
464 kmem_cache_create_usercopy(const char *name
,
465 unsigned int size
, unsigned int align
,
467 unsigned int useroffset
, unsigned int usersize
,
468 void (*ctor
)(void *))
470 struct kmem_cache
*s
= NULL
;
471 const char *cache_name
;
476 memcg_get_cache_ids();
478 mutex_lock(&slab_mutex
);
480 err
= kmem_cache_sanity_check(name
, size
);
485 /* Refuse requests with allocator specific flags */
486 if (flags
& ~SLAB_FLAGS_PERMITTED
) {
492 * Some allocators will constraint the set of valid flags to a subset
493 * of all flags. We expect them to define CACHE_CREATE_MASK in this
494 * case, and we'll just provide them with a sanitized version of the
497 flags
&= CACHE_CREATE_MASK
;
499 /* Fail closed on bad usersize of useroffset values. */
500 if (WARN_ON(!usersize
&& useroffset
) ||
501 WARN_ON(size
< usersize
|| size
- usersize
< useroffset
))
502 usersize
= useroffset
= 0;
505 s
= __kmem_cache_alias(name
, size
, align
, flags
, ctor
);
509 cache_name
= kstrdup_const(name
, GFP_KERNEL
);
515 s
= create_cache(cache_name
, size
,
516 calculate_alignment(flags
, align
, size
),
517 flags
, useroffset
, usersize
, ctor
, NULL
, NULL
);
520 kfree_const(cache_name
);
524 mutex_unlock(&slab_mutex
);
526 memcg_put_cache_ids();
531 if (flags
& SLAB_PANIC
)
532 panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n",
535 pr_warn("kmem_cache_create(%s) failed with error %d\n",
543 EXPORT_SYMBOL(kmem_cache_create_usercopy
);
546 * kmem_cache_create - Create a cache.
547 * @name: A string which is used in /proc/slabinfo to identify this cache.
548 * @size: The size of objects to be created in this cache.
549 * @align: The required alignment for the objects.
551 * @ctor: A constructor for the objects.
553 * Cannot be called within a interrupt, but can be interrupted.
554 * The @ctor is run when new pages are allocated by the cache.
558 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
559 * to catch references to uninitialised memory.
561 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
562 * for buffer overruns.
564 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
565 * cacheline. This can be beneficial if you're counting cycles as closely
568 * Return: a pointer to the cache on success, NULL on failure.
571 kmem_cache_create(const char *name
, unsigned int size
, unsigned int align
,
572 slab_flags_t flags
, void (*ctor
)(void *))
574 return kmem_cache_create_usercopy(name
, size
, align
, flags
, 0, 0,
577 EXPORT_SYMBOL(kmem_cache_create
);
579 static void slab_caches_to_rcu_destroy_workfn(struct work_struct
*work
)
581 LIST_HEAD(to_destroy
);
582 struct kmem_cache
*s
, *s2
;
585 * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the
586 * @slab_caches_to_rcu_destroy list. The slab pages are freed
587 * through RCU and and the associated kmem_cache are dereferenced
588 * while freeing the pages, so the kmem_caches should be freed only
589 * after the pending RCU operations are finished. As rcu_barrier()
590 * is a pretty slow operation, we batch all pending destructions
593 mutex_lock(&slab_mutex
);
594 list_splice_init(&slab_caches_to_rcu_destroy
, &to_destroy
);
595 mutex_unlock(&slab_mutex
);
597 if (list_empty(&to_destroy
))
602 list_for_each_entry_safe(s
, s2
, &to_destroy
, list
) {
603 #ifdef SLAB_SUPPORTS_SYSFS
604 sysfs_slab_release(s
);
606 slab_kmem_cache_release(s
);
611 static int shutdown_cache(struct kmem_cache
*s
)
613 /* free asan quarantined objects */
614 kasan_cache_shutdown(s
);
616 if (__kmem_cache_shutdown(s
) != 0)
619 memcg_unlink_cache(s
);
622 if (s
->flags
& SLAB_TYPESAFE_BY_RCU
) {
623 #ifdef SLAB_SUPPORTS_SYSFS
624 sysfs_slab_unlink(s
);
626 list_add_tail(&s
->list
, &slab_caches_to_rcu_destroy
);
627 schedule_work(&slab_caches_to_rcu_destroy_work
);
629 #ifdef SLAB_SUPPORTS_SYSFS
630 sysfs_slab_unlink(s
);
631 sysfs_slab_release(s
);
633 slab_kmem_cache_release(s
);
640 #ifdef CONFIG_MEMCG_KMEM
642 * memcg_create_kmem_cache - Create a cache for a memory cgroup.
643 * @memcg: The memory cgroup the new cache is for.
644 * @root_cache: The parent of the new cache.
646 * This function attempts to create a kmem cache that will serve allocation
647 * requests going from @memcg to @root_cache. The new cache inherits properties
650 void memcg_create_kmem_cache(struct mem_cgroup
*memcg
,
651 struct kmem_cache
*root_cache
)
653 static char memcg_name_buf
[NAME_MAX
+ 1]; /* protected by slab_mutex */
654 struct cgroup_subsys_state
*css
= &memcg
->css
;
655 struct memcg_cache_array
*arr
;
656 struct kmem_cache
*s
= NULL
;
663 mutex_lock(&slab_mutex
);
666 * The memory cgroup could have been offlined while the cache
667 * creation work was pending.
