xref: /linux/mm/slab_common.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
1 /*
2  * Slab allocator functions that are independent of the allocator strategy
3  *
4  * (C) 2012 Christoph Lameter <cl@linux.com>
5  */
6 #include <linux/slab.h>
7 
8 #include <linux/mm.h>
9 #include <linux/poison.h>
10 #include <linux/interrupt.h>
11 #include <linux/memory.h>
12 #include <linux/compiler.h>
13 #include <linux/module.h>
14 #include <linux/cpu.h>
15 #include <linux/uaccess.h>
16 #include <linux/seq_file.h>
17 #include <linux/proc_fs.h>
18 #include <asm/cacheflush.h>
19 #include <asm/tlbflush.h>
20 #include <asm/page.h>
21 #include <linux/memcontrol.h>
22 
23 #define CREATE_TRACE_POINTS
24 #include <trace/events/kmem.h>
25 
26 #include "slab.h"
27 
28 enum slab_state slab_state;
29 LIST_HEAD(slab_caches);
30 DEFINE_MUTEX(slab_mutex);
31 struct kmem_cache *kmem_cache;
32 
33 /*
34  * Set of flags that will prevent slab merging
35  */
36 #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
37 		SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \
38 		SLAB_FAILSLAB)
39 
40 #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | SLAB_NOTRACK)
41 
42 /*
43  * Merge control. If this is set then no merging of slab caches will occur.
44  * (Could be removed. This was introduced to pacify the merge skeptics.)
45  */
46 static int slab_nomerge;
47 
48 static int __init setup_slab_nomerge(char *str)
49 {
50 	slab_nomerge = 1;
51 	return 1;
52 }
53 
54 #ifdef CONFIG_SLUB
55 __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0);
56 #endif
57 
58 __setup("slab_nomerge", setup_slab_nomerge);
59 
60 /*
61  * Determine the size of a slab object
62  */
63 unsigned int kmem_cache_size(struct kmem_cache *s)
64 {
65 	return s->object_size;
66 }
67 EXPORT_SYMBOL(kmem_cache_size);
68 
69 #ifdef CONFIG_DEBUG_VM
70 static int kmem_cache_sanity_check(const char *name, size_t size)
71 {
72 	struct kmem_cache *s = NULL;
73 
74 	if (!name || in_interrupt() || size < sizeof(void *) ||
75 		size > KMALLOC_MAX_SIZE) {
76 		pr_err("kmem_cache_create(%s) integrity check failed\n", name);
77 		return -EINVAL;
78 	}
79 
80 	list_for_each_entry(s, &slab_caches, list) {
81 		char tmp;
82 		int res;
83 
84 		/*
85 		 * This happens when the module gets unloaded and doesn't
86 		 * destroy its slab cache and no-one else reuses the vmalloc
87 		 * area of the module.  Print a warning.
88 		 */
89 		res = probe_kernel_address(s->name, tmp);
90 		if (res) {
91 			pr_err("Slab cache with size %d has lost its name\n",
92 			       s->object_size);
93 			continue;
94 		}
95 	}
96 
97 	WARN_ON(strchr(name, ' '));	/* It confuses parsers */
98 	return 0;
99 }
100 #else
101 static inline int kmem_cache_sanity_check(const char *name, size_t size)
102 {
103 	return 0;
104 }
105 #endif
106 
107 void __kmem_cache_free_bulk(struct kmem_cache *s, size_t nr, void **p)
108 {
109 	size_t i;
110 
111 	for (i = 0; i < nr; i++)
112 		kmem_cache_free(s, p[i]);
113 }
114 
115 bool __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 false;
125 		}
126 	}
127 	return true;
128 }
129 
130 #ifdef CONFIG_MEMCG_KMEM
131 void slab_init_memcg_params(struct kmem_cache *s)
132 {
133 	s->memcg_params.is_root_cache = true;
134 	INIT_LIST_HEAD(&s->memcg_params.list);
135 	RCU_INIT_POINTER(s->memcg_params.memcg_caches, NULL);
136 }
137 
138 static int init_memcg_params(struct kmem_cache *s,
139 		struct mem_cgroup *memcg, struct kmem_cache *root_cache)
140 {
141 	struct memcg_cache_array *arr;
142 
143 	if (memcg) {
144 		s->memcg_params.is_root_cache = false;
145 		s->memcg_params.memcg = memcg;
146 		s->memcg_params.root_cache = root_cache;
