xref: /linux/mm/slab_common.c (revision a2b0fe7435faee6f6fbb27409878013bc4727e98)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Slab allocator functions that are independent of the allocator strategy
4  *
5  * (C) 2012 Christoph Lameter <cl@linux.com>
6  */
7 #include <linux/slab.h>
8 
9 #include <linux/mm.h>
10 #include <linux/poison.h>
11 #include <linux/interrupt.h>
12 #include <linux/memory.h>
13 #include <linux/compiler.h>
14 #include <linux/module.h>
15 #include <linux/cpu.h>
16 #include <linux/uaccess.h>
17 #include <linux/seq_file.h>
18 #include <linux/proc_fs.h>
19 #include <asm/cacheflush.h>
20 #include <asm/tlbflush.h>
21 #include <asm/page.h>
22 #include <linux/memcontrol.h>
23 
24 #define CREATE_TRACE_POINTS
25 #include <trace/events/kmem.h>
26 
27 #include "slab.h"
28 
29 enum slab_state slab_state;
30 LIST_HEAD(slab_caches);
31 DEFINE_MUTEX(slab_mutex);
32 struct kmem_cache *kmem_cache;
33 
34 static LIST_HEAD(slab_caches_to_rcu_destroy);
35 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work);
36 static DECLARE_WORK(slab_caches_to_rcu_destroy_work,
37 		    slab_caches_to_rcu_destroy_workfn);
38 
39 /*
40  * Set of flags that will prevent slab merging
41  */
42 #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
43 		SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \
44 		SLAB_FAILSLAB | SLAB_KASAN)
45 
46 #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \
47 			 SLAB_ACCOUNT)
48 
49 /*
50  * Merge control. If this is set then no merging of slab caches will occur.
51  */
52 static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT);
53 
54 static int __init setup_slab_nomerge(char *str)
55 {
56 	slab_nomerge = true;
57 	return 1;
58 }
59 
60 #ifdef CONFIG_SLUB
61 __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0);
62 #endif
63 
64 __setup("slab_nomerge", setup_slab_nomerge);
65 
66 /*
67  * Determine the size of a slab object
68  */
69 unsigned int kmem_cache_size(struct kmem_cache *s)
70 {
71 	return s->object_size;
72 }
73 EXPORT_SYMBOL(kmem_cache_size);
74 
75 #ifdef CONFIG_DEBUG_VM
76 static int kmem_cache_sanity_check(const char *name, size_t size)
77 {
78 	struct kmem_cache *s = NULL;
79 
80 	if (!name || in_interrupt() || size < sizeof(void *) ||
81 		size > KMALLOC_MAX_SIZE) {
82 		pr_err("kmem_cache_create(%s) integrity check failed\n", name);
83 		return -EINVAL;
84 	}
85 
86 	list_for_each_entry(s, &slab_caches, list) {
87 		char tmp;
88 		int res;
89 
90 		/*
91 		 * This happens when the module gets unloaded and doesn't
92 		 * destroy its slab cache and no-one else reuses the vmalloc
93 		 * area of the module.  Print a warning.
94 		 */
95 		res = probe_kernel_address(s->name, tmp);
96 		if (res) {
97 			pr_err("Slab cache with size %d has lost its name\n",
98 			       s->object_size);
99 			continue;
100 		}
101 	}
102 
103 	WARN_ON(strchr(name, ' '));	/* It confuses parsers */
104 	return 0;
105 }
106 #else
107 static inline int kmem_cache_sanity_check(const char *name, size_t size)
108 {
109 	return 0;
110 }
111 #endif
112 
113 void __kmem_cache_free_bulk(struct kmem_cache *s, size_t nr, void **p)
114 {
115 	size_t i;
116 
117 	for (i = 0; i < nr; i++) {
118 		if (s)
119 			kmem_cache_free(s, p[i]);
120 		else
121 			kfree(p[i]);
122 	}
123 }
124 
125 int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t nr,
126 								void **p)
127 {
128 	size_t i;
129 
130 	for (i = 0; i < nr; i++) {
131 		void *x = p[i] = kmem_cache_alloc(s, flags);
132 		if (!x) {
133 			__kmem_cache_free_bulk(s, i, p);
134 			return 0;
135 		}
136 	}
137 	return i;
138 }
139 
140 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
141 
142 LIST_HEAD(slab_root_caches);
143 
144 void slab_init_memcg_params(struct kmem_cache *s)
145 {
146 	s->memcg_params.root_cache = NULL;
147 	RCU_INIT_POINTER(s->memcg_params.memcg_caches, NULL);
148 	INIT_LIST_HEAD(&s->memcg_params.children);
149 }
150 
151 static int init_memcg_params(struct kmem_cache *s,
152 		struct mem_cgroup *memcg, struct kmem_cache *root_cache)
153 {
154 	struct memcg_cache_array *arr;
155 
156 	if (root_cache) {
157 		s->memcg_params.root_cache = root_cache;
158 		s->memcg_params.memcg = memcg;
159 		INIT_LIST_HEAD(&s->memcg_params.children_node);
160 		INIT_LIST_HEAD(&s->memcg_params.kmem_caches_node);
161 		return 0;
162 	}
163 
164 	slab_init_memcg_params(s);
165 
166 	if (!memcg_nr_cache_ids)
167 		return 0;
168 
169 	arr = kvzalloc(sizeof(struct memcg_cache_array) +
170 		       memcg_nr_cache_ids * sizeof(void *),
171 		       GFP_KERNEL);
172 	if (!arr)
173 		return -ENOMEM;
174 
175 	RCU_INIT_POINTER(s->memcg_params.memcg_caches, arr);
176 	return 0;
177 }
178 
179 static void destroy_memcg_params(struct kmem_cache *s)
180 {
181 	if (is_root_cache(s))
182 		kvfree(rcu_access_pointer(s->memcg_params.memcg_caches));
183 }
184 
185 static void free_memcg_params(struct rcu_head *rcu)
186 {
187 	struct memcg_cache_array *old;
188 
189 	old = container_of(rcu, struct memcg_cache_array, rcu);
190 	kvfree(old);
191 }
192 
193 static int update_memcg_params(struct kmem_cache *s, int new_array_size)
194 {
195 	struct memcg_cache_array *old, *new;
196 
197 	new = kvzalloc(sizeof(struct memcg_cache_array) +
198 		       new_array_size * sizeof(void *), GFP_KERNEL);
199 	if (!new)
200 		return -ENOMEM;
201 
202 	old = rcu_dereference_protected(s->memcg_params.memcg_caches,
203 					lockdep_is_held(&slab_mutex));
204 	if (old)
205 		memcpy(new->entries, old->entries,
206 		       memcg_nr_cache_ids * sizeof(void *));
207 
208 	rcu_assign_pointer(s->memcg_params.memcg_caches, new);
209 	if (old)
210 		call_rcu(&old->rcu, free_memcg_params);
211 	return 0;
212 }
213 
214 int memcg_update_all_caches(int num_memcgs)
215 {
216 	struct kmem_cache *s;
217 	int ret = 0;
218 
219 	mutex_lock(&slab_mutex);
220 	list_for_each_entry(s, &slab_root_caches, root_caches_node) {
221 		ret = update_memcg_params(s, num_memcgs);
222 		/*
223 		 * Instead of freeing the memory, we'll just leave the caches
224 		 * up to this point in an updated state.
