xref: /openbmc/linux/mm/slab_common.c (revision 5d331b7f)
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/cache.h>
14 #include <linux/compiler.h>
15 #include <linux/module.h>
16 #include <linux/cpu.h>
17 #include <linux/uaccess.h>
18 #include <linux/seq_file.h>
19 #include <linux/proc_fs.h>
20 #include <asm/cacheflush.h>
21 #include <asm/tlbflush.h>
22 #include <asm/page.h>
23 #include <linux/memcontrol.h>
24 
25 #define CREATE_TRACE_POINTS
26 #include <trace/events/kmem.h>
27 
28 #include "slab.h"
29 
30 enum slab_state slab_state;
31 LIST_HEAD(slab_caches);
32 DEFINE_MUTEX(slab_mutex);
33 struct kmem_cache *kmem_cache;
34 
35 #ifdef CONFIG_HARDENED_USERCOPY
36 bool usercopy_fallback __ro_after_init =
37 		IS_ENABLED(CONFIG_HARDENED_USERCOPY_FALLBACK);
38 module_param(usercopy_fallback, bool, 0400);
39 MODULE_PARM_DESC(usercopy_fallback,
40 		"WARN instead of reject usercopy whitelist violations");
41 #endif
42 
43 static LIST_HEAD(slab_caches_to_rcu_destroy);
44 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work);
45 static DECLARE_WORK(slab_caches_to_rcu_destroy_work,
46 		    slab_caches_to_rcu_destroy_workfn);
47 
48 /*
49  * Set of flags that will prevent slab merging
50  */
51 #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
52 		SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \
53 		SLAB_FAILSLAB | SLAB_KASAN)
54 
55 #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \
56 			 SLAB_ACCOUNT)
57 
58 /*
59  * Merge control. If this is set then no merging of slab caches will occur.
60  */
61 static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT);
62 
63 static int __init setup_slab_nomerge(char *str)
64 {
65 	slab_nomerge = true;
66 	return 1;
67 }
68 
69 #ifdef CONFIG_SLUB
70 __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0);
71 #endif
72 
73 __setup("slab_nomerge", setup_slab_nomerge);
74 
75 /*
76  * Determine the size of a slab object
77  */
78 unsigned int kmem_cache_size(struct kmem_cache *s)
79 {
80 	return s->object_size;
81 }
82 EXPORT_SYMBOL(kmem_cache_size);
83 
84 #ifdef CONFIG_DEBUG_VM
85 static int kmem_cache_sanity_check(const char *name, unsigned int size)
86 {
87 	if (!name || in_interrupt() || size < sizeof(void *) ||
88 		size > KMALLOC_MAX_SIZE) {
89 		pr_err("kmem_cache_create(%s) integrity check failed\n", name);
90 		return -EINVAL;
91 	}
92 
93 	WARN_ON(strchr(name, ' '));	/* It confuses parsers */
94 	return 0;
95 }
96 #else
97 static inline int kmem_cache_sanity_check(const char *name, unsigned int size)
98 {
99 	return 0;
100 }
101 #endif
102 
103 void __kmem_cache_free_bulk(struct kmem_cache *s, size_t nr, void **p)
104 {
105 	size_t i;
106 
107 	for (i = 0; i < nr; i++) {
108 		if (s)
109 			kmem_cache_free(s, p[i]);
110 		else
111 			kfree(p[i]);
112 	}
113 }
114 
115 int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t nr,
116 								void **p)
117 {
118 	size_t i;
119 
120 	for (i = 0; i < nr; i++) {
121 		void *x = p[i] = kmem_cache_alloc(s, flags);
122 		if (!x) {
123 			__kmem_cache_free_bulk(s, i, p);
124 			return 0;
125 		}
126 	}
127 	return i;
128 }
129 
130 #ifdef CONFIG_MEMCG_KMEM
131 
132 LIST_HEAD(slab_root_caches);
133 
134 void slab_init_memcg_params(struct kmem_cache *s)
135 {
136 	s->memcg_params.root_cache = NULL;
137 	RCU_INIT_POINTER(s->memcg_params.memcg_caches, NULL);
138 	INIT_LIST_HEAD(&s->memcg_params.children);
139 	s->memcg_params.dying = false;
140 }
141 
142 static int init_memcg_params(struct kmem_cache *s,
143 		struct mem_cgroup *memcg, struct kmem_cache *root_cache)
144 {
145 	struct memcg_cache_array *arr;
146 
147 	if (root_cache) {
148 		s->memcg_params.root_cache = root_cache;
149 		s->memcg_params.memcg = memcg;
150 		INIT_LIST_HEAD(&s->memcg_params.children_node);
151 		INIT_LIST_HEAD(&s->memcg_params.kmem_caches_node);
152 		return 0;
153 	}
154 
155 	slab_init_memcg_params(s);
156 
157 	if (!memcg_nr_cache_ids)
158 		return 0;
159 
160 	arr = kvzalloc(sizeof(struct memcg_cache_array) +
161 		       memcg_nr_cache_ids * sizeof(void *),
162 		       GFP_KERNEL);
163 	if (!arr)
164 		return -ENOMEM;
165 
166 	RCU_INIT_POINTER(s->memcg_params.memcg_caches, arr);
167 	return 0;
168 }
169 
170 static void destroy_memcg_params(struct kmem_cache *s)
171 {
172 	if (is_root_cache(s))
173 		kvfree(rcu_access_pointer(s->memcg_params.memcg_caches));
174 }
175 
176 static void free_memcg_params(struct rcu_head *rcu)
177 {
178 	struct memcg_cache_array *old;
179 
180 	old = container_of(rcu, struct memcg_cache_array, rcu);
181 	kvfree(old);
182 }
183 
184 static int update_memcg_params(struct kmem_cache *s, int new_array_size)
185 {
186 	struct memcg_cache_array *old, *new;
187 
188 	new = kvzalloc(sizeof(struct memcg_cache_array) +
189 		       new_array_size * sizeof(void *), GFP_KERNEL);
190 	if (!new)
191 		return -ENOMEM;
192 
193 	old = rcu_dereference_protected(s->memcg_params.memcg_caches,
194 					lockdep_is_held(&slab_mutex));
195 	if (old)
196 		memcpy(new->entries, old->entries,
197 		       memcg_nr_cache_ids * sizeof(void *));
198 
199 	rcu_assign_pointer(s->memcg_params.memcg_caches, new);
200 	if (old)
201 		call_rcu(&old->rcu, free_memcg_params);
202 	return 0;
203 }
204 
205 int memcg_update_all_caches(int num_memcgs)
206 {
207 	struct kmem_cache *s;
208 	int ret = 0;
209 
210 	mutex_lock(&slab_mutex);
211 	list_for_each_entry(s, &slab_root_caches, root_caches_node) {
212 		ret = update_memcg_params(s, num_memcgs);
213 		/*
214 		 * Instead of freeing the memory, we'll just leave the caches
215 		 * up to this point in an updated state.