147 		return 0;
148 	}
149 
150 	slab_init_memcg_params(s);
151 
152 	if (!memcg_nr_cache_ids)
153 		return 0;
154 
155 	arr = kzalloc(sizeof(struct memcg_cache_array) +
156 		      memcg_nr_cache_ids * sizeof(void *),
157 		      GFP_KERNEL);
158 	if (!arr)
159 		return -ENOMEM;
160 
161 	RCU_INIT_POINTER(s->memcg_params.memcg_caches, arr);
162 	return 0;
163 }
164 
165 static void destroy_memcg_params(struct kmem_cache *s)
166 {
167 	if (is_root_cache(s))
168 		kfree(rcu_access_pointer(s->memcg_params.memcg_caches));
169 }
170 
171 static int update_memcg_params(struct kmem_cache *s, int new_array_size)
172 {
173 	struct memcg_cache_array *old, *new;
174 
175 	if (!is_root_cache(s))
176 		return 0;
177 
178 	new = kzalloc(sizeof(struct memcg_cache_array) +
179 		      new_array_size * sizeof(void *), GFP_KERNEL);
180 	if (!new)
181 		return -ENOMEM;
182 
183 	old = rcu_dereference_protected(s->memcg_params.memcg_caches,
184 					lockdep_is_held(&slab_mutex));
185 	if (old)
186 		memcpy(new->entries, old->entries,
187 		       memcg_nr_cache_ids * sizeof(void *));
188 
189 	rcu_assign_pointer(s->memcg_params.memcg_caches, new);
190 	if (old)
191 		kfree_rcu(old, rcu);
192 	return 0;
193 }
194 
195 int memcg_update_all_caches(int num_memcgs)
196 {
197 	struct kmem_cache *s;
198 	int ret = 0;
199 
200 	mutex_lock(&slab_mutex);
201 	list_for_each_entry(s, &slab_caches, list) {
202 		ret = update_memcg_params(s, num_memcgs);
203 		/*
204 		 * Instead of freeing the memory, we'll just leave the caches
205 		 * up to this point in an updated state.
206 		 */
207 		if (ret)
208 			break;
209 	}
210 	mutex_unlock(&slab_mutex);
211 	return ret;
212 }
213 #else
214 static inline int init_memcg_params(struct kmem_cache *s,
215 		struct mem_cgroup *memcg, struct kmem_cache *root_cache)
216 {
217 	return 0;
218 }
219 
220 static inline void destroy_memcg_params(struct kmem_cache *s)
221 {
222 }
223 #endif /* CONFIG_MEMCG_KMEM */
224 
225 /*
226  * Find a mergeable slab cache
227  */
228 int slab_unmergeable(struct kmem_cache *s)
229 {
230 	if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
231 		return 1;
232 
233 	if (!is_root_cache(s))
234 		return 1;
235 
236 	if (s->ctor)
237 		return 1;
238 
239 	/*
240 	 * We may have set a slab to be unmergeable during bootstrap.
241 	 */
242 	if (s->refcount < 0)
243 		return 1;
244 
245 	return 0;
246 }
247 
248 struct kmem_cache *find_mergeable(size_t size, size_t align,
249 		unsigned long flags, const char *name, void (*ctor)(void *))
250 {
251 	struct kmem_cache *s;
252 
253 	if (slab_nomerge || (flags & SLAB_NEVER_MERGE))
254 		return NULL;
255 
256 	if (ctor)
257 		return NULL;
258 
259 	size = ALIGN(size, sizeof(void *));
260 	align = calculate_alignment(flags, align, size);
261 	size = ALIGN(size, align);
262 	flags = kmem_cache_flags(size, flags, name, NULL);
263 
264 	list_for_each_entry_reverse(s, &slab_caches, list) {
265 		if (slab_unmergeable(s))
266 			continue;
267 
268 		if (size > s->size)
269 			continue;
270 
271 		if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
272 			continue;
273 		/*
274 		 * Check if alignment is compatible.
275 		 * Courtesy of Adrian Drzewiecki
276 		 */
277 		if ((s->size & ~(align - 1)) != s->size)
278 			continue;
279 
280 		if (s->size - size >= sizeof(void *))
281 			continue;
282 
283 		if (IS_ENABLED(CONFIG_SLAB) && align &&
284 			(align > s->align || s->align % align))
285 			continue;
286 
287 		return s;
288 	}
289 	return NULL;
290 }
291 
292 /*
293  * Figure out what the alignment of the objects will be given a set of
294  * flags, a user specified alignment and the size of the objects.