225 		 */
226 		if (ret)
227 			break;
228 	}
229 	mutex_unlock(&slab_mutex);
230 	return ret;
231 }
232 
233 void memcg_link_cache(struct kmem_cache *s)
234 {
235 	if (is_root_cache(s)) {
236 		list_add(&s->root_caches_node, &slab_root_caches);
237 	} else {
238 		list_add(&s->memcg_params.children_node,
239 			 &s->memcg_params.root_cache->memcg_params.children);
240 		list_add(&s->memcg_params.kmem_caches_node,
241 			 &s->memcg_params.memcg->kmem_caches);
242 	}
243 }
244 
245 static void memcg_unlink_cache(struct kmem_cache *s)
246 {
247 	if (is_root_cache(s)) {
248 		list_del(&s->root_caches_node);
249 	} else {
250 		list_del(&s->memcg_params.children_node);
251 		list_del(&s->memcg_params.kmem_caches_node);
252 	}
253 }
254 #else
255 static inline int init_memcg_params(struct kmem_cache *s,
256 		struct mem_cgroup *memcg, struct kmem_cache *root_cache)
257 {
258 	return 0;
259 }
260 
261 static inline void destroy_memcg_params(struct kmem_cache *s)
262 {
263 }
264 
265 static inline void memcg_unlink_cache(struct kmem_cache *s)
266 {
267 }
268 #endif /* CONFIG_MEMCG && !CONFIG_SLOB */
269 
270 /*
271  * Figure out what the alignment of the objects will be given a set of
272  * flags, a user specified alignment and the size of the objects.
273  */
274 static unsigned long calculate_alignment(unsigned long flags,
275 		unsigned long align, unsigned long size)
276 {
277 	/*
278 	 * If the user wants hardware cache aligned objects then follow that
279 	 * suggestion if the object is sufficiently large.
280 	 *
281 	 * The hardware cache alignment cannot override the specified
282 	 * alignment though. If that is greater then use it.
283 	 */
284 	if (flags & SLAB_HWCACHE_ALIGN) {
285 		unsigned long ralign;
286 
287 		ralign = cache_line_size();
288 		while (size <= ralign / 2)
289 			ralign /= 2;
290 		align = max(align, ralign);
291 	}
292 
293 	if (align < ARCH_SLAB_MINALIGN)
294 		align = ARCH_SLAB_MINALIGN;
295 
296 	return ALIGN(align, sizeof(void *));
297 }
298 
299 /*
300  * Find a mergeable slab cache
301  */
302 int slab_unmergeable(struct kmem_cache *s)
303 {
304 	if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
305 		return 1;
306 
307 	if (!is_root_cache(s))
308 		return 1;
309 
310 	if (s->ctor)
311 		return 1;
312 
313 	/*
314 	 * We may have set a slab to be unmergeable during bootstrap.
315 	 */
316 	if (s->refcount < 0)
317 		return 1;
318 
319 	return 0;
320 }
321 
322 struct kmem_cache *find_mergeable(size_t size, size_t align,
323 		slab_flags_t flags, const char *name, void (*ctor)(void *))
324 {
325 	struct kmem_cache *s;
326 
327 	if (slab_nomerge)
328 		return NULL;
329 
330 	if (ctor)
331 		return NULL;
332 
333 	size = ALIGN(size, sizeof(void *));
334 	align = calculate_alignment(flags, align, size);
335 	size = ALIGN(size, align);
336 	flags = kmem_cache_flags(size, flags, name, NULL);
337 
338 	if (flags & SLAB_NEVER_MERGE)
339 		return NULL;
340 
341 	list_for_each_entry_reverse(s, &slab_root_caches, root_caches_node) {
342 		if (slab_unmergeable(s))
343 			continue;
344 
345 		if (size > s->size)
346 			continue;
347 
348 		if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
349 			continue;
350 		/*
351 		 * Check if alignment is compatible.
352 		 * Courtesy of Adrian Drzewiecki
353 		 */
354 		if ((s->size & ~(align - 1)) != s->size)
355 			continue;
356 
357 		if (s->size - size >= sizeof(void *))
358 			continue;
359 
360 		if (IS_ENABLED(CONFIG_SLAB) && align &&
361 			(align > s->align || s->align % align))
362 			continue;
363 
364 		return s;
365 	}
366 	return NULL;
367 }
368 
369 static struct kmem_cache *create_cache(const char *name,
370 		size_t object_size, size_t size, size_t align,
371 		slab_flags_t flags, void (*ctor)(void *),
372 		struct mem_cgroup *memcg, struct kmem_cache *root_cache)
373 {
374 	struct kmem_cache *s;
375 	int err;
376 
377 	err = -ENOMEM;
378 	s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
379 	if (!s)
380 		goto out;
381 
382 	s->name = name;
383 	s->object_size = object_size;
384 	s->size = size;
385 	s->align = align;
386 	s->ctor = ctor;
387 
388 	err = init_memcg_params(s, memcg, root_cache);
389 	if (err)
390 		goto out_free_cache;
391 
392 	err = __kmem_cache_create(s, flags);
393 	if (err)
394 		goto out_free_cache;
395 
396 	s->refcount = 1;
397 	list_add(&s->list, &slab_caches);
398 	memcg_link_cache(s);
399 out:
400 	if (err)
401 		return ERR_PTR(err);
402 	return s;
403 
404 out_free_cache:
405 	destroy_memcg_params(s);
406 	kmem_cache_free(kmem_cache, s);
407 	goto out;
408 }
409 
410 /*
411  * kmem_cache_create - Create a cache.