669 if (memcg
->kmem_state
!= KMEM_ONLINE
)
672 idx
= memcg_cache_id(memcg
);
673 arr
= rcu_dereference_protected(root_cache
->memcg_params
.memcg_caches
,
674 lockdep_is_held(&slab_mutex
));
677 * Since per-memcg caches are created asynchronously on first
678 * allocation (see memcg_kmem_get_cache()), several threads can try to
679 * create the same cache, but only one of them may succeed.
681 if (arr
->entries
[idx
])
684 cgroup_name(css
->cgroup
, memcg_name_buf
, sizeof(memcg_name_buf
));
685 cache_name
= kasprintf(GFP_KERNEL
, "%s(%llu:%s)", root_cache
->name
,
686 css
->serial_nr
, memcg_name_buf
);
690 s
= create_cache(cache_name
, root_cache
->object_size
,
692 root_cache
->flags
& CACHE_CREATE_MASK
,
693 root_cache
->useroffset
, root_cache
->usersize
,
694 root_cache
->ctor
, memcg
, root_cache
);
696 * If we could not create a memcg cache, do not complain, because
697 * that's not critical at all as we can always proceed with the root
706 * Since readers won't lock (see memcg_kmem_get_cache()), we need a
707 * barrier here to ensure nobody will see the kmem_cache partially
711 arr
->entries
[idx
] = s
;
714 mutex_unlock(&slab_mutex
);
720 static void kmemcg_workfn(struct work_struct
*work
)
722 struct kmem_cache
*s
= container_of(work
, struct kmem_cache
,
728 mutex_lock(&slab_mutex
);
729 s
->memcg_params
.work_fn(s
);
730 mutex_unlock(&slab_mutex
);
736 static void kmemcg_rcufn(struct rcu_head
*head
)
738 struct kmem_cache
*s
= container_of(head
, struct kmem_cache
,
739 memcg_params
.rcu_head
);
742 * We need to grab blocking locks. Bounce to ->work. The
743 * work item shares the space with the RCU head and can't be
744 * initialized earlier.
746 INIT_WORK(&s
->memcg_params
.work
, kmemcg_workfn
);
747 queue_work(memcg_kmem_cache_wq
, &s
->memcg_params
.work
);
750 static void kmemcg_cache_shutdown_fn(struct kmem_cache
*s
)
752 WARN_ON(shutdown_cache(s
));
755 static void kmemcg_cache_shutdown(struct percpu_ref
*percpu_ref
)
757 struct kmem_cache
*s
= container_of(percpu_ref
, struct kmem_cache
,
758 memcg_params
.refcnt
);
761 spin_lock_irqsave(&memcg_kmem_wq_lock
, flags
);
762 if (s
->memcg_params
.root_cache
->memcg_params
.dying
)
765 s
->memcg_params
.work_fn
= kmemcg_cache_shutdown_fn
;
766 INIT_WORK(&s
->memcg_params
.work
, kmemcg_workfn
);
767 queue_work(memcg_kmem_cache_wq
, &s
->memcg_params
.work
);
770 spin_unlock_irqrestore(&memcg_kmem_wq_lock
, flags
);
773 static void kmemcg_cache_deactivate_after_rcu(struct kmem_cache
*s
)
775 __kmemcg_cache_deactivate_after_rcu(s
);
776 percpu_ref_kill(&s
->memcg_params
.refcnt
);
779 static void kmemcg_cache_deactivate(struct kmem_cache
*s
)
781 if (WARN_ON_ONCE(is_root_cache(s
)))
784 __kmemcg_cache_deactivate(s
);
785 s
->flags
|= SLAB_DEACTIVATED
;
788 * memcg_kmem_wq_lock is used to synchronize memcg_params.dying
789 * flag and make sure that no new kmem_cache deactivation tasks
790 * are queued (see flush_memcg_workqueue() ).