216 		 */
217 		if (ret)
218 			break;
219 	}
220 	mutex_unlock(&slab_mutex);
221 	return ret;
222 }
223 
224 void memcg_link_cache(struct kmem_cache *s)
225 {
226 	if (is_root_cache(s)) {
227 		list_add(&s->root_caches_node, &slab_root_caches);
228 	} else {
229 		list_add(&s->memcg_params.children_node,
230 			 &s->memcg_params.root_cache->memcg_params.children);
231 		list_add(&s->memcg_params.kmem_caches_node,
232 			 &s->memcg_params.memcg->kmem_caches);
233 	}
234 }
235 
236 static void memcg_unlink_cache(struct kmem_cache *s)
237 {
238 	if (is_root_cache(s)) {
239 		list_del(&s->root_caches_node);
240 	} else {
241 		list_del(&s->memcg_params.children_node);
242 		list_del(&s->memcg_params.kmem_caches_node);
243 	}
244 }
245 #else
246 static inline int init_memcg_params(struct kmem_cache *s,
247 		struct mem_cgroup *memcg, struct kmem_cache *root_cache)
248 {
249 	return 0;
250 }
251 
252 static inline void destroy_memcg_params(struct kmem_cache *s)
253 {
254 }
255 
256 static inline void memcg_unlink_cache(struct kmem_cache *s)
257 {
258 }
259 #endif /* CONFIG_MEMCG_KMEM */
260 
261 /*
262  * Figure out what the alignment of the objects will be given a set of
263  * flags, a user specified alignment and the size of the objects.
264  */
265 static unsigned int calculate_alignment(slab_flags_t flags,
266 		unsigned int align, unsigned int size)
267 {
268 	/*
269 	 * If the user wants hardware cache aligned objects then follow that
270 	 * suggestion if the object is sufficiently large.
271 	 *
272 	 * The hardware cache alignment cannot override the specified
273 	 * alignment though. If that is greater then use it.
274 	 */
275 	if (flags & SLAB_HWCACHE_ALIGN) {
276 		unsigned int ralign;
277 
278 		ralign = cache_line_size();
279 		while (size <= ralign / 2)
280 			ralign /= 2;
281 		align = max(align, ralign);
282 	}
283 
284 	if (align < ARCH_SLAB_MINALIGN)
285 		align = ARCH_SLAB_MINALIGN;
286 
287 	return ALIGN(align, sizeof(void *));
288 }
289 
290 /*
291  * Find a mergeable slab cache
292  */
293 int slab_unmergeable(struct kmem_cache *s)
294 {
295 	if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
296 		return 1;
297 
298 	if (!is_root_cache(s))
299 		return 1;
300 
301 	if (s->ctor)
302 		return 1;
303 
304 	if (s->usersize)
305 		return 1;
306 
307 	/*
308 	 * We may have set a slab to be unmergeable during bootstrap.
309 	 */
310 	if (s->refcount < 0)
311 		return 1;
312 
313 	return 0;
314 }
315 
316 struct kmem_cache *find_mergeable(unsigned int size, unsigned int align,
317 		slab_flags_t flags, const char *name, void (*ctor)(void *))
318 {
319 	struct kmem_cache *s;
320 
321 	if (slab_nomerge)
322 		return NULL;
323 
324 	if (ctor)
325 		return NULL;
326 
327 	size = ALIGN(size, sizeof(void *));
328 	align = calculate_alignment(flags, align, size);
329 	size = ALIGN(size, align);
330 	flags = kmem_cache_flags(size, flags, name, NULL);
331 
332 	if (flags & SLAB_NEVER_MERGE)
333 		return NULL;
334 
335 	list_for_each_entry_reverse(s, &slab_root_caches, root_caches_node) {
336 		if (slab_unmergeable(s))
337 			continue;
338 
339 		if (size > s->size)
340 			continue;
341 
342 		if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
343 			continue;
344 		/*
345 		 * Check if alignment is compatible.
346 		 * Courtesy of Adrian Drzewiecki
347 		 */
348 		if ((s->size & ~(align - 1)) != s->size)
349 			continue;
350 
351 		if (s->size - size >= sizeof(void *))
352 			continue;
353 
354 		if (IS_ENABLED(CONFIG_SLAB) && align &&
355 			(align > s->align || s->align % align))
356 			continue;
357 
358 		return s;
359 	}
360 	return NULL;
361 }
362 
363 static struct kmem_cache *create_cache(const char *name,
364 		unsigned int object_size, unsigned int align,
365 		slab_flags_t flags, unsigned int useroffset,
366 		unsigned int usersize, void (*ctor)(void *),
367 		struct mem_cgroup *memcg, struct kmem_cache *root_cache)
368 {
369 	struct kmem_cache *s;
370 	int err;
371 
372 	if (WARN_ON(useroffset + usersize > object_size))
373 		useroffset = usersize = 0;
374 
375 	err = -ENOMEM;
376 	s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
377 	if (!s)
378 		goto out;
379 
380 	s->name = name;
381 	s->size = s->object_size = object_size;
382 	s->align = align;
383 	s->ctor = ctor;
384 	s->useroffset = useroffset;
385 	s->usersize = usersize;
386 
387 	err = init_memcg_params(s, memcg, root_cache);
388 	if (err)
389 		goto out_free_cache;
390 
391 	err = __kmem_cache_create(s, flags);
392 	if (err)
393 		goto out_free_cache;
394 
395 	s->refcount = 1;
396 	list_add(&s->list, &slab_caches);
397 	memcg_link_cache(s);
398 out:
399 	if (err)
400 		return ERR_PTR(err);
401 	return s;
402 
403 out_free_cache:
404 	destroy_memcg_params(s);
405 	kmem_cache_free(kmem_cache, s);
406 	goto out;
407 }
408 
409 /*
410  * kmem_cache_create_usercopy - Create a cache.
411  * @name: A string which is used in /proc/slabinfo to identify this cache.
412  * @size: The size of objects to be created in this cache.
413  * @align: The required alignment for the objects.
414  * @flags: SLAB flags
415  * @useroffset: Usercopy region offset
416  * @usersize: Usercopy region size
417  * @ctor: A constructor for the objects.
418  *
419  * Returns a ptr to the cache on success, NULL on failure.
420  * Cannot be called within a interrupt, but can be interrupted.
421  * The @ctor is run when new pages are allocated by the cache.
422  *
423  * The flags are
424  *
425  * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
426  * to catch references to uninitialised memory.
427  *
428  * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check
429  * for buffer overruns.
430  *
431  * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
432  * cacheline.  This can be beneficial if you're counting cycles as closely
433  * as davem.