412  * @name: A string which is used in /proc/slabinfo to identify this cache.
413  * @size: The size of objects to be created in this cache.
414  * @align: The required alignment for the objects.
415  * @flags: SLAB flags
416  * @ctor: A constructor for the objects.
417  *
418  * Returns a ptr to the cache on success, NULL on failure.
419  * Cannot be called within a interrupt, but can be interrupted.
420  * The @ctor is run when new pages are allocated by the cache.
421  *
422  * The flags are
423  *
424  * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
425  * to catch references to uninitialised memory.
426  *
427  * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check
428  * for buffer overruns.
429  *
430  * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
431  * cacheline.  This can be beneficial if you're counting cycles as closely
432  * as davem.
433  */
434 struct kmem_cache *
435 kmem_cache_create(const char *name, size_t size, size_t align,
436 		  slab_flags_t flags, void (*ctor)(void *))
437 {
438 	struct kmem_cache *s = NULL;
439 	const char *cache_name;
440 	int err;
441 
442 	get_online_cpus();
443 	get_online_mems();
444 	memcg_get_cache_ids();
445 
446 	mutex_lock(&slab_mutex);
447 
448 	err = kmem_cache_sanity_check(name, size);
449 	if (err) {
450 		goto out_unlock;
451 	}
452 
453 	/* Refuse requests with allocator specific flags */
454 	if (flags & ~SLAB_FLAGS_PERMITTED) {
455 		err = -EINVAL;
456 		goto out_unlock;
457 	}
458 
459 	/*
460 	 * Some allocators will constraint the set of valid flags to a subset
461 	 * of all flags. We expect them to define CACHE_CREATE_MASK in this
462 	 * case, and we'll just provide them with a sanitized version of the
463 	 * passed flags.
464 	 */
465 	flags &= CACHE_CREATE_MASK;
466 
467 	s = __kmem_cache_alias(name, size, align, flags, ctor);
468 	if (s)
469 		goto out_unlock;
470 
471 	cache_name = kstrdup_const(name, GFP_KERNEL);
472 	if (!cache_name) {
473 		err = -ENOMEM;
474 		goto out_unlock;
475 	}
476 
477 	s = create_cache(cache_name, size, size,
478 			 calculate_alignment(flags, align, size),
479 			 flags, ctor, NULL, NULL);
480 	if (IS_ERR(s)) {
481 		err = PTR_ERR(s);
482 		kfree_const(cache_name);
483 	}
484 
485 out_unlock:
486 	mutex_unlock(&slab_mutex);
487 
488 	memcg_put_cache_ids();
489 	put_online_mems();
490 	put_online_cpus();
491 
492 	if (err) {
493 		if (flags & SLAB_PANIC)
494 			panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n",
495 				name, err);
496 		else {
497 			pr_warn("kmem_cache_create(%s) failed with error %d\n",
498 				name, err);
499 			dump_stack();
500 		}
501 		return NULL;
502 	}
503 	return s;
504 }
505 EXPORT_SYMBOL(kmem_cache_create);
506 
507 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work)
508 {
509 	LIST_HEAD(to_destroy);
510 	struct kmem_cache *s, *s2;
511 
512 	/*
513 	 * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the
514 	 * @slab_caches_to_rcu_destroy list.  The slab pages are freed
515 	 * through RCU and and the associated kmem_cache are dereferenced
516 	 * while freeing the pages, so the kmem_caches should be freed only
517 	 * after the pending RCU operations are finished.  As rcu_barrier()
518 	 * is a pretty slow operation, we batch all pending destructions
519 	 * asynchronously.
520 	 */
521 	mutex_lock(&slab_mutex);
522 	list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy);
523 	mutex_unlock(&slab_mutex);
524 
525 	if (list_empty(&to_destroy))
526 		return;
527 
528 	rcu_barrier();
529 
530 	list_for_each_entry_safe(s, s2, &to_destroy, list) {
531 #ifdef SLAB_SUPPORTS_SYSFS
532 		sysfs_slab_release(s);
533 #else
534 		slab_kmem_cache_release(s);
535 #endif
536 	}
537 }
538 
539 static int shutdown_cache(struct kmem_cache *s)
540 {
541 	/* free asan quarantined objects */
542 	kasan_cache_shutdown(s);
543 
544 	if (__kmem_cache_shutdown(s) != 0)
545 		return -EBUSY;
546 
547 	memcg_unlink_cache(s);
548 	list_del(&s->list);
549 
550 	if (s->flags & SLAB_TYPESAFE_BY_RCU) {
551 		list_add_tail(&s->list, &slab_caches_to_rcu_destroy);
552 		schedule_work(&slab_caches_to_rcu_destroy_work);
553 	} else {
554 #ifdef SLAB_SUPPORTS_SYSFS
555 		sysfs_slab_release(s);
556 #else
557 		slab_kmem_cache_release(s);
558 #endif
559 	}
560 
561 	return 0;
562 }
563 
564 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
565 /*
566  * memcg_create_kmem_cache - Create a cache for a memory cgroup.
567  * @memcg: The memory cgroup the new cache is for.
568  * @root_cache: The parent of the new cache.
569  *
570  * This function attempts to create a kmem cache that will serve allocation
571  * requests going from @memcg to @root_cache. The new cache inherits properties
572  * from its parent.
573  */
574 void memcg_create_kmem_cache(struct mem_cgroup *memcg,
575 			     struct kmem_cache *root_cache)
576 {
577 	static char memcg_name_buf[NAME_MAX + 1]; /* protected by slab_mutex */
578 	struct cgroup_subsys_state *css = &memcg->css;
579 	struct memcg_cache_array *arr;
580 	struct kmem_cache *s = NULL;
581 	char *cache_name;
582 	int idx;
583 
584 	get_online_cpus();
585 	get_online_mems();
586 
587 	mutex_lock(&slab_mutex);
588 
589 	/*
590 	 * The memory cgroup could have been offlined while the cache
591 	 * creation work was pending.