792 spin_lock_irq(&memcg_kmem_wq_lock
);
793 if (s
->memcg_params
.root_cache
->memcg_params
.dying
)
796 s
->memcg_params
.work_fn
= kmemcg_cache_deactivate_after_rcu
;
797 call_rcu(&s
->memcg_params
.rcu_head
, kmemcg_rcufn
);
799 spin_unlock_irq(&memcg_kmem_wq_lock
);
802 void memcg_deactivate_kmem_caches(struct mem_cgroup
*memcg
,
803 struct mem_cgroup
*parent
)
806 struct memcg_cache_array
*arr
;
807 struct kmem_cache
*s
, *c
;
808 unsigned int nr_reparented
;
810 idx
= memcg_cache_id(memcg
);
815 mutex_lock(&slab_mutex
);
816 list_for_each_entry(s
, &slab_root_caches
, root_caches_node
) {
817 arr
= rcu_dereference_protected(s
->memcg_params
.memcg_caches
,
818 lockdep_is_held(&slab_mutex
));
819 c
= arr
->entries
[idx
];
823 kmemcg_cache_deactivate(c
);
824 arr
->entries
[idx
] = NULL
;
827 list_for_each_entry(s
, &memcg
->kmem_caches
,
828 memcg_params
.kmem_caches_node
) {
829 WRITE_ONCE(s
->memcg_params
.memcg
, parent
);
830 css_put(&memcg
->css
);
834 list_splice_init(&memcg
->kmem_caches
,
835 &parent
->kmem_caches
);
836 css_get_many(&parent
->css
, nr_reparented
);
838 mutex_unlock(&slab_mutex
);
844 static int shutdown_memcg_caches(struct kmem_cache
*s
)
846 struct memcg_cache_array
*arr
;
847 struct kmem_cache
*c
, *c2
;
851 BUG_ON(!is_root_cache(s
));
854 * First, shutdown active caches, i.e. caches that belong to online
857 arr
= rcu_dereference_protected(s
->memcg_params
.memcg_caches
,
858 lockdep_is_held(&slab_mutex
));
859 for_each_memcg_cache_index(i
) {
863 if (shutdown_cache(c
))
865 * The cache still has objects. Move it to a temporary
866 * list so as not to try to destroy it for a second
867 * time while iterating over inactive caches below.
869 list_move(&c
->memcg_params
.children_node
, &busy
);
872 * The cache is empty and will be destroyed soon. Clear
873 * the pointer to it in the memcg_caches array so that
874 * it will never be accessed even if the root cache
877 arr
->entries
[i
] = NULL
;
881 * Second, shutdown all caches left from memory cgroups that are now
884 list_for_each_entry_safe(c
, c2
, &s
->memcg_params
.children
,
885 memcg_params
.children_node
)
888 list_splice(&busy
, &s
->memcg_params
.children
);
891 * A cache being destroyed must be empty. In particular, this means
892 * that all per memcg caches attached to it must be empty too.
894 if (!list_empty(&s
->memcg_params
.children
))
899 static void memcg_set_kmem_cache_dying(struct kmem_cache
*s
)
901 spin_lock_irq(&memcg_kmem_wq_lock
);
902 s
->memcg_params
.dying
= true;
903 spin_unlock_irq(&memcg_kmem_wq_lock
);
906 static void flush_memcg_workqueue(struct kmem_cache
*s
)
909 * SLAB and SLUB deactivate the kmem_caches through call_rcu. Make
910 * sure all registered rcu callbacks have been invoked.
915 * SLAB and SLUB create memcg kmem_caches through workqueue and SLUB
916 * deactivates the memcg kmem_caches through workqueue. Make sure all
917 * previous workitems on workqueue are processed.
919 if (likely(memcg_kmem_cache_wq
))
920 flush_workqueue(memcg_kmem_cache_wq
);
923 * If we're racing with children kmem_cache deactivation, it might
924 * take another rcu grace period to complete their destruction.
925 * At this moment the corresponding percpu_ref_kill() call should be
926 * done, but it might take another rcu grace period to complete
927 * switching to the atomic mode.
928 * Please, note that we check without grabbing the slab_mutex. It's safe
929 * because at this moment the children list can't grow.
931 if (!list_empty(&s
->memcg_params
.children
))
935 static inline int shutdown_memcg_caches(struct kmem_cache
*s
)
939 #endif /* CONFIG_MEMCG_KMEM */
941 void slab_kmem_cache_release(struct kmem_cache
*s
)
943 __kmem_cache_release(s
);
944 destroy_memcg_params(s
);
945 kfree_const(s
->name
);
946 kmem_cache_free(kmem_cache
, s
);
949 void kmem_cache_destroy(struct kmem_cache
*s
)
959 mutex_lock(&slab_mutex
);
965 #ifdef CONFIG_MEMCG_KMEM
966 memcg_set_kmem_cache_dying(s
);
968 mutex_unlock(&slab_mutex
);
973 flush_memcg_workqueue(s
);
978 mutex_lock(&slab_mutex
);
981 err
= shutdown_memcg_caches(s
);
983 err
= shutdown_cache(s
);
986 pr_err("kmem_cache_destroy %s: Slab cache still has objects\n",
991 mutex_unlock(&slab_mutex
);
996 EXPORT_SYMBOL(kmem_cache_destroy
);
999 * kmem_cache_shrink - Shrink a cache.
1000 * @cachep: The cache to shrink.
1002 * Releases as many slabs as possible for a cache.
1003 * To help debugging, a zero exit status indicates all slabs were released.
1005 * Return: %0 if all slabs were released, non-zero otherwise
1007 int kmem_cache_shrink(struct kmem_cache
*cachep
)
1013 kasan_cache_shrink(cachep
);
1014 ret
= __kmem_cache_shrink(cachep
);
1019 EXPORT_SYMBOL(kmem_cache_shrink
);
1022 * kmem_cache_shrink_all - shrink a cache and all memcg caches for root cache
1023 * @s: The cache pointer
1025 void kmem_cache_shrink_all(struct kmem_cache
*s
)
1027 struct kmem_cache
*c
;
1029 if (!IS_ENABLED(CONFIG_MEMCG_KMEM
) || !is_root_cache(s
)) {
1030 kmem_cache_shrink(s
);
1036 kasan_cache_shrink(s
);
1037 __kmem_cache_shrink(s
);
1040 * We have to take the slab_mutex to protect from the memcg list
1043 mutex_lock(&slab_mutex
);
1044 for_each_memcg_cache(c
, s
) {
1046 * Don't need to shrink deactivated memcg caches.