295  */
296 unsigned long calculate_alignment(unsigned long flags,
297 		unsigned long align, unsigned long size)
298 {
299 	/*
300 	 * If the user wants hardware cache aligned objects then follow that
301 	 * suggestion if the object is sufficiently large.
302 	 *
303 	 * The hardware cache alignment cannot override the specified
304 	 * alignment though. If that is greater then use it.
305 	 */
306 	if (flags & SLAB_HWCACHE_ALIGN) {
307 		unsigned long ralign = cache_line_size();
308 		while (size <= ralign / 2)
309 			ralign /= 2;
310 		align = max(align, ralign);
311 	}
312 
313 	if (align < ARCH_SLAB_MINALIGN)
314 		align = ARCH_SLAB_MINALIGN;
315 
316 	return ALIGN(align, sizeof(void *));
317 }
318 
319 static struct kmem_cache *
320 do_kmem_cache_create(const char *name, size_t object_size, size_t size,
321 		     size_t align, unsigned long flags, void (*ctor)(void *),
322 		     struct mem_cgroup *memcg, struct kmem_cache *root_cache)
323 {
324 	struct kmem_cache *s;
325 	int err;
326 
327 	err = -ENOMEM;
328 	s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
329 	if (!s)
330 		goto out;
331 
332 	s->name = name;
333 	s->object_size = object_size;
334 	s->size = size;
335 	s->align = align;
336 	s->ctor = ctor;
337 
338 	err = init_memcg_params(s, memcg, root_cache);
339 	if (err)
340 		goto out_free_cache;
341 
342 	err = __kmem_cache_create(s, flags);
343 	if (err)
344 		goto out_free_cache;
345 
346 	s->refcount = 1;
347 	list_add(&s->list, &slab_caches);
348 out:
349 	if (err)
350 		return ERR_PTR(err);
351 	return s;
352 
353 out_free_cache:
354 	destroy_memcg_params(s);
355 	kmem_cache_free(kmem_cache, s);
356 	goto out;
357 }
358 
359 /*
360  * kmem_cache_create - Create a cache.
361  * @name: A string which is used in /proc/slabinfo to identify this cache.
362  * @size: The size of objects to be created in this cache.
363  * @align: The required alignment for the objects.
364  * @flags: SLAB flags
365  * @ctor: A constructor for the objects.
366  *
367  * Returns a ptr to the cache on success, NULL on failure.
368  * Cannot be called within a interrupt, but can be interrupted.
369  * The @ctor is run when new pages are allocated by the cache.
370  *
371  * The flags are
372  *
373  * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
374  * to catch references to uninitialised memory.
375  *
376  * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check
377  * for buffer overruns.
378  *
379  * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
380  * cacheline.  This can be beneficial if you're counting cycles as closely
381  * as davem.
382  */
383 struct kmem_cache *
384 kmem_cache_create(const char *name, size_t size, size_t align,
385 		  unsigned long flags, void (*ctor)(void *))
386 {
387 	struct kmem_cache *s;
388 	const char *cache_name;
389 	int err;
390 
391 	get_online_cpus();
392 	get_online_mems();
393 	memcg_get_cache_ids();
394 
395 	mutex_lock(&slab_mutex);
396 
397 	err = kmem_cache_sanity_check(name, size);
398 	if (err) {
399 		s = NULL;	/* suppress uninit var warning */
400 		goto out_unlock;
401 	}
402 
403 	/*
404 	 * Some allocators will constraint the set of valid flags to a subset
405 	 * of all flags. We expect them to define CACHE_CREATE_MASK in this
406 	 * case, and we'll just provide them with a sanitized version of the
407 	 * passed flags.