592 	 */
593 	if (memcg->kmem_state != KMEM_ONLINE)
594 		goto out_unlock;
595 
596 	idx = memcg_cache_id(memcg);
597 	arr = rcu_dereference_protected(root_cache->memcg_params.memcg_caches,
598 					lockdep_is_held(&slab_mutex));
599 
600 	/*
601 	 * Since per-memcg caches are created asynchronously on first
602 	 * allocation (see memcg_kmem_get_cache()), several threads can try to
603 	 * create the same cache, but only one of them may succeed.
604 	 */
605 	if (arr->entries[idx])
606 		goto out_unlock;
607 
608 	cgroup_name(css->cgroup, memcg_name_buf, sizeof(memcg_name_buf));
609 	cache_name = kasprintf(GFP_KERNEL, "%s(%llu:%s)", root_cache->name,
610 			       css->serial_nr, memcg_name_buf);
611 	if (!cache_name)
612 		goto out_unlock;
613 
614 	s = create_cache(cache_name, root_cache->object_size,
615 			 root_cache->size, root_cache->align,
616 			 root_cache->flags & CACHE_CREATE_MASK,
617 			 root_cache->ctor, memcg, root_cache);
618 	/*
619 	 * If we could not create a memcg cache, do not complain, because
620 	 * that's not critical at all as we can always proceed with the root
621 	 * cache.
622 	 */
623 	if (IS_ERR(s)) {
624 		kfree(cache_name);
625 		goto out_unlock;
626 	}
627 
628 	/*
629 	 * Since readers won't lock (see cache_from_memcg_idx()), we need a
630 	 * barrier here to ensure nobody will see the kmem_cache partially
631 	 * initialized.
632 	 */
633 	smp_wmb();
634 	arr->entries[idx] = s;
635 
636 out_unlock:
637 	mutex_unlock(&slab_mutex);
638 
639 	put_online_mems();
640 	put_online_cpus();
641 }
642 
643 static void kmemcg_deactivate_workfn(struct work_struct *work)
644 {
645 	struct kmem_cache *s = container_of(work, struct kmem_cache,
646 					    memcg_params.deact_work);
647 
648 	get_online_cpus();
649 	get_online_mems();
650 
651 	mutex_lock(&slab_mutex);
652 
653 	s->memcg_params.deact_fn(s);
654 
655 	mutex_unlock(&slab_mutex);
656 
657 	put_online_mems();
658 	put_online_cpus();
659 
660 	/* done, put the ref from slab_deactivate_memcg_cache_rcu_sched() */
661 	css_put(&s->memcg_params.memcg->css);
662 }
663 
664 static void kmemcg_deactivate_rcufn(struct rcu_head *head)
665 {
666 	struct kmem_cache *s = container_of(head, struct kmem_cache,
667 					    memcg_params.deact_rcu_head);
668 
669 	/*
670 	 * We need to grab blocking locks.  Bounce to ->deact_work.  The
671 	 * work item shares the space with the RCU head and can't be
672 	 * initialized eariler.
673 	 */
674 	INIT_WORK(&s->memcg_params.deact_work, kmemcg_deactivate_workfn);
675 	queue_work(memcg_kmem_cache_wq, &s->memcg_params.deact_work);
676 }
677 
678 /**
679  * slab_deactivate_memcg_cache_rcu_sched - schedule deactivation after a
680  *					   sched RCU grace period
681  * @s: target kmem_cache
682  * @deact_fn: deactivation function to call
683  *
684  * Schedule @deact_fn to be invoked with online cpus, mems and slab_mutex
685  * held after a sched RCU grace period.  The slab is guaranteed to stay
686  * alive until @deact_fn is finished.  This is to be used from
687  * __kmemcg_cache_deactivate().
688  */
689 void slab_deactivate_memcg_cache_rcu_sched(struct kmem_cache *s,
690 					   void (*deact_fn)(struct kmem_cache *))
691 {
692 	if (WARN_ON_ONCE(is_root_cache(s)) ||
693 	    WARN_ON_ONCE(s->memcg_params.deact_fn))
694 		return;
695 
696 	/* pin memcg so that @s doesn't get destroyed in the middle */
697 	css_get(&s->memcg_params.memcg->css);
698 
699 	s->memcg_params.deact_fn = deact_fn;
700 	call_rcu_sched(&s->memcg_params.deact_rcu_head, kmemcg_deactivate_rcufn);
701 }
702 
703 void memcg_deactivate_kmem_caches(struct mem_cgroup *memcg)
704 {
705 	int idx;
706 	struct memcg_cache_array *arr;
707 	struct kmem_cache *s, *c;
708 
709 	idx = memcg_cache_id(memcg);
710 
711 	get_online_cpus();
712 	get_online_mems();
713 
714 	mutex_lock(&slab_mutex);
715 	list_for_each_entry(s, &slab_root_caches, root_caches_node) {
716 		arr = rcu_dereference_protected(s->memcg_params.memcg_caches,
717 						lockdep_is_held(&slab_mutex));
718 		c = arr->entries[idx];
719 		if (!c)
720 			continue;
721 
722 		__kmemcg_cache_deactivate(c);
723 		arr->entries[idx] = NULL;
724 	}
725 	mutex_unlock(&slab_mutex);
726 
727 	put_online_mems();
728 	put_online_cpus();
729 }
730 
731 void memcg_destroy_kmem_caches(struct mem_cgroup *memcg)
732 {
733 	struct kmem_cache *s, *s2;
734 
735 	get_online_cpus();
736 	get_online_mems();
737 
738 	mutex_lock(&slab_mutex);
739 	list_for_each_entry_safe(s, s2, &memcg->kmem_caches,
740 				 memcg_params.kmem_caches_node) {
741 		/*
742 		 * The cgroup is about to be freed and therefore has no charges
743 		 * left. Hence, all its caches must be empty by now.
744 		 */
745 		BUG_ON(shutdown_cache(s));
746 	}
747 	mutex_unlock(&slab_mutex);
748 
749 	put_online_mems();
750 	put_online_cpus();
751 }
752 
753 static int shutdown_memcg_caches(struct kmem_cache *s)
754 {
755 	struct memcg_cache_array *arr;
756 	struct kmem_cache *c, *c2;
757 	LIST_HEAD(busy);
758 	int i;
759 
760 	BUG_ON(!is_root_cache(s));
761 
762 	/*
763 	 * First, shutdown active caches, i.e. caches that belong to online
764 	 * memory cgroups.