1048 if (s
->flags
& SLAB_DEACTIVATED
)
1050 kasan_cache_shrink(c
);
1051 __kmem_cache_shrink(c
);
1053 mutex_unlock(&slab_mutex
);
1058 bool slab_is_available(void)
1060 return slab_state
>= UP
;
1064 /* Create a cache during boot when no slab services are available yet */
1065 void __init
create_boot_cache(struct kmem_cache
*s
, const char *name
,
1066 unsigned int size
, slab_flags_t flags
,
1067 unsigned int useroffset
, unsigned int usersize
)
1070 unsigned int align
= ARCH_KMALLOC_MINALIGN
;
1073 s
->size
= s
->object_size
= size
;
1076 * For power of two sizes, guarantee natural alignment for kmalloc
1077 * caches, regardless of SL*B debugging options.
1079 if (is_power_of_2(size
))
1080 align
= max(align
, size
);
1081 s
->align
= calculate_alignment(flags
, align
, size
);
1083 s
->useroffset
= useroffset
;
1084 s
->usersize
= usersize
;
1086 slab_init_memcg_params(s
);
1088 err
= __kmem_cache_create(s
, flags
);
1091 panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n",
1094 s
->refcount
= -1; /* Exempt from merging for now */
1097 struct kmem_cache
*__init
create_kmalloc_cache(const char *name
,
1098 unsigned int size
, slab_flags_t flags
,
1099 unsigned int useroffset
, unsigned int usersize
)
1101 struct kmem_cache
*s
= kmem_cache_zalloc(kmem_cache
, GFP_NOWAIT
);
1104 panic("Out of memory when creating slab %s\n", name
);
1106 create_boot_cache(s
, name
, size
, flags
, useroffset
, usersize
);
1107 list_add(&s
->list
, &slab_caches
);
1108 memcg_link_cache(s
, NULL
);
1114 kmalloc_caches
[NR_KMALLOC_TYPES
][KMALLOC_SHIFT_HIGH
+ 1] __ro_after_init
=
1115 { /* initialization for https://bugs.llvm.org/show_bug.cgi?id=42570 */ };
1116 EXPORT_SYMBOL(kmalloc_caches
);
1119 * Conversion table for small slabs sizes / 8 to the index in the
1120 * kmalloc array. This is necessary for slabs < 192 since we have non power
1121 * of two cache sizes there. The size of larger slabs can be determined using
1124 static u8 size_index
[24] __ro_after_init
= {
1151 static inline unsigned int size_index_elem(unsigned int bytes
)
1153 return (bytes
- 1) / 8;
1157 * Find the kmem_cache structure that serves a given size of
1160 struct kmem_cache
*kmalloc_slab(size_t size
, gfp_t flags
)
1166 return ZERO_SIZE_PTR
;
1168 index
= size_index
[size_index_elem(size
)];
1170 if (WARN_ON_ONCE(size
> KMALLOC_MAX_CACHE_SIZE
))
1172 index
= fls(size
- 1);
1175 return kmalloc_caches
[kmalloc_type(flags
)][index
];
1178 #ifdef CONFIG_ZONE_DMA
1179 #define INIT_KMALLOC_INFO(__size, __short_size) \
1181 .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \
1182 .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #__short_size, \
1183 .name[KMALLOC_DMA] = "dma-kmalloc-" #__short_size, \
1187 #define INIT_KMALLOC_INFO(__size, __short_size) \
1189 .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \
1190 .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #__short_size, \
1196 * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
1197 * kmalloc_index() supports up to 2^26=64MB, so the final entry of the table is
1200 const struct kmalloc_info_struct kmalloc_info
[] __initconst
= {
1201 INIT_KMALLOC_INFO(0, 0),
1202 INIT_KMALLOC_INFO(96, 96),
1203 INIT_KMALLOC_INFO(192, 192),
1204 INIT_KMALLOC_INFO(8, 8),
1205 INIT_KMALLOC_INFO(16, 16),
1206 INIT_KMALLOC_INFO(32, 32),
1207 INIT_KMALLOC_INFO(64, 64),
1208 INIT_KMALLOC_INFO(128, 128),
1209 INIT_KMALLOC_INFO(256, 256),
1210 INIT_KMALLOC_INFO(512, 512),
1211 INIT_KMALLOC_INFO(1024, 1k
),
1212 INIT_KMALLOC_INFO(2048, 2k
),
1213 INIT_KMALLOC_INFO(4096, 4k
),
1214 INIT_KMALLOC_INFO(8192, 8k
),
1215 INIT_KMALLOC_INFO(16384, 16k
),
1216 INIT_KMALLOC_INFO(32768, 32k
),
1217 INIT_KMALLOC_INFO(65536, 64k
),
1218 INIT_KMALLOC_INFO(131072, 128k
),
1219 INIT_KMALLOC_INFO(262144, 256k
),
1220 INIT_KMALLOC_INFO(524288, 512k
),
1221 INIT_KMALLOC_INFO(1048576, 1M
),
1222 INIT_KMALLOC_INFO(2097152, 2M
),
1223 INIT_KMALLOC_INFO(4194304, 4M
),
1224 INIT_KMALLOC_INFO(8388608, 8M
),
1225 INIT_KMALLOC_INFO(16777216, 16M
),
1226 INIT_KMALLOC_INFO(33554432, 32M
),
1227 INIT_KMALLOC_INFO(67108864, 64M
)
1231 * Patch up the size_index table if we have strange large alignment
1232 * requirements for the kmalloc array. This is only the case for
1233 * MIPS it seems. The standard arches will not generate any code here.