434  */
435 struct kmem_cache *
436 kmem_cache_create_usercopy(const char *name,
437 		  unsigned int size, unsigned int align,
438 		  slab_flags_t flags,
439 		  unsigned int useroffset, unsigned int usersize,
440 		  void (*ctor)(void *))
441 {
442 	struct kmem_cache *s = NULL;
443 	const char *cache_name;
444 	int err;
445 
446 	get_online_cpus();
447 	get_online_mems();
448 	memcg_get_cache_ids();
449 
450 	mutex_lock(&slab_mutex);
451 
452 	err = kmem_cache_sanity_check(name, size);
453 	if (err) {
454 		goto out_unlock;
455 	}
456 
457 	/* Refuse requests with allocator specific flags */
458 	if (flags & ~SLAB_FLAGS_PERMITTED) {
459 		err = -EINVAL;
460 		goto out_unlock;
461 	}
462 
463 	/*
464 	 * Some allocators will constraint the set of valid flags to a subset
465 	 * of all flags. We expect them to define CACHE_CREATE_MASK in this
466 	 * case, and we'll just provide them with a sanitized version of the
467 	 * passed flags.
468 	 */
469 	flags &= CACHE_CREATE_MASK;
470 
471 	/* Fail closed on bad usersize of useroffset values. */
472 	if (WARN_ON(!usersize && useroffset) ||
473 	    WARN_ON(size < usersize || size - usersize < useroffset))
474 		usersize = useroffset = 0;
475 
476 	if (!usersize)
477 		s = __kmem_cache_alias(name, size, align, flags, ctor);
478 	if (s)
479 		goto out_unlock;
480 
481 	cache_name = kstrdup_const(name, GFP_KERNEL);
482 	if (!cache_name) {
483 		err = -ENOMEM;
484 		goto out_unlock;
485 	}
486 
487 	s = create_cache(cache_name, size,
488 			 calculate_alignment(flags, align, size),
489 			 flags, useroffset, usersize, ctor, NULL, NULL);
490 	if (IS_ERR(s)) {
491 		err = PTR_ERR(s);
492 		kfree_const(cache_name);
493 	}
494 
495 out_unlock:
496 	mutex_unlock(&slab_mutex);
497 
498 	memcg_put_cache_ids();
499 	put_online_mems();
500 	put_online_cpus();
501 
502 	if (err) {
503 		if (flags & SLAB_PANIC)
504 			panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n",
505 				name, err);
506 		else {
507 			pr_warn("kmem_cache_create(%s) failed with error %d\n",
508 				name, err);
509 			dump_stack();
510 		}
511 		return NULL;
512 	}
513 	return s;
514 }
515 EXPORT_SYMBOL(kmem_cache_create_usercopy);
516 
517 struct kmem_cache *
518 kmem_cache_create(const char *name, unsigned int size, unsigned int align,
519 		slab_flags_t flags, void (*ctor)(void *))
520 {
521 	return kmem_cache_create_usercopy(name, size, align, flags, 0, 0,
522 					  ctor);
523 }
524 EXPORT_SYMBOL(kmem_cache_create);
525 
526 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work)
527 {
528 	LIST_HEAD(to_destroy);
529 	struct kmem_cache *s, *s2;
530 
531 	/*
532 	 * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the
533 	 * @slab_caches_to_rcu_destroy list.  The slab pages are freed
534 	 * through RCU and and the associated kmem_cache are dereferenced
535 	 * while freeing the pages, so the kmem_caches should be freed only
536 	 * after the pending RCU operations are finished.  As rcu_barrier()
537 	 * is a pretty slow operation, we batch all pending destructions
538 	 * asynchronously.
539 	 */
540 	mutex_lock(&slab_mutex);
541 	list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy);
542 	mutex_unlock(&slab_mutex);
543 
544 	if (list_empty(&to_destroy))
545 		return;
546 
547 	rcu_barrier();
548 
549 	list_for_each_entry_safe(s, s2, &to_destroy, list) {
550 #ifdef SLAB_SUPPORTS_SYSFS
551 		sysfs_slab_release(s);
552 #else
553 		slab_kmem_cache_release(s);
554 #endif
555 	}
556 }
557 
558 static int shutdown_cache(struct kmem_cache *s)
559 {
560 	/* free asan quarantined objects */
561 	kasan_cache_shutdown(s);
562 
563 	if (__kmem_cache_shutdown(s) != 0)
564 		return -EBUSY;
565 
566 	memcg_unlink_cache(s);
567 	list_del(&s->list);
568 
569 	if (s->flags & SLAB_TYPESAFE_BY_RCU) {
570 #ifdef SLAB_SUPPORTS_SYSFS
571 		sysfs_slab_unlink(s);
572 #endif
573 		list_add_tail(&s->list, &slab_caches_to_rcu_destroy);
574 		schedule_work(&slab_caches_to_rcu_destroy_work);
575 	} else {
576 #ifdef SLAB_SUPPORTS_SYSFS
577 		sysfs_slab_unlink(s);
578 		sysfs_slab_release(s);
579 #else
580 		slab_kmem_cache_release(s);
581 #endif
582 	}
583 
584 	return 0;
585 }
586 
587 #ifdef CONFIG_MEMCG_KMEM
588 /*
589  * memcg_create_kmem_cache - Create a cache for a memory cgroup.
590  * @memcg: The memory cgroup the new cache is for.
591  * @root_cache: The parent of the new cache.
592  *
593  * This function attempts to create a kmem cache that will serve allocation
594  * requests going from @memcg to @root_cache. The new cache inherits properties
595  * from its parent.
596  */
597 void memcg_create_kmem_cache(struct mem_cgroup *memcg,
598 			     struct kmem_cache *root_cache)
599 {
600 	static char memcg_name_buf[NAME_MAX + 1]; /* protected by slab_mutex */
601 	struct cgroup_subsys_state *css = &memcg->css;
602 	struct memcg_cache_array *arr;
603 	struct kmem_cache *s = NULL;
604 	char *cache_name;
605 	int idx;
606 
607 	get_online_cpus();
608 	get_online_mems();
609 
610 	mutex_lock(&slab_mutex);
611 
612 	/*
613 	 * The memory cgroup could have been offlined while the cache
614 	 * creation work was pending.
615 	 */
616 	if (memcg->kmem_state != KMEM_ONLINE || root_cache->memcg_params.dying)
617 		goto out_unlock;
618 
619 	idx = memcg_cache_id(memcg);
620 	arr = rcu_dereference_protected(root_cache->memcg_params.memcg_caches,
621 					lockdep_is_held(&slab_mutex));
622 
623 	/*
624 	 * Since per-memcg caches are created asynchronously on first
625 	 * allocation (see memcg_kmem_get_cache()), several threads can try to
626 	 * create the same cache, but only one of them may succeed.