765 	 */
766 	arr = rcu_dereference_protected(s->memcg_params.memcg_caches,
767 					lockdep_is_held(&slab_mutex));
768 	for_each_memcg_cache_index(i) {
769 		c = arr->entries[i];
770 		if (!c)
771 			continue;
772 		if (shutdown_cache(c))
773 			/*
774 			 * The cache still has objects. Move it to a temporary
775 			 * list so as not to try to destroy it for a second
776 			 * time while iterating over inactive caches below.
777 			 */
778 			list_move(&c->memcg_params.children_node, &busy);
779 		else
780 			/*
781 			 * The cache is empty and will be destroyed soon. Clear
782 			 * the pointer to it in the memcg_caches array so that
783 			 * it will never be accessed even if the root cache
784 			 * stays alive.
785 			 */
786 			arr->entries[i] = NULL;
787 	}
788 
789 	/*
790 	 * Second, shutdown all caches left from memory cgroups that are now
791 	 * offline.
792 	 */
793 	list_for_each_entry_safe(c, c2, &s->memcg_params.children,
794 				 memcg_params.children_node)
795 		shutdown_cache(c);
796 
797 	list_splice(&busy, &s->memcg_params.children);
798 
799 	/*
800 	 * A cache being destroyed must be empty. In particular, this means
801 	 * that all per memcg caches attached to it must be empty too.
802 	 */
803 	if (!list_empty(&s->memcg_params.children))
804 		return -EBUSY;
805 	return 0;
806 }
807 #else
808 static inline int shutdown_memcg_caches(struct kmem_cache *s)
809 {
810 	return 0;
811 }
812 #endif /* CONFIG_MEMCG && !CONFIG_SLOB */
813 
814 void slab_kmem_cache_release(struct kmem_cache *s)
815 {
816 	__kmem_cache_release(s);
817 	destroy_memcg_params(s);
818 	kfree_const(s->name);
819 	kmem_cache_free(kmem_cache, s);
820 }
821 
822 void kmem_cache_destroy(struct kmem_cache *s)
823 {
824 	int err;
825 
826 	if (unlikely(!s))
827 		return;
828 
829 	get_online_cpus();
830 	get_online_mems();
831 
832 	mutex_lock(&slab_mutex);
833 
834 	s->refcount--;
835 	if (s->refcount)
836 		goto out_unlock;
837 
838 	err = shutdown_memcg_caches(s);
839 	if (!err)
840 		err = shutdown_cache(s);
841 
842 	if (err) {
843 		pr_err("kmem_cache_destroy %s: Slab cache still has objects\n",
844 		       s->name);
845 		dump_stack();
846 	}
847 out_unlock:
848 	mutex_unlock(&slab_mutex);
849 
850 	put_online_mems();
851 	put_online_cpus();
852 }
853 EXPORT_SYMBOL(kmem_cache_destroy);
854 
855 /**
856  * kmem_cache_shrink - Shrink a cache.
857  * @cachep: The cache to shrink.
858  *
859  * Releases as many slabs as possible for a cache.
860  * To help debugging, a zero exit status indicates all slabs were released.
861  */
862 int kmem_cache_shrink(struct kmem_cache *cachep)
863 {
864 	int ret;
865 
866 	get_online_cpus();
867 	get_online_mems();
868 	kasan_cache_shrink(cachep);
869 	ret = __kmem_cache_shrink(cachep);
870 	put_online_mems();
871 	put_online_cpus();
872 	return ret;
873 }
874 EXPORT_SYMBOL(kmem_cache_shrink);
875 
876 bool slab_is_available(void)
877 {
878 	return slab_state >= UP;
879 }
880 
881 #ifndef CONFIG_SLOB
882 /* Create a cache during boot when no slab services are available yet */
883 void __init create_boot_cache(struct kmem_cache *s, const char *name, size_t size,
884 		slab_flags_t flags)
885 {
886 	int err;
887 
888 	s->name = name;
889 	s->size = s->object_size = size;
890 	s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size);
891 
892 	slab_init_memcg_params(s);
893 
894 	err = __kmem_cache_create(s, flags);
895 
896 	if (err)
897 		panic("Creation of kmalloc slab %s size=%zu failed. Reason %d\n",
898 					name, size, err);
899 
900 	s->refcount = -1;	/* Exempt from merging for now */
901 }
902 
903 struct kmem_cache *__init create_kmalloc_cache(const char *name, size_t size,
904 				slab_flags_t flags)
905 {
906 	struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
907 
908 	if (!s)
909 		panic("Out of memory when creating slab %s\n", name);
910 
911 	create_boot_cache(s, name, size, flags);
912 	list_add(&s->list, &slab_caches);
913 	memcg_link_cache(s);
914 	s->refcount = 1;
915 	return s;
916 }
917 
918 struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1];
919 EXPORT_SYMBOL(kmalloc_caches);
920 
921 #ifdef CONFIG_ZONE_DMA
922 struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1];
923 EXPORT_SYMBOL(kmalloc_dma_caches);
924 #endif
925 
926 /*
927  * Conversion table for small slabs sizes / 8 to the index in the
928  * kmalloc array. This is necessary for slabs < 192 since we have non power
929  * of two cache sizes there. The size of larger slabs can be determined using
930  * fls.
931  */
932 static s8 size_index[24] = {
933 	3,	/* 8 */
934 	4,	/* 16 */
935 	5,	/* 24 */
936 	5,	/* 32 */
937 	6,	/* 40 */
938 	6,	/* 48 */
939 	6,	/* 56 */
940 	6,	/* 64 */
941 	1,	/* 72 */
942 	1,	/* 80 */
943 	1,	/* 88 */
944 	1,	/* 96 */
945 	7,	/* 104 */
946 	7,	/* 112 */
947 	7,	/* 120 */
948 	7,	/* 128 */
949 	2,	/* 136 */
950 	2,	/* 144 */
951 	2,	/* 152 */
952 	2,	/* 160 */
953 	2,	/* 168 */
954 	2,	/* 176 */
955 	2,	/* 184 */
956 	2	/* 192 */
957 };
958 
959 static inline int size_index_elem(size_t bytes)
960 {
961 	return (bytes - 1) / 8;
962 }
963 
964 /*
965  * Find the kmem_cache structure that serves a given size of
966  * allocation
967  */
968 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
969 {
970 	int index;
971 
972 	if (unlikely(size > KMALLOC_MAX_SIZE)) {
973 		WARN_ON_ONCE(!(flags & __GFP_NOWARN));
974 		return NULL;
975 	}
976 
977 	if (size <= 192) {
978 		if (!size)
979 			return ZERO_SIZE_PTR;
980 
981 		index = size_index[size_index_elem(size)];
982 	} else
983 		index = fls(size - 1);
984 
985 #ifdef CONFIG_ZONE_DMA
986 	if (unlikely((flags & GFP_DMA)))
987 		return kmalloc_dma_caches[index];
988 
989 #endif
990 	return kmalloc_caches[index];
991 }
992 
993 /*
994  * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
995  * kmalloc_index() supports up to 2^26=64MB, so the final entry of the table is
996  * kmalloc-67108864.