1235 * Largest permitted alignment is 256 bytes due to the way we
1236 * handle the index determination for the smaller caches.
1238 * Make sure that nothing crazy happens if someone starts tinkering
1239 * around with ARCH_KMALLOC_MINALIGN
1241 void __init
setup_kmalloc_cache_index_table(void)
1245 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 256 ||
1246 (KMALLOC_MIN_SIZE
& (KMALLOC_MIN_SIZE
- 1)));
1248 for (i
= 8; i
< KMALLOC_MIN_SIZE
; i
+= 8) {
1249 unsigned int elem
= size_index_elem(i
);
1251 if (elem
>= ARRAY_SIZE(size_index
))
1253 size_index
[elem
] = KMALLOC_SHIFT_LOW
;
1256 if (KMALLOC_MIN_SIZE
>= 64) {
1258 * The 96 byte size cache is not used if the alignment
1261 for (i
= 64 + 8; i
<= 96; i
+= 8)
1262 size_index
[size_index_elem(i
)] = 7;
1266 if (KMALLOC_MIN_SIZE
>= 128) {
1268 * The 192 byte sized cache is not used if the alignment
1269 * is 128 byte. Redirect kmalloc to use the 256 byte cache
1272 for (i
= 128 + 8; i
<= 192; i
+= 8)
1273 size_index
[size_index_elem(i
)] = 8;
1278 new_kmalloc_cache(int idx
, enum kmalloc_cache_type type
, slab_flags_t flags
)
1280 if (type
== KMALLOC_RECLAIM
)
1281 flags
|= SLAB_RECLAIM_ACCOUNT
;
1283 kmalloc_caches
[type
][idx
] = create_kmalloc_cache(
1284 kmalloc_info
[idx
].name
[type
],
1285 kmalloc_info
[idx
].size
, flags
, 0,
1286 kmalloc_info
[idx
].size
);
1290 * Create the kmalloc array. Some of the regular kmalloc arrays
1291 * may already have been created because they were needed to
1292 * enable allocations for slab creation.
1294 void __init
create_kmalloc_caches(slab_flags_t flags
)
1297 enum kmalloc_cache_type type
;
1299 for (type
= KMALLOC_NORMAL
; type
<= KMALLOC_RECLAIM
; type
++) {
1300 for (i
= KMALLOC_SHIFT_LOW
; i
<= KMALLOC_SHIFT_HIGH
; i
++) {
1301 if (!kmalloc_caches
[type
][i
])
1302 new_kmalloc_cache(i
, type
, flags
);
1305 * Caches that are not of the two-to-the-power-of size.
1306 * These have to be created immediately after the
1307 * earlier power of two caches
1309 if (KMALLOC_MIN_SIZE
<= 32 && i
== 6 &&
1310 !kmalloc_caches
[type
][1])
1311 new_kmalloc_cache(1, type
, flags
);
1312 if (KMALLOC_MIN_SIZE
<= 64 && i
== 7 &&
1313 !kmalloc_caches
[type
][2])
1314 new_kmalloc_cache(2, type
, flags
);
1318 /* Kmalloc array is now usable */
1321 #ifdef CONFIG_ZONE_DMA
1322 for (i
= 0; i
<= KMALLOC_SHIFT_HIGH
; i
++) {
1323 struct kmem_cache
*s
= kmalloc_caches
[KMALLOC_NORMAL
][i
];
1326 kmalloc_caches
[KMALLOC_DMA
][i
] = create_kmalloc_cache(
1327 kmalloc_info
[i
].name
[KMALLOC_DMA
],
1328 kmalloc_info
[i
].size
,
1329 SLAB_CACHE_DMA
| flags
, 0,
1330 kmalloc_info
[i
].size
);
1335 #endif /* !CONFIG_SLOB */
1337 gfp_t
kmalloc_fix_flags(gfp_t flags
)
1339 gfp_t invalid_mask
= flags
& GFP_SLAB_BUG_MASK
;
1341 flags
&= ~GFP_SLAB_BUG_MASK
;
1342 pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
1343 invalid_mask
, &invalid_mask
, flags
, &flags
);
1350 * To avoid unnecessary overhead, we pass through large allocation requests
1351 * directly to the page allocator. We use __GFP_COMP, because we will need to
1352 * know the allocation order to free the pages properly in kfree.