627 	 */
628 	if (arr->entries[idx])
629 		goto out_unlock;
630 
631 	cgroup_name(css->cgroup, memcg_name_buf, sizeof(memcg_name_buf));
632 	cache_name = kasprintf(GFP_KERNEL, "%s(%llu:%s)", root_cache->name,
633 			       css->serial_nr, memcg_name_buf);
634 	if (!cache_name)
635 		goto out_unlock;
636 
637 	s = create_cache(cache_name, root_cache->object_size,
638 			 root_cache->align,
639 			 root_cache->flags & CACHE_CREATE_MASK,
640 			 root_cache->useroffset, root_cache->usersize,
641 			 root_cache->ctor, memcg, root_cache);
642 	/*
643 	 * If we could not create a memcg cache, do not complain, because
644 	 * that's not critical at all as we can always proceed with the root
645 	 * cache.
646 	 */
647 	if (IS_ERR(s)) {
648 		kfree(cache_name);
649 		goto out_unlock;
650 	}
651 
652 	/*
653 	 * Since readers won't lock (see cache_from_memcg_idx()), we need a
654 	 * barrier here to ensure nobody will see the kmem_cache partially
655 	 * initialized.
656 	 */
657 	smp_wmb();
658 	arr->entries[idx] = s;
659 
660 out_unlock:
661 	mutex_unlock(&slab_mutex);
662 
663 	put_online_mems();
664 	put_online_cpus();
665 }
666 
667 static void kmemcg_deactivate_workfn(struct work_struct *work)
668 {
669 	struct kmem_cache *s = container_of(work, struct kmem_cache,
670 					    memcg_params.deact_work);
671 
672 	get_online_cpus();
673 	get_online_mems();
674 
675 	mutex_lock(&slab_mutex);
676 
677 	s->memcg_params.deact_fn(s);
678 
679 	mutex_unlock(&slab_mutex);
680 
681 	put_online_mems();
682 	put_online_cpus();
683 
684 	/* done, put the ref from slab_deactivate_memcg_cache_rcu_sched() */
685 	css_put(&s->memcg_params.memcg->css);
686 }
687 
688 static void kmemcg_deactivate_rcufn(struct rcu_head *head)
689 {
690 	struct kmem_cache *s = container_of(head, struct kmem_cache,
691 					    memcg_params.deact_rcu_head);
692 
693 	/*
694 	 * We need to grab blocking locks.  Bounce to ->deact_work.  The
695 	 * work item shares the space with the RCU head and can't be
696 	 * initialized eariler.
697 	 */
698 	INIT_WORK(&s->memcg_params.deact_work, kmemcg_deactivate_workfn);
699 	queue_work(memcg_kmem_cache_wq, &s->memcg_params.deact_work);
700 }
701 
702 /**
703  * slab_deactivate_memcg_cache_rcu_sched - schedule deactivation after a
704  *					   sched RCU grace period
705  * @s: target kmem_cache
706  * @deact_fn: deactivation function to call
707  *
708  * Schedule @deact_fn to be invoked with online cpus, mems and slab_mutex
709  * held after a sched RCU grace period.  The slab is guaranteed to stay
710  * alive until @deact_fn is finished.  This is to be used from
711  * __kmemcg_cache_deactivate().
712  */
713 void slab_deactivate_memcg_cache_rcu_sched(struct kmem_cache *s,
714 					   void (*deact_fn)(struct kmem_cache *))
715 {
716 	if (WARN_ON_ONCE(is_root_cache(s)) ||
717 	    WARN_ON_ONCE(s->memcg_params.deact_fn))
718 		return;
719 
720 	if (s->memcg_params.root_cache->memcg_params.dying)
721 		return;
722 
723 	/* pin memcg so that @s doesn't get destroyed in the middle */
724 	css_get(&s->memcg_params.memcg->css);
725 
726 	s->memcg_params.deact_fn = deact_fn;
727 	call_rcu_sched(&s->memcg_params.deact_rcu_head, kmemcg_deactivate_rcufn);
728 }
729 
730 void memcg_deactivate_kmem_caches(struct mem_cgroup *memcg)
731 {
732 	int idx;
733 	struct memcg_cache_array *arr;
734 	struct kmem_cache *s, *c;
735 
736 	idx = memcg_cache_id(memcg);
737 
738 	get_online_cpus();
739 	get_online_mems();
740 
741 	mutex_lock(&slab_mutex);
742 	list_for_each_entry(s, &slab_root_caches, root_caches_node) {
743 		arr = rcu_dereference_protected(s->memcg_params.memcg_caches,
744 						lockdep_is_held(&slab_mutex));
745 		c = arr->entries[idx];
746 		if (!c)
747 			continue;
748 
749 		__kmemcg_cache_deactivate(c);
750 		arr->entries[idx] = NULL;
751 	}
752 	mutex_unlock(&slab_mutex);
753 
754 	put_online_mems();
755 	put_online_cpus();
756 }
757 
758 void memcg_destroy_kmem_caches(struct mem_cgroup *memcg)
759 {
760 	struct kmem_cache *s, *s2;
761 
762 	get_online_cpus();
763 	get_online_mems();
764 
765 	mutex_lock(&slab_mutex);
766 	list_for_each_entry_safe(s, s2, &memcg->kmem_caches,
767 				 memcg_params.kmem_caches_node) {
768 		/*
769 		 * The cgroup is about to be freed and therefore has no charges
770 		 * left. Hence, all its caches must be empty by now.
771 		 */
772 		BUG_ON(shutdown_cache(s));
773 	}
774 	mutex_unlock(&slab_mutex);
775 
776 	put_online_mems();
777 	put_online_cpus();
778 }
779 
780 static int shutdown_memcg_caches(struct kmem_cache *s)
781 {
782 	struct memcg_cache_array *arr;
783 	struct kmem_cache *c, *c2;
784 	LIST_HEAD(busy);
785 	int i;
786 
787 	BUG_ON(!is_root_cache(s));
788 
789 	/*
790 	 * First, shutdown active caches, i.e. caches that belong to online
791 	 * memory cgroups.
792 	 */
793 	arr = rcu_dereference_protected(s->memcg_params.memcg_caches,
794 					lockdep_is_held(&slab_mutex));
795 	for_each_memcg_cache_index(i) {
796 		c = arr->entries[i];
797 		if (!c)
798 			continue;
799 		if (shutdown_cache(c))
800 			/*
801 			 * The cache still has objects. Move it to a temporary
802 			 * list so as not to try to destroy it for a second
803 			 * time while iterating over inactive caches below.
804 			 */
805 			list_move(&c->memcg_params.children_node, &busy);
806 		else
807 			/*
808 			 * The cache is empty and will be destroyed soon. Clear
809 			 * the pointer to it in the memcg_caches array so that
810 			 * it will never be accessed even if the root cache
811 			 * stays alive.