997  */
998 const struct kmalloc_info_struct kmalloc_info[] __initconst = {
999 	{NULL,                      0},		{"kmalloc-96",             96},
1000 	{"kmalloc-192",           192},		{"kmalloc-8",               8},
1001 	{"kmalloc-16",             16},		{"kmalloc-32",             32},
1002 	{"kmalloc-64",             64},		{"kmalloc-128",           128},
1003 	{"kmalloc-256",           256},		{"kmalloc-512",           512},
1004 	{"kmalloc-1024",         1024},		{"kmalloc-2048",         2048},
1005 	{"kmalloc-4096",         4096},		{"kmalloc-8192",         8192},
1006 	{"kmalloc-16384",       16384},		{"kmalloc-32768",       32768},
1007 	{"kmalloc-65536",       65536},		{"kmalloc-131072",     131072},
1008 	{"kmalloc-262144",     262144},		{"kmalloc-524288",     524288},
1009 	{"kmalloc-1048576",   1048576},		{"kmalloc-2097152",   2097152},
1010 	{"kmalloc-4194304",   4194304},		{"kmalloc-8388608",   8388608},
1011 	{"kmalloc-16777216", 16777216},		{"kmalloc-33554432", 33554432},
1012 	{"kmalloc-67108864", 67108864}
1013 };
1014 
1015 /*
1016  * Patch up the size_index table if we have strange large alignment
1017  * requirements for the kmalloc array. This is only the case for
1018  * MIPS it seems. The standard arches will not generate any code here.
1019  *
1020  * Largest permitted alignment is 256 bytes due to the way we
1021  * handle the index determination for the smaller caches.
1022  *
1023  * Make sure that nothing crazy happens if someone starts tinkering
1024  * around with ARCH_KMALLOC_MINALIGN
1025  */
1026 void __init setup_kmalloc_cache_index_table(void)
1027 {
1028 	int i;
1029 
1030 	BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
1031 		(KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
1032 
1033 	for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
1034 		int elem = size_index_elem(i);
1035 
1036 		if (elem >= ARRAY_SIZE(size_index))
1037 			break;
1038 		size_index[elem] = KMALLOC_SHIFT_LOW;
1039 	}
1040 
1041 	if (KMALLOC_MIN_SIZE >= 64) {
1042 		/*
1043 		 * The 96 byte size cache is not used if the alignment
1044 		 * is 64 byte.
1045 		 */
1046 		for (i = 64 + 8; i <= 96; i += 8)
1047 			size_index[size_index_elem(i)] = 7;
1048 
1049 	}
1050 
1051 	if (KMALLOC_MIN_SIZE >= 128) {
1052 		/*
1053 		 * The 192 byte sized cache is not used if the alignment
1054 		 * is 128 byte. Redirect kmalloc to use the 256 byte cache
1055 		 * instead.
1056 		 */
1057 		for (i = 128 + 8; i <= 192; i += 8)
1058 			size_index[size_index_elem(i)] = 8;
1059 	}
1060 }
1061 
1062 static void __init new_kmalloc_cache(int idx, slab_flags_t flags)
1063 {
1064 	kmalloc_caches[idx] = create_kmalloc_cache(kmalloc_info[idx].name,
1065 					kmalloc_info[idx].size, flags);
1066 }
1067 
1068 /*
1069  * Create the kmalloc array. Some of the regular kmalloc arrays
1070  * may already have been created because they were needed to
1071  * enable allocations for slab creation.
1072  */
1073 void __init create_kmalloc_caches(slab_flags_t flags)
1074 {
1075 	int i;
1076 
1077 	for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
1078 		if (!kmalloc_caches[i])
1079 			new_kmalloc_cache(i, flags);
1080 
1081 		/*
1082 		 * Caches that are not of the two-to-the-power-of size.
1083 		 * These have to be created immediately after the
1084 		 * earlier power of two caches
1085 		 */
1086 		if (KMALLOC_MIN_SIZE <= 32 && !kmalloc_caches[1] && i == 6)
1087 			new_kmalloc_cache(1, flags);
1088 		if (KMALLOC_MIN_SIZE <= 64 && !kmalloc_caches[2] && i == 7)
1089 			new_kmalloc_cache(2, flags);
1090 	}
1091 
1092 	/* Kmalloc array is now usable */
1093 	slab_state = UP;
1094 
1095 #ifdef CONFIG_ZONE_DMA
1096 	for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
1097 		struct kmem_cache *s = kmalloc_caches[i];
1098 
1099 		if (s) {
1100 			int size = kmalloc_size(i);
1101 			char *n = kasprintf(GFP_NOWAIT,
1102 				 "dma-kmalloc-%d", size);
1103 
1104 			BUG_ON(!n);
1105 			kmalloc_dma_caches[i] = create_kmalloc_cache(n,
1106 				size, SLAB_CACHE_DMA | flags);
1107 		}
1108 	}
1109 #endif
1110 }
1111 #endif /* !CONFIG_SLOB */
1112 
1113 /*
1114  * To avoid unnecessary overhead, we pass through large allocation requests
1115  * directly to the page allocator. We use __GFP_COMP, because we will need to
1116  * know the allocation order to free the pages properly in kfree.