1354 void *kmalloc_order(size_t size
, gfp_t flags
, unsigned int order
)
1359 if (unlikely(flags
& GFP_SLAB_BUG_MASK
))
1360 flags
= kmalloc_fix_flags(flags
);
1362 flags
|= __GFP_COMP
;
1363 page
= alloc_pages(flags
, order
);
1365 ret
= page_address(page
);
1366 mod_node_page_state(page_pgdat(page
), NR_SLAB_UNRECLAIMABLE_B
,
1367 PAGE_SIZE
<< order
);
1369 ret
= kasan_kmalloc_large(ret
, size
, flags
);
1370 /* As ret might get tagged, call kmemleak hook after KASAN. */
1371 kmemleak_alloc(ret
, size
, 1, flags
);
1374 EXPORT_SYMBOL(kmalloc_order
);
1376 #ifdef CONFIG_TRACING
1377 void *kmalloc_order_trace(size_t size
, gfp_t flags
, unsigned int order
)
1379 void *ret
= kmalloc_order(size
, flags
, order
);
1380 trace_kmalloc(_RET_IP_
, ret
, size
, PAGE_SIZE
<< order
, flags
);
1383 EXPORT_SYMBOL(kmalloc_order_trace
);
1386 #ifdef CONFIG_SLAB_FREELIST_RANDOM
1387 /* Randomize a generic freelist */
1388 static void freelist_randomize(struct rnd_state
*state
, unsigned int *list
,
1394 for (i
= 0; i
< count
; i
++)
1397 /* Fisher-Yates shuffle */
1398 for (i
= count
- 1; i
> 0; i
--) {
1399 rand
= prandom_u32_state(state
);
1401 swap(list
[i
], list
[rand
]);
1405 /* Create a random sequence per cache */
1406 int cache_random_seq_create(struct kmem_cache
*cachep
, unsigned int count
,
1409 struct rnd_state state
;
1411 if (count
< 2 || cachep
->random_seq
)
1414 cachep
->random_seq
= kcalloc(count
, sizeof(unsigned int), gfp
);
1415 if (!cachep
->random_seq
)
1418 /* Get best entropy at this stage of boot */
1419 prandom_seed_state(&state
, get_random_long());
1421 freelist_randomize(&state
, cachep
->random_seq
, count
);
1425 /* Destroy the per-cache random freelist sequence */
1426 void cache_random_seq_destroy(struct kmem_cache
*cachep
)
1428 kfree(cachep
->random_seq
);
1429 cachep
->random_seq
= NULL
;
1431 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
1433 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
1435 #define SLABINFO_RIGHTS (0600)
1437 #define SLABINFO_RIGHTS (0400)
1440 static void print_slabinfo_header(struct seq_file
*m
)
1443 * Output format version, so at least we can change it
1444 * without _too_ many complaints.
1446 #ifdef CONFIG_DEBUG_SLAB
1447 seq_puts(m
, "slabinfo - version: 2.1 (statistics)\n");
1449 seq_puts(m
, "slabinfo - version: 2.1\n");
1451 seq_puts(m
, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
1452 seq_puts(m
, " : tunables <limit> <batchcount> <sharedfactor>");
1453 seq_puts(m
, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
1454 #ifdef CONFIG_DEBUG_SLAB
1455 seq_puts(m
, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
1456 seq_puts(m
, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
1461 void *slab_start(struct seq_file
*m
, loff_t
*pos
)
1463 mutex_lock(&slab_mutex
);
1464 return seq_list_start(&slab_root_caches
, *pos
);
1467 void *slab_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
1469 return seq_list_next(p
, &slab_root_caches
, pos
);
1472 void slab_stop(struct seq_file
*m
, void *p
)
1474 mutex_unlock(&slab_mutex
);
1478 memcg_accumulate_slabinfo(struct kmem_cache
*s
, struct slabinfo
*info
)
1480 struct kmem_cache
*c
;
1481 struct slabinfo sinfo
;
1483 if (!is_root_cache(s
))
1486 for_each_memcg_cache(c
, s
) {
1487 memset(&sinfo
, 0, sizeof(sinfo
));
1488 get_slabinfo(c
, &sinfo
);
1490 info
->active_slabs
+= sinfo
.active_slabs
;
1491 info
->num_slabs
+= sinfo
.num_slabs
;
1492 info
->shared_avail
+= sinfo
.shared_avail
;
1493 info
->active_objs
+= sinfo
.active_objs
;
1494 info
->num_objs
+= sinfo
.num_objs
;
1498 static void cache_show(struct kmem_cache
*s
, struct seq_file
*m
)
1500 struct slabinfo sinfo
;
1502 memset(&sinfo
, 0, sizeof(sinfo
));
1503 get_slabinfo(s
, &sinfo
);
1505 memcg_accumulate_slabinfo(s
, &sinfo
);
1507 seq_printf(m
, "%-17s %6lu %6lu %6u %4u %4d",
1508 cache_name(s
), sinfo
.active_objs
, sinfo
.num_objs
, s
->size
,
1509 sinfo
.objects_per_slab
, (1 << sinfo
.cache_order
));
1511 seq_printf(m
, " : tunables %4u %4u %4u",
1512 sinfo
.limit
, sinfo
.batchcount
, sinfo
.shared
);
1513 seq_printf(m
, " : slabdata %6lu %6lu %6lu",
1514 sinfo
.active_slabs
, sinfo
.num_slabs
, sinfo
.shared_avail
);
1515 slabinfo_show_stats(m
, s
);
1519 static int slab_show(struct seq_file
*m
, void *p
)
1521 struct kmem_cache
*s
= list_entry(p
, struct kmem_cache
, root_caches_node
);
1523 if (p
== slab_root_caches
.next
)
1524 print_slabinfo_header(m
);
1529 void dump_unreclaimable_slab(void)
1531 struct kmem_cache
*s
, *s2
;
1532 struct slabinfo sinfo
;
1535 * Here acquiring slab_mutex is risky since we don't prefer to get
1536 * sleep in oom path. But, without mutex hold, it may introduce a
1538 * Use mutex_trylock to protect the list traverse, dump nothing
1539 * without acquiring the mutex.