812 			 */
813 			arr->entries[i] = NULL;
814 	}
815 
816 	/*
817 	 * Second, shutdown all caches left from memory cgroups that are now
818 	 * offline.
819 	 */
820 	list_for_each_entry_safe(c, c2, &s->memcg_params.children,
821 				 memcg_params.children_node)
822 		shutdown_cache(c);
823 
824 	list_splice(&busy, &s->memcg_params.children);
825 
826 	/*
827 	 * A cache being destroyed must be empty. In particular, this means
828 	 * that all per memcg caches attached to it must be empty too.
829 	 */
830 	if (!list_empty(&s->memcg_params.children))
831 		return -EBUSY;
832 	return 0;
833 }
834 
835 static void flush_memcg_workqueue(struct kmem_cache *s)
836 {
837 	mutex_lock(&slab_mutex);
838 	s->memcg_params.dying = true;
839 	mutex_unlock(&slab_mutex);
840 
841 	/*
842 	 * SLUB deactivates the kmem_caches through call_rcu_sched. Make
843 	 * sure all registered rcu callbacks have been invoked.
844 	 */
845 	if (IS_ENABLED(CONFIG_SLUB))
846 		rcu_barrier_sched();
847 
848 	/*
849 	 * SLAB and SLUB create memcg kmem_caches through workqueue and SLUB
850 	 * deactivates the memcg kmem_caches through workqueue. Make sure all
851 	 * previous workitems on workqueue are processed.
852 	 */
853 	flush_workqueue(memcg_kmem_cache_wq);
854 }
855 #else
856 static inline int shutdown_memcg_caches(struct kmem_cache *s)
857 {
858 	return 0;
859 }
860 
861 static inline void flush_memcg_workqueue(struct kmem_cache *s)
862 {
863 }
864 #endif /* CONFIG_MEMCG_KMEM */
865 
866 void slab_kmem_cache_release(struct kmem_cache *s)
867 {
868 	__kmem_cache_release(s);
869 	destroy_memcg_params(s);
870 	kfree_const(s->name);
871 	kmem_cache_free(kmem_cache, s);
872 }
873 
874 void kmem_cache_destroy(struct kmem_cache *s)
875 {
876 	int err;
877 
878 	if (unlikely(!s))
879 		return;
880 
881 	flush_memcg_workqueue(s);
882 
883 	get_online_cpus();
884 	get_online_mems();
885 
886 	mutex_lock(&slab_mutex);
887 
888 	s->refcount--;
889 	if (s->refcount)
890 		goto out_unlock;
891 
892 	err = shutdown_memcg_caches(s);
893 	if (!err)
894 		err = shutdown_cache(s);
895 
896 	if (err) {
897 		pr_err("kmem_cache_destroy %s: Slab cache still has objects\n",
898 		       s->name);
899 		dump_stack();
900 	}
901 out_unlock:
902 	mutex_unlock(&slab_mutex);
903 
904 	put_online_mems();
905 	put_online_cpus();
906 }
907 EXPORT_SYMBOL(kmem_cache_destroy);
908 
909 /**
910  * kmem_cache_shrink - Shrink a cache.
911  * @cachep: The cache to shrink.
912  *
913  * Releases as many slabs as possible for a cache.
914  * To help debugging, a zero exit status indicates all slabs were released.
915  */
916 int kmem_cache_shrink(struct kmem_cache *cachep)
917 {
918 	int ret;
919 
920 	get_online_cpus();
921 	get_online_mems();
922 	kasan_cache_shrink(cachep);
923 	ret = __kmem_cache_shrink(cachep);
924 	put_online_mems();
925 	put_online_cpus();
926 	return ret;
927 }
928 EXPORT_SYMBOL(kmem_cache_shrink);
929 
930 bool slab_is_available(void)
931 {
932 	return slab_state >= UP;
933 }
934 
935 #ifndef CONFIG_SLOB
936 /* Create a cache during boot when no slab services are available yet */
937 void __init create_boot_cache(struct kmem_cache *s, const char *name,
938 		unsigned int size, slab_flags_t flags,
939 		unsigned int useroffset, unsigned int usersize)
940 {
941 	int err;
942 
943 	s->name = name;
944 	s->size = s->object_size = size;
945 	s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size);
946 	s->useroffset = useroffset;
947 	s->usersize = usersize;
948 
949 	slab_init_memcg_params(s);
950 
951 	err = __kmem_cache_create(s, flags);
952 
953 	if (err)
954 		panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n",
955 					name, size, err);
956 
957 	s->refcount = -1;	/* Exempt from merging for now */
958 }
959 
960 struct kmem_cache *__init create_kmalloc_cache(const char *name,
961 		unsigned int size, slab_flags_t flags,
962 		unsigned int useroffset, unsigned int usersize)
963 {
964 	struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
965 
966 	if (!s)
967 		panic("Out of memory when creating slab %s\n", name);
968 
969 	create_boot_cache(s, name, size, flags, useroffset, usersize);
970 	list_add(&s->list, &slab_caches);
971 	memcg_link_cache(s);
972 	s->refcount = 1;
973 	return s;
974 }
975 
976 struct kmem_cache *
977 kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1] __ro_after_init;
978 EXPORT_SYMBOL(kmalloc_caches);
979 
980 /*
981  * Conversion table for small slabs sizes / 8 to the index in the
982  * kmalloc array. This is necessary for slabs < 192 since we have non power
983  * of two cache sizes there. The size of larger slabs can be determined using
984  * fls.
985  */
986 static u8 size_index[24] __ro_after_init = {
987 	3,	/* 8 */
988 	4,	/* 16 */
989 	5,	/* 24 */
990 	5,	/* 32 */
991 	6,	/* 40 */
992 	6,	/* 48 */
993 	6,	/* 56 */
994 	6,	/* 64 */
995 	1,	/* 72 */
996 	1,	/* 80 */
997 	1,	/* 88 */
998 	1,	/* 96 */
999 	7,	/* 104 */
1000 	7,	/* 112 */
1001 	7,	/* 120 */
1002 	7,	/* 128 */
1003 	2,	/* 136 */
1004 	2,	/* 144 */
1005 	2,	/* 152 */
1006 	2,	/* 160 */
1007 	2,	/* 168 */
1008 	2,	/* 176 */
1009 	2,	/* 184 */
1010 	2	/* 192 */
1011 };
1012 
1013 static inline unsigned int size_index_elem(unsigned int bytes)
1014 {
1015 	return (bytes - 1) / 8;
1016 }
1017 
1018 /*
1019  * Find the kmem_cache structure that serves a given size of
1020  * allocation
1021  */
1022 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
1023 {
1024 	unsigned int index;
1025 
1026 	if (size <= 192) {
1027 		if (!size)
1028 			return ZERO_SIZE_PTR;
1029 
1030 		index = size_index[size_index_elem(size)];
1031 	} else {
1032 		if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
1033 			WARN_ON(1);
1034 			return NULL;
1035 		}
1036 		index = fls(size - 1);
1037 	}
1038 
1039 	return kmalloc_caches[kmalloc_type(flags)][index];
1040 }
1041 
1042 /*
1043  * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
1044  * kmalloc_index() supports up to 2^26=64MB, so the final entry of the table is
1045  * kmalloc-67108864.