1117  */
1118 void *kmalloc_order(size_t size, gfp_t flags, unsigned int order)
1119 {
1120 	void *ret;
1121 	struct page *page;
1122 
1123 	flags |= __GFP_COMP;
1124 	page = alloc_pages(flags, order);
1125 	ret = page ? page_address(page) : NULL;
1126 	kmemleak_alloc(ret, size, 1, flags);
1127 	kasan_kmalloc_large(ret, size, flags);
1128 	return ret;
1129 }
1130 EXPORT_SYMBOL(kmalloc_order);
1131 
1132 #ifdef CONFIG_TRACING
1133 void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
1134 {
1135 	void *ret = kmalloc_order(size, flags, order);
1136 	trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
1137 	return ret;
1138 }
1139 EXPORT_SYMBOL(kmalloc_order_trace);
1140 #endif
1141 
1142 #ifdef CONFIG_SLAB_FREELIST_RANDOM
1143 /* Randomize a generic freelist */
1144 static void freelist_randomize(struct rnd_state *state, unsigned int *list,
1145 			size_t count)
1146 {
1147 	size_t i;
1148 	unsigned int rand;
1149 
1150 	for (i = 0; i < count; i++)
1151 		list[i] = i;
1152 
1153 	/* Fisher-Yates shuffle */
1154 	for (i = count - 1; i > 0; i--) {
1155 		rand = prandom_u32_state(state);
1156 		rand %= (i + 1);
1157 		swap(list[i], list[rand]);
1158 	}
1159 }
1160 
1161 /* Create a random sequence per cache */
1162 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
1163 				    gfp_t gfp)
1164 {
1165 	struct rnd_state state;
1166 
1167 	if (count < 2 || cachep->random_seq)
1168 		return 0;
1169 
1170 	cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp);
1171 	if (!cachep->random_seq)
1172 		return -ENOMEM;
1173 
1174 	/* Get best entropy at this stage of boot */
1175 	prandom_seed_state(&state, get_random_long());
1176 
1177 	freelist_randomize(&state, cachep->random_seq, count);
1178 	return 0;
1179 }
1180 
1181 /* Destroy the per-cache random freelist sequence */
1182 void cache_random_seq_destroy(struct kmem_cache *cachep)
1183 {
1184 	kfree(cachep->random_seq);
1185 	cachep->random_seq = NULL;
1186 }
1187 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
1188 
1189 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
1190 #ifdef CONFIG_SLAB
1191 #define SLABINFO_RIGHTS (S_IWUSR | S_IRUSR)
1192 #else
1193 #define SLABINFO_RIGHTS S_IRUSR
1194 #endif
1195 
1196 static void print_slabinfo_header(struct seq_file *m)
1197 {
1198 	/*
1199 	 * Output format version, so at least we can change it
1200 	 * without _too_ many complaints.
1201 	 */
1202 #ifdef CONFIG_DEBUG_SLAB
1203 	seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
1204 #else
1205 	seq_puts(m, "slabinfo - version: 2.1\n");
1206 #endif
1207 	seq_puts(m, "# name            <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
1208 	seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
1209 	seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
1210 #ifdef CONFIG_DEBUG_SLAB
1211 	seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
1212 	seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
1213 #endif
1214 	seq_putc(m, '\n');
1215 }
1216 
1217 void *slab_start(struct seq_file *m, loff_t *pos)
1218 {
1219 	mutex_lock(&slab_mutex);
1220 	return seq_list_start(&slab_root_caches, *pos);
1221 }
1222 
1223 void *slab_next(struct seq_file *m, void *p, loff_t *pos)
1224 {
1225 	return seq_list_next(p, &slab_root_caches, pos);
1226 }
1227 
1228 void slab_stop(struct seq_file *m, void *p)
1229 {
1230 	mutex_unlock(&slab_mutex);
1231 }
1232 
1233 static void
1234 memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info)
1235 {
1236 	struct kmem_cache *c;
1237 	struct slabinfo sinfo;
1238 
1239 	if (!is_root_cache(s))
1240 		return;
1241 
1242 	for_each_memcg_cache(c, s) {
1243 		memset(&sinfo, 0, sizeof(sinfo));
1244 		get_slabinfo(c, &sinfo);
1245 
1246 		info->active_slabs += sinfo.active_slabs;
1247 		info->num_slabs += sinfo.num_slabs;
1248 		info->shared_avail += sinfo.shared_avail;
1249 		info->active_objs += sinfo.active_objs;
1250 		info->num_objs += sinfo.num_objs;
1251 	}
1252 }
1253 
1254 static void cache_show(struct kmem_cache *s, struct seq_file *m)
1255 {
1256 	struct slabinfo sinfo;
1257 
1258 	memset(&sinfo, 0, sizeof(sinfo));
1259 	get_slabinfo(s, &sinfo);
1260 
1261 	memcg_accumulate_slabinfo(s, &sinfo);
1262 
1263 	seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
1264 		   cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size,
1265 		   sinfo.objects_per_slab, (1 << sinfo.cache_order));
1266 
1267 	seq_printf(m, " : tunables %4u %4u %4u",
1268 		   sinfo.limit, sinfo.batchcount, sinfo.shared);
1269 	seq_printf(m, " : slabdata %6lu %6lu %6lu",
1270 		   sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
1271 	slabinfo_show_stats(m, s);
1272 	seq_putc(m, '\n');
1273 }
1274 
1275 static int slab_show(struct seq_file *m, void *p)
1276 {
1277 	struct kmem_cache *s = list_entry(p, struct kmem_cache, root_caches_node);
1278 
1279 	if (p == slab_root_caches.next)
1280 		print_slabinfo_header(m);
1281 	cache_show(s, m);
1282 	return 0;
1283 }
1284 
1285 void dump_unreclaimable_slab(void)
1286 {
1287 	struct kmem_cache *s, *s2;
1288 	struct slabinfo sinfo;
1289 
1290 	/*
1291 	 * Here acquiring slab_mutex is risky since we don't prefer to get
1292 	 * sleep in oom path. But, without mutex hold, it may introduce a
1293 	 * risk of crash.
1294 	 * Use mutex_trylock to protect the list traverse, dump nothing
1295 	 * without acquiring the mutex.