1541 if (!mutex_trylock(&slab_mutex
)) {
1542 pr_warn("excessive unreclaimable slab but cannot dump stats\n");
1546 pr_info("Unreclaimable slab info:\n");
1547 pr_info("Name Used Total\n");
1549 list_for_each_entry_safe(s
, s2
, &slab_caches
, list
) {
1550 if (!is_root_cache(s
) || (s
->flags
& SLAB_RECLAIM_ACCOUNT
))
1553 get_slabinfo(s
, &sinfo
);
1555 if (sinfo
.num_objs
> 0)
1556 pr_info("%-17s %10luKB %10luKB\n", cache_name(s
),
1557 (sinfo
.active_objs
* s
->size
) / 1024,
1558 (sinfo
.num_objs
* s
->size
) / 1024);
1560 mutex_unlock(&slab_mutex
);
1563 #if defined(CONFIG_MEMCG_KMEM)
1564 void *memcg_slab_start(struct seq_file
*m
, loff_t
*pos
)
1566 struct mem_cgroup
*memcg
= mem_cgroup_from_seq(m
);
1568 mutex_lock(&slab_mutex
);
1569 return seq_list_start(&memcg
->kmem_caches
, *pos
);
1572 void *memcg_slab_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
1574 struct mem_cgroup
*memcg
= mem_cgroup_from_seq(m
);
1576 return seq_list_next(p
, &memcg
->kmem_caches
, pos
);
1579 void memcg_slab_stop(struct seq_file
*m
, void *p
)
1581 mutex_unlock(&slab_mutex
);
1584 int memcg_slab_show(struct seq_file
*m
, void *p
)
1586 struct kmem_cache
*s
= list_entry(p
, struct kmem_cache
,
1587 memcg_params
.kmem_caches_node
);
1588 struct mem_cgroup
*memcg
= mem_cgroup_from_seq(m
);
1590 if (p
== memcg
->kmem_caches
.next
)
1591 print_slabinfo_header(m
);
1598 * slabinfo_op - iterator that generates /proc/slabinfo
1607 * num-pages-per-slab
1608 * + further values on SMP and with statistics enabled
1610 static const struct seq_operations slabinfo_op
= {
1611 .start
= slab_start
,
1617 static int slabinfo_open(struct inode
*inode
, struct file
*file
)
1619 return seq_open(file
, &slabinfo_op
);
1622 static const struct proc_ops slabinfo_proc_ops
= {
1623 .proc_flags
= PROC_ENTRY_PERMANENT
,
1624 .proc_open
= slabinfo_open
,
1625 .proc_read
= seq_read
,
1626 .proc_write
= slabinfo_write
,
1627 .proc_lseek
= seq_lseek
,
1628 .proc_release
= seq_release
,
1631 static int __init
slab_proc_init(void)
1633 proc_create("slabinfo", SLABINFO_RIGHTS
, NULL
, &slabinfo_proc_ops
);
1636 module_init(slab_proc_init
);
1638 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_MEMCG_KMEM)
1640 * Display information about kmem caches that have child memcg caches.
1642 static int memcg_slabinfo_show(struct seq_file
*m
, void *unused
)
1644 struct kmem_cache
*s
, *c
;
1645 struct slabinfo sinfo
;
1647 mutex_lock(&slab_mutex
);
1648 seq_puts(m
, "# <name> <css_id[:dead|deact]> <active_objs> <num_objs>");
1649 seq_puts(m
, " <active_slabs> <num_slabs>\n");
1650 list_for_each_entry(s
, &slab_root_caches
, root_caches_node
) {
1652 * Skip kmem caches that don't have any memcg children.