1046  */
1047 const struct kmalloc_info_struct kmalloc_info[] __initconst = {
1048 	{NULL,                      0},		{"kmalloc-96",             96},
1049 	{"kmalloc-192",           192},		{"kmalloc-8",               8},
1050 	{"kmalloc-16",             16},		{"kmalloc-32",             32},
1051 	{"kmalloc-64",             64},		{"kmalloc-128",           128},
1052 	{"kmalloc-256",           256},		{"kmalloc-512",           512},
1053 	{"kmalloc-1k",           1024},		{"kmalloc-2k",           2048},
1054 	{"kmalloc-4k",           4096},		{"kmalloc-8k",           8192},
1055 	{"kmalloc-16k",         16384},		{"kmalloc-32k",         32768},
1056 	{"kmalloc-64k",         65536},		{"kmalloc-128k",       131072},
1057 	{"kmalloc-256k",       262144},		{"kmalloc-512k",       524288},
1058 	{"kmalloc-1M",        1048576},		{"kmalloc-2M",        2097152},
1059 	{"kmalloc-4M",        4194304},		{"kmalloc-8M",        8388608},
1060 	{"kmalloc-16M",      16777216},		{"kmalloc-32M",      33554432},
1061 	{"kmalloc-64M",      67108864}
1062 };
1063 
1064 /*
1065  * Patch up the size_index table if we have strange large alignment
1066  * requirements for the kmalloc array. This is only the case for
1067  * MIPS it seems. The standard arches will not generate any code here.
1068  *
1069  * Largest permitted alignment is 256 bytes due to the way we
1070  * handle the index determination for the smaller caches.
1071  *
1072  * Make sure that nothing crazy happens if someone starts tinkering
1073  * around with ARCH_KMALLOC_MINALIGN
1074  */
1075 void __init setup_kmalloc_cache_index_table(void)
1076 {
1077 	unsigned int i;
1078 
1079 	BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
1080 		(KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
1081 
1082 	for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
1083 		unsigned int elem = size_index_elem(i);
1084 
1085 		if (elem >= ARRAY_SIZE(size_index))
1086 			break;
1087 		size_index[elem] = KMALLOC_SHIFT_LOW;
1088 	}
1089 
1090 	if (KMALLOC_MIN_SIZE >= 64) {
1091 		/*
1092 		 * The 96 byte size cache is not used if the alignment
1093 		 * is 64 byte.
1094 		 */
1095 		for (i = 64 + 8; i <= 96; i += 8)
1096 			size_index[size_index_elem(i)] = 7;
1097 
1098 	}
1099 
1100 	if (KMALLOC_MIN_SIZE >= 128) {
1101 		/*
1102 		 * The 192 byte sized cache is not used if the alignment
1103 		 * is 128 byte. Redirect kmalloc to use the 256 byte cache
1104 		 * instead.
1105 		 */
1106 		for (i = 128 + 8; i <= 192; i += 8)
1107 			size_index[size_index_elem(i)] = 8;
1108 	}
1109 }
1110 
1111 static const char *
1112 kmalloc_cache_name(const char *prefix, unsigned int size)
1113 {
1114 
1115 	static const char units[3] = "\0kM";
1116 	int idx = 0;
1117 
1118 	while (size >= 1024 && (size % 1024 == 0)) {
1119 		size /= 1024;
1120 		idx++;
1121 	}
1122 
1123 	return kasprintf(GFP_NOWAIT, "%s-%u%c", prefix, size, units[idx]);
1124 }
1125 
1126 static void __init
1127 new_kmalloc_cache(int idx, int type, slab_flags_t flags)
1128 {
1129 	const char *name;
1130 
1131 	if (type == KMALLOC_RECLAIM) {
1132 		flags |= SLAB_RECLAIM_ACCOUNT;
1133 		name = kmalloc_cache_name("kmalloc-rcl",
1134 						kmalloc_info[idx].size);
1135 		BUG_ON(!name);
1136 	} else {
1137 		name = kmalloc_info[idx].name;
1138 	}
1139 
1140 	kmalloc_caches[type][idx] = create_kmalloc_cache(name,
1141 					kmalloc_info[idx].size, flags, 0,
1142 					kmalloc_info[idx].size);
1143 }
1144 
1145 /*
1146  * Create the kmalloc array. Some of the regular kmalloc arrays
1147  * may already have been created because they were needed to
1148  * enable allocations for slab creation.
1149  */
1150 void __init create_kmalloc_caches(slab_flags_t flags)
1151 {
1152 	int i, type;
1153 
1154 	for (type = KMALLOC_NORMAL; type <= KMALLOC_RECLAIM; type++) {
1155 		for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
1156 			if (!kmalloc_caches[type][i])
1157 				new_kmalloc_cache(i, type, flags);
1158 
1159 			/*
1160 			 * Caches that are not of the two-to-the-power-of size.
1161 			 * These have to be created immediately after the
1162 			 * earlier power of two caches
1163 			 */
1164 			if (KMALLOC_MIN_SIZE <= 32 && i == 6 &&
1165 					!kmalloc_caches[type][1])
1166 				new_kmalloc_cache(1, type, flags);
1167 			if (KMALLOC_MIN_SIZE <= 64 && i == 7 &&
1168 					!kmalloc_caches[type][2])
1169 				new_kmalloc_cache(2, type, flags);
1170 		}
1171 	}
1172 
1173 	/* Kmalloc array is now usable */
1174 	slab_state = UP;
1175 
1176 #ifdef CONFIG_ZONE_DMA
1177 	for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
1178 		struct kmem_cache *s = kmalloc_caches[KMALLOC_NORMAL][i];
1179 
1180 		if (s) {
1181 			unsigned int size = kmalloc_size(i);
1182 			const char *n = kmalloc_cache_name("dma-kmalloc", size);
1183 
1184 			BUG_ON(!n);
1185 			kmalloc_caches[KMALLOC_DMA][i] = create_kmalloc_cache(
1186 				n, size, SLAB_CACHE_DMA | flags, 0, 0);
1187 		}
1188 	}
1189 #endif
1190 }
1191 #endif /* !CONFIG_SLOB */
1192 
1193 /*
1194  * To avoid unnecessary overhead, we pass through large allocation requests
1195  * directly to the page allocator. We use __GFP_COMP, because we will need to
1196  * know the allocation order to free the pages properly in kfree.