1296 	 */
1297 	if (!mutex_trylock(&slab_mutex)) {
1298 		pr_warn("excessive unreclaimable slab but cannot dump stats\n");
1299 		return;
1300 	}
1301 
1302 	pr_info("Unreclaimable slab info:\n");
1303 	pr_info("Name                      Used          Total\n");
1304 
1305 	list_for_each_entry_safe(s, s2, &slab_caches, list) {
1306 		if (!is_root_cache(s) || (s->flags & SLAB_RECLAIM_ACCOUNT))
1307 			continue;
1308 
1309 		get_slabinfo(s, &sinfo);
1310 
1311 		if (sinfo.num_objs > 0)
1312 			pr_info("%-17s %10luKB %10luKB\n", cache_name(s),
1313 				(sinfo.active_objs * s->size) / 1024,
1314 				(sinfo.num_objs * s->size) / 1024);
1315 	}
1316 	mutex_unlock(&slab_mutex);
1317 }
1318 
1319 #if defined(CONFIG_MEMCG)
1320 void *memcg_slab_start(struct seq_file *m, loff_t *pos)
1321 {
1322 	struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
1323 
1324 	mutex_lock(&slab_mutex);
1325 	return seq_list_start(&memcg->kmem_caches, *pos);
1326 }
1327 
1328 void *memcg_slab_next(struct seq_file *m, void *p, loff_t *pos)
1329 {
1330 	struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
1331 
1332 	return seq_list_next(p, &memcg->kmem_caches, pos);
1333 }
1334 
1335 void memcg_slab_stop(struct seq_file *m, void *p)
1336 {
1337 	mutex_unlock(&slab_mutex);
1338 }
1339 
1340 int memcg_slab_show(struct seq_file *m, void *p)
1341 {
1342 	struct kmem_cache *s = list_entry(p, struct kmem_cache,
1343 					  memcg_params.kmem_caches_node);
1344 	struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
1345 
1346 	if (p == memcg->kmem_caches.next)
1347 		print_slabinfo_header(m);
1348 	cache_show(s, m);
1349 	return 0;
1350 }
1351 #endif
1352 
1353 /*
1354  * slabinfo_op - iterator that generates /proc/slabinfo
1355  *
1356  * Output layout:
1357  * cache-name
1358  * num-active-objs
1359  * total-objs
1360  * object size
1361  * num-active-slabs
1362  * total-slabs
1363  * num-pages-per-slab
1364  * + further values on SMP and with statistics enabled
1365  */
1366 static const struct seq_operations slabinfo_op = {
1367 	.start = slab_start,
1368 	.next = slab_next,
1369 	.stop = slab_stop,
1370 	.show = slab_show,
1371 };
1372 
1373 static int slabinfo_open(struct inode *inode, struct file *file)
1374 {
1375 	return seq_open(file, &slabinfo_op);
1376 }
1377 
1378 static const struct file_operations proc_slabinfo_operations = {
1379 	.open		= slabinfo_open,
1380 	.read		= seq_read,
1381 	.write          = slabinfo_write,
1382 	.llseek		= seq_lseek,
1383 	.release	= seq_release,
1384 };
1385 
1386 static int __init slab_proc_init(void)
1387 {
1388 	proc_create("slabinfo", SLABINFO_RIGHTS, NULL,
1389 						&proc_slabinfo_operations);
1390 	return 0;
1391 }
1392 module_init(slab_proc_init);
1393 #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */
1394 
1395 static __always_inline void *__do_krealloc(const void *p, size_t new_size,
1396 					   gfp_t flags)
1397 {
1398 	void *ret;
1399 	size_t ks = 0;
1400 
1401 	if (p)
1402 		ks = ksize(p);
1403 
1404 	if (ks >= new_size) {
1405 		kasan_krealloc((void *)p, new_size, flags);
1406 		return (void *)p;
1407 	}
1408 
1409 	ret = kmalloc_track_caller(new_size, flags);
1410 	if (ret && p)
1411 		memcpy(ret, p, ks);
1412 
1413 	return ret;
1414 }
1415 
1416 /**
1417  * __krealloc - like krealloc() but don't free @p.
1418  * @p: object to reallocate memory for.
1419  * @new_size: how many bytes of memory are required.
1420  * @flags: the type of memory to allocate.
1421  *
1422  * This function is like krealloc() except it never frees the originally
1423  * allocated buffer. Use this if you don't want to free the buffer immediately
1424  * like, for example, with RCU.
1425  */
1426 void *__krealloc(const void *p, size_t new_size, gfp_t flags)
1427 {
1428 	if (unlikely(!new_size))
1429 		return ZERO_SIZE_PTR;
1430 
1431 	return __do_krealloc(p, new_size, flags);
1432 
1433 }
1434 EXPORT_SYMBOL(__krealloc);
1435 
1436 /**
1437  * krealloc - reallocate memory. The contents will remain unchanged.
1438  * @p: object to reallocate memory for.
1439  * @new_size: how many bytes of memory are required.
1440  * @flags: the type of memory to allocate.
1441  *
1442  * The contents of the object pointed to are preserved up to the
1443  * lesser of the new and old sizes.  If @p is %NULL, krealloc()
1444  * behaves exactly like kmalloc().  If @new_size is 0 and @p is not a
1445  * %NULL pointer, the object pointed to is freed.
1446  */
1447 void *krealloc(const void *p, size_t new_size, gfp_t flags)
1448 {
1449 	void *ret;
1450 
1451 	if (unlikely(!new_size)) {
1452 		kfree(p);
1453 		return ZERO_SIZE_PTR;
1454 	}
1455 
1456 	ret = __do_krealloc(p, new_size, flags);
1457 	if (ret && p != ret)
1458 		kfree(p);
1459 
1460 	return ret;
1461 }
1462 EXPORT_SYMBOL(krealloc);
1463 
1464 /**
1465  * kzfree - like kfree but zero memory
1466  * @p: object to free memory of
1467  *
1468  * The memory of the object @p points to is zeroed before freed.
1469  * If @p is %NULL, kzfree() does nothing.
1470  *
1471  * Note: this function zeroes the whole allocated buffer which can be a good
1472  * deal bigger than the requested buffer size passed to kmalloc(). So be
1473  * careful when using this function in performance sensitive code.
1474  */
1475 void kzfree(const void *p)
1476 {
1477 	size_t ks;
1478 	void *mem = (void *)p;
1479 
1480 	if (unlikely(ZERO_OR_NULL_PTR(mem)))
1481 		return;
1482 	ks = ksize(mem);
1483 	memset(mem, 0, ks);
1484 	kfree(mem);
1485 }
1486 EXPORT_SYMBOL(kzfree);
1487 
1488 /* Tracepoints definitions. */
1489 EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1490 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1491 EXPORT_TRACEPOINT_SYMBOL(kmalloc_node);
1492 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node);
1493 EXPORT_TRACEPOINT_SYMBOL(kfree);
1494 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);
1495