1654 if (list_empty(&s
->memcg_params
.children
))
1657 memset(&sinfo
, 0, sizeof(sinfo
));
1658 get_slabinfo(s
, &sinfo
);
1659 seq_printf(m
, "%-17s root %6lu %6lu %6lu %6lu\n",
1660 cache_name(s
), sinfo
.active_objs
, sinfo
.num_objs
,
1661 sinfo
.active_slabs
, sinfo
.num_slabs
);
1663 for_each_memcg_cache(c
, s
) {
1664 struct cgroup_subsys_state
*css
;
1667 css
= &c
->memcg_params
.memcg
->css
;
1668 if (!(css
->flags
& CSS_ONLINE
))
1670 else if (c
->flags
& SLAB_DEACTIVATED
)
1673 memset(&sinfo
, 0, sizeof(sinfo
));
1674 get_slabinfo(c
, &sinfo
);
1675 seq_printf(m
, "%-17s %4d%-6s %6lu %6lu %6lu %6lu\n",
1676 cache_name(c
), css
->id
, status
,
1677 sinfo
.active_objs
, sinfo
.num_objs
,
1678 sinfo
.active_slabs
, sinfo
.num_slabs
);
1681 mutex_unlock(&slab_mutex
);
1684 DEFINE_SHOW_ATTRIBUTE(memcg_slabinfo
);
1686 static int __init
memcg_slabinfo_init(void)
1688 debugfs_create_file("memcg_slabinfo", S_IFREG
| S_IRUGO
,
1689 NULL
, NULL
, &memcg_slabinfo_fops
);
1693 late_initcall(memcg_slabinfo_init
);
1694 #endif /* CONFIG_DEBUG_FS && CONFIG_MEMCG_KMEM */
1695 #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */
1697 static __always_inline
void *__do_krealloc(const void *p
, size_t new_size
,
1705 if (ks
>= new_size
) {
1706 p
= kasan_krealloc((void *)p
, new_size
, flags
);
1710 ret
= kmalloc_track_caller(new_size
, flags
);
1718 * krealloc - reallocate memory. The contents will remain unchanged.
1719 * @p: object to reallocate memory for.
1720 * @new_size: how many bytes of memory are required.
1721 * @flags: the type of memory to allocate.
1723 * The contents of the object pointed to are preserved up to the
1724 * lesser of the new and old sizes. If @p is %NULL, krealloc()
1725 * behaves exactly like kmalloc(). If @new_size is 0 and @p is not a
1726 * %NULL pointer, the object pointed to is freed.
1728 * Return: pointer to the allocated memory or %NULL in case of error
1730 void *krealloc(const void *p
, size_t new_size
, gfp_t flags
)
1734 if (unlikely(!new_size
)) {
1736 return ZERO_SIZE_PTR
;
1739 ret
= __do_krealloc(p
, new_size
, flags
);
1740 if (ret
&& kasan_reset_tag(p
) != kasan_reset_tag(ret
))
1745 EXPORT_SYMBOL(krealloc
);
1748 * kfree_sensitive - Clear sensitive information in memory before freeing
1749 * @p: object to free memory of
1751 * The memory of the object @p points to is zeroed before freed.
1752 * If @p is %NULL, kfree_sensitive() does nothing.
1754 * Note: this function zeroes the whole allocated buffer which can be a good
1755 * deal bigger than the requested buffer size passed to kmalloc(). So be
1756 * careful when using this function in performance sensitive code.
1758 void kfree_sensitive(const void *p
)
1761 void *mem
= (void *)p
;
1765 memzero_explicit(mem
, ks
);
1768 EXPORT_SYMBOL(kfree_sensitive
);
1771 * ksize - get the actual amount of memory allocated for a given object
1772 * @objp: Pointer to the object
1774 * kmalloc may internally round up allocations and return more memory
1775 * than requested. ksize() can be used to determine the actual amount of
1776 * memory allocated. The caller may use this additional memory, even though
1777 * a smaller amount of memory was initially specified with the kmalloc call.
1778 * The caller must guarantee that objp points to a valid object previously
1779 * allocated with either kmalloc() or kmem_cache_alloc(). The object
1780 * must not be freed during the duration of the call.
1782 * Return: size of the actual memory used by @objp in bytes
1784 size_t ksize(const void *objp
)
1789 * We need to check that the pointed to object is valid, and only then
1790 * unpoison the shadow memory below. We use __kasan_check_read(), to
1791 * generate a more useful report at the time ksize() is called (rather
1792 * than later where behaviour is undefined due to potential
1793 * use-after-free or double-free).
1795 * If the pointed to memory is invalid we return 0, to avoid users of
1796 * ksize() writing to and potentially corrupting the memory region.
1798 * We want to perform the check before __ksize(), to avoid potentially
1799 * crashing in __ksize() due to accessing invalid metadata.
1801 if (unlikely(ZERO_OR_NULL_PTR(objp
)) || !__kasan_check_read(objp
, 1))
1804 size
= __ksize(objp
);
1806 * We assume that ksize callers could use whole allocated area,
1807 * so we need to unpoison this area.
1809 kasan_unpoison_shadow(objp
, size
);
1812 EXPORT_SYMBOL(ksize
);
1814 /* Tracepoints definitions. */
1815 EXPORT_TRACEPOINT_SYMBOL(kmalloc
);
1816 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc
);
1817 EXPORT_TRACEPOINT_SYMBOL(kmalloc_node
);
1818 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node
);
1819 EXPORT_TRACEPOINT_SYMBOL(kfree
);
1820 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free
);
1822 int should_failslab(struct kmem_cache
*s
, gfp_t gfpflags
)
1824 if (__should_failslab(s
, gfpflags
))
1828 ALLOW_ERROR_INJECTION(should_failslab
, ERRNO
);