1197  */
1198 void *kmalloc_order(size_t size, gfp_t flags, unsigned int order)
1199 {
1200 	void *ret;
1201 	struct page *page;
1202 
1203 	flags |= __GFP_COMP;
1204 	page = alloc_pages(flags, order);
1205 	ret = page ? page_address(page) : NULL;
1206 	kmemleak_alloc(ret, size, 1, flags);
1207 	kasan_kmalloc_large(ret, size, flags);
1208 	return ret;
1209 }
1210 EXPORT_SYMBOL(kmalloc_order);
1211 
1212 #ifdef CONFIG_TRACING
1213 void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
1214 {
1215 	void *ret = kmalloc_order(size, flags, order);
1216 	trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
1217 	return ret;
1218 }
1219 EXPORT_SYMBOL(kmalloc_order_trace);
1220 #endif
1221 
1222 #ifdef CONFIG_SLAB_FREELIST_RANDOM
1223 /* Randomize a generic freelist */
1224 static void freelist_randomize(struct rnd_state *state, unsigned int *list,
1225 			       unsigned int count)
1226 {
1227 	unsigned int rand;
1228 	unsigned int i;
1229 
1230 	for (i = 0; i < count; i++)
1231 		list[i] = i;
1232 
1233 	/* Fisher-Yates shuffle */
1234 	for (i = count - 1; i > 0; i--) {
1235 		rand = prandom_u32_state(state);
1236 		rand %= (i + 1);
1237 		swap(list[i], list[rand]);
1238 	}
1239 }
1240 
1241 /* Create a random sequence per cache */
1242 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
1243 				    gfp_t gfp)
1244 {
1245 	struct rnd_state state;
1246 
1247 	if (count < 2 || cachep->random_seq)
1248 		return 0;
1249 
1250 	cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp);
1251 	if (!cachep->random_seq)
1252 		return -ENOMEM;
1253 
1254 	/* Get best entropy at this stage of boot */
1255 	prandom_seed_state(&state, get_random_long());
1256 
1257 	freelist_randomize(&state, cachep->random_seq, count);
1258 	return 0;
1259 }
1260 
1261 /* Destroy the per-cache random freelist sequence */
1262 void cache_random_seq_destroy(struct kmem_cache *cachep)
1263 {
1264 	kfree(cachep->random_seq);
1265 	cachep->random_seq = NULL;
1266 }
1267 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
1268 
1269 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
1270 #ifdef CONFIG_SLAB
1271 #define SLABINFO_RIGHTS (0600)
1272 #else
1273 #define SLABINFO_RIGHTS (0400)
1274 #endif
1275 
1276 static void print_slabinfo_header(struct seq_file *m)
1277 {
1278 	/*
1279 	 * Output format version, so at least we can change it
1280 	 * without _too_ many complaints.
1281 	 */
1282 #ifdef CONFIG_DEBUG_SLAB
1283 	seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
1284 #else
1285 	seq_puts(m, "slabinfo - version: 2.1\n");
1286 #endif
1287 	seq_puts(m, "# name            <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
1288 	seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
1289 	seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
1290 #ifdef CONFIG_DEBUG_SLAB
1291 	seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
1292 	seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
1293 #endif
1294 	seq_putc(m, '\n');
1295 }
1296 
1297 void *slab_start(struct seq_file *m, loff_t *pos)
1298 {
1299 	mutex_lock(&slab_mutex);
1300 	return seq_list_start(&slab_root_caches, *pos);
1301 }
1302 
1303 void *slab_next(struct seq_file *m, void *p, loff_t *pos)
1304 {
1305 	return seq_list_next(p, &slab_root_caches, pos);
1306 }
1307 
1308 void slab_stop(struct seq_file *m, void *p)
1309 {
1310 	mutex_unlock(&slab_mutex);
1311 }
1312 
1313 static void
1314 memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info)
1315 {
1316 	struct kmem_cache *c;
1317 	struct slabinfo sinfo;
1318 
1319 	if (!is_root_cache(s))
1320 		return;
1321 
1322 	for_each_memcg_cache(c, s) {
1323 		memset(&sinfo, 0, sizeof(sinfo));
1324 		get_slabinfo(c, &sinfo);
1325 
1326 		info->active_slabs += sinfo.active_slabs;
1327 		info->num_slabs += sinfo.num_slabs;
1328 		info->shared_avail += sinfo.shared_avail;
1329 		info->active_objs += sinfo.active_objs;
1330 		info->num_objs += sinfo.num_objs;
1331 	}
1332 }
1333 
1334 static void cache_show(struct kmem_cache *s, struct seq_file *m)
1335 {
1336 	struct slabinfo sinfo;
1337 
1338 	memset(&sinfo, 0, sizeof(sinfo));
1339 	get_slabinfo(s, &sinfo);
1340 
1341 	memcg_accumulate_slabinfo(s, &sinfo);
1342 
1343 	seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
1344 		   cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size,
1345 		   sinfo.objects_per_slab, (1 << sinfo.cache_order));
1346 
1347 	seq_printf(m, " : tunables %4u %4u %4u",
1348 		   sinfo.limit, sinfo.batchcount, sinfo.shared);
1349 	seq_printf(m, " : slabdata %6lu %6lu %6lu",
1350 		   sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
1351 	slabinfo_show_stats(m, s);
1352 	seq_putc(m, '\n');
1353 }
1354 
1355 static int slab_show(struct seq_file *m, void *p)
1356 {
1357 	struct kmem_cache *s = list_entry(p, struct kmem_cache, root_caches_node);
1358 
1359 	if (p == slab_root_caches.next)
1360 		print_slabinfo_header(m);
1361 	cache_show(s, m);
1362 	return 0;
1363 }
1364 
1365 void dump_unreclaimable_slab(void)
1366 {
1367 	struct kmem_cache *s, *s2;
1368 	struct slabinfo sinfo;
1369 
1370 	/*
1371 	 * Here acquiring slab_mutex is risky since we don't prefer to get
1372 	 * sleep in oom path. But, without mutex hold, it may introduce a
1373 	 * risk of crash.
1374 	 * Use mutex_trylock to protect the list traverse, dump nothing
1375 	 * without acquiring the mutex.
1376 	 */
1377 	if (!mutex_trylock(&slab_mutex)) {
1378 		pr_warn("excessive unreclaimable slab but cannot dump stats\n");
1379 		return;
1380 	}
1381 
1382 	pr_info("Unreclaimable slab info:\n");
1383 	pr_info("Name                      Used          Total\n");
1384 
1385 	list_for_each_entry_safe(s, s2, &slab_caches, list) {
1386 		if (!is_root_cache(s) || (s->flags & SLAB_RECLAIM_ACCOUNT))
1387 			continue;
1388 
1389 		get_slabinfo(s, &sinfo);
1390 
1391 		if (sinfo.num_objs > 0)
1392 			pr_info("%-17s %10luKB %10luKB\n", cache_name(s),
1393 				(sinfo.active_objs * s->size) / 1024,
1394 				(sinfo.num_objs * s->size) / 1024);
1395 	}
1396 	mutex_unlock(&slab_mutex);
1397 }
1398 
1399 #if defined(CONFIG_MEMCG)
1400 void *memcg_slab_start(struct seq_file *m, loff_t *pos)
1401 {
1402 	struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
1403 
1404 	mutex_lock(&slab_mutex);
1405 	return seq_list_start(&memcg->kmem_caches, *pos);
1406 }
1407 
1408 void *memcg_slab_next(struct seq_file *m, void *p, loff_t *pos)
1409 {
1410 	struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
1411 
1412 	return seq_list_next(p, &memcg->kmem_caches, pos);
1413 }
1414 
1415 void memcg_slab_stop(struct seq_file *m, void *p)
1416 {
1417 	mutex_unlock(&slab_mutex);
1418 }
1419 
1420 int memcg_slab_show(struct seq_file *m, void *p)
1421 {
1422 	struct kmem_cache *s = list_entry(p, struct kmem_cache,
1423 					  memcg_params.kmem_caches_node);
1424 	struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
1425 
1426 	if (p == memcg->kmem_caches.next)
1427 		print_slabinfo_header(m);
1428 	cache_show(s, m);
1429 	return 0;
1430 }
1431 #endif
1432 
1433 /*
1434  * slabinfo_op - iterator that generates /proc/slabinfo
1435  *
1436  * Output layout:
1437  * cache-name
1438  * num-active-objs
1439  * total-objs
1440  * object size
1441  * num-active-slabs
1442  * total-slabs
1443  * num-pages-per-slab
1444  * + further values on SMP and with statistics enabled
1445  */
1446 static const struct seq_operations slabinfo_op = {
1447 	.start = slab_start,
1448 	.next = slab_next,
1449 	.stop = slab_stop,
1450 	.show = slab_show,
1451 };
1452 
1453 static int slabinfo_open(struct inode *inode, struct file *file)
1454 {
1455 	return seq_open(file, &slabinfo_op);
1456 }
1457 
1458 static const struct file_operations proc_slabinfo_operations = {
1459 	.open		= slabinfo_open,
1460 	.read		= seq_read,
1461 	.write          = slabinfo_write,
1462 	.llseek		= seq_lseek,
1463 	.release	= seq_release,
1464 };
1465 
1466 static int __init slab_proc_init(void)
1467 {
1468 	proc_create("slabinfo", SLABINFO_RIGHTS, NULL,
1469 						&proc_slabinfo_operations);
1470 	return 0;
1471 }
1472 module_init(slab_proc_init);
1473 #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */
1474 
1475 static __always_inline void *__do_krealloc(const void *p, size_t new_size,
1476 					   gfp_t flags)
1477 {
1478 	void *ret;
1479 	size_t ks = 0;
1480 
1481 	if (p)
1482 		ks = ksize(p);
1483 
1484 	if (ks >= new_size) {
1485 		kasan_krealloc((void *)p, new_size, flags);
1486 		return (void *)p;
1487 	}
1488 
1489 	ret = kmalloc_track_caller(new_size, flags);
1490 	if (ret && p)
1491 		memcpy(ret, p, ks);
1492 
1493 	return ret;
1494 }
1495 
1496 /**
1497  * __krealloc - like krealloc() but don't free @p.
1498  * @p: object to reallocate memory for.
1499  * @new_size: how many bytes of memory are required.
1500  * @flags: the type of memory to allocate.
1501  *
1502  * This function is like krealloc() except it never frees the originally
1503  * allocated buffer. Use this if you don't want to free the buffer immediately
1504  * like, for example, with RCU.
1505  */
1506 void *__krealloc(const void *p, size_t new_size, gfp_t flags)
1507 {
1508 	if (unlikely(!new_size))
1509 		return ZERO_SIZE_PTR;
1510 
1511 	return __do_krealloc(p, new_size, flags);
1512 
1513 }
1514 EXPORT_SYMBOL(__krealloc);
1515 
1516 /**
1517  * krealloc - reallocate memory. The contents will remain unchanged.
1518  * @p: object to reallocate memory for.
1519  * @new_size: how many bytes of memory are required.
1520  * @flags: the type of memory to allocate.
1521  *
1522  * The contents of the object pointed to are preserved up to the
1523  * lesser of the new and old sizes.  If @p is %NULL, krealloc()
1524  * behaves exactly like kmalloc().  If @new_size is 0 and @p is not a
1525  * %NULL pointer, the object pointed to is freed.
1526  */
1527 void *krealloc(const void *p, size_t new_size, gfp_t flags)
1528 {
1529 	void *ret;
1530 
1531 	if (unlikely(!new_size)) {
1532 		kfree(p);
1533 		return ZERO_SIZE_PTR;
1534 	}
1535 
1536 	ret = __do_krealloc(p, new_size, flags);
1537 	if (ret && p != ret)
1538 		kfree(p);
1539 
1540 	return ret;
1541 }
1542 EXPORT_SYMBOL(krealloc);
1543 
1544 /**
1545  * kzfree - like kfree but zero memory
1546  * @p: object to free memory of
1547  *
1548  * The memory of the object @p points to is zeroed before freed.
1549  * If @p is %NULL, kzfree() does nothing.
1550  *
1551  * Note: this function zeroes the whole allocated buffer which can be a good
1552  * deal bigger than the requested buffer size passed to kmalloc(). So be
1553  * careful when using this function in performance sensitive code.
1554  */
1555 void kzfree(const void *p)
1556 {
1557 	size_t ks;
1558 	void *mem = (void *)p;
1559 
1560 	if (unlikely(ZERO_OR_NULL_PTR(mem)))
1561 		return;
1562 	ks = ksize(mem);
1563 	memset(mem, 0, ks);
1564 	kfree(mem);
1565 }
1566 EXPORT_SYMBOL(kzfree);
1567 
1568 /* Tracepoints definitions. */
1569 EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1570 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1571 EXPORT_TRACEPOINT_SYMBOL(kmalloc_node);
1572 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node);
1573 EXPORT_TRACEPOINT_SYMBOL(kfree);
1574 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);
1575 
1576 int should_failslab(struct kmem_cache *s, gfp_t gfpflags)
1577 {
1578 	if (__should_failslab(s, gfpflags))
1579 		return -ENOMEM;
1580 	return 0;
1581 }
1582 ALLOW_ERROR_INJECTION(should_failslab, ERRNO);
1583