408 	 */
409 	flags &= CACHE_CREATE_MASK;
410 
411 	s = __kmem_cache_alias(name, size, align, flags, ctor);
412 	if (s)
413 		goto out_unlock;
414 
415 	cache_name = kstrdup_const(name, GFP_KERNEL);
416 	if (!cache_name) {
417 		err = -ENOMEM;
418 		goto out_unlock;
419 	}
420 
421 	s = do_kmem_cache_create(cache_name, size, size,
422 				 calculate_alignment(flags, align, size),
423 				 flags, ctor, NULL, NULL);
424 	if (IS_ERR(s)) {
425 		err = PTR_ERR(s);
426 		kfree_const(cache_name);
427 	}
428 
429 out_unlock:
430 	mutex_unlock(&slab_mutex);
431 
432 	memcg_put_cache_ids();
433 	put_online_mems();
434 	put_online_cpus();
435 
436 	if (err) {
437 		if (flags & SLAB_PANIC)
438 			panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n",
439 				name, err);
440 		else {
441 			printk(KERN_WARNING "kmem_cache_create(%s) failed with error %d",
442 				name, err);
443 			dump_stack();
444 		}
445 		return NULL;
446 	}
447 	return s;
448 }
449 EXPORT_SYMBOL(kmem_cache_create);
450 
451 static int do_kmem_cache_shutdown(struct kmem_cache *s,
452 		struct list_head *release, bool *need_rcu_barrier)
453 {
454 	if (__kmem_cache_shutdown(s) != 0) {
455 		printk(KERN_ERR "kmem_cache_destroy %s: "
456 		       "Slab cache still has objects\n", s->name);
457 		dump_stack();
458 		return -EBUSY;
459 	}
460 
461 	if (s->flags & SLAB_DESTROY_BY_RCU)
462 		*need_rcu_barrier = true;
463 
464 #ifdef CONFIG_MEMCG_KMEM
465 	if (!is_root_cache(s))
466 		list_del(&s->memcg_params.list);
467 #endif
468 	list_move(&s->list, release);
469 	return 0;
470 }
471 
472 static void do_kmem_cache_release(struct list_head *release,
473 				  bool need_rcu_barrier)
474 {
475 	struct kmem_cache *s, *s2;
476 
477 	if (need_rcu_barrier)
478 		rcu_barrier();
479 
480 	list_for_each_entry_safe(s, s2, release, list) {
481 #ifdef SLAB_SUPPORTS_SYSFS
482 		sysfs_slab_remove(s);
483 #else
484 		slab_kmem_cache_release(s);
485 #endif
486 	}
487 }
488 
489 #ifdef CONFIG_MEMCG_KMEM
490 /*
491  * memcg_create_kmem_cache - Create a cache for a memory cgroup.
492  * @memcg: The memory cgroup the new cache is for.
493  * @root_cache: The parent of the new cache.
494  *
495  * This function attempts to create a kmem cache that will serve allocation
496  * requests going from @memcg to @root_cache. The new cache inherits properties
497  * from its parent.
498  */
499 void memcg_create_kmem_cache(struct mem_cgroup *memcg,
500 			     struct kmem_cache *root_cache)
501 {
502 	static char memcg_name_buf[NAME_MAX + 1]; /* protected by slab_mutex */
503 	struct cgroup_subsys_state *css = &memcg->css;
504 	struct memcg_cache_array *arr;
505 	struct kmem_cache *s = NULL;
506 	char *cache_name;
507 	int idx;
508 
509 	get_online_cpus();
510 	get_online_mems();
511 
512 	mutex_lock(&slab_mutex);
513 
514 	/*
515 	 * The memory cgroup could have been deactivated while the cache
516 	 * creation work was pending.
517 	 */
518 	if (!memcg_kmem_is_active(memcg))
519 		goto out_unlock;
520 
521 	idx = memcg_cache_id(memcg);
522 	arr = rcu_dereference_protected(root_cache->memcg_params.memcg_caches,
523 					lockdep_is_held(&slab_mutex));
524 
525 	/*
526 	 * Since per-memcg caches are created asynchronously on first
527 	 * allocation (see memcg_kmem_get_cache()), several threads can try to
528 	 * create the same cache, but only one of them may succeed.
529 	 */
530 	if (arr->entries[idx])
531 		goto out_unlock;
532 
533 	cgroup_name(css->cgroup, memcg_name_buf, sizeof(memcg_name_buf));
534 	cache_name = kasprintf(GFP_KERNEL, "%s(%d:%s)", root_cache->name,
535 			       css->id, memcg_name_buf);
536 	if (!cache_name)
537 		goto out_unlock;
538 
539 	s = do_kmem_cache_create(cache_name, root_cache->object_size,
540 				 root_cache->size, root_cache->align,
541 				 root_cache->flags, root_cache->ctor,
542 				 memcg, root_cache);
543 	/*
544 	 * If we could not create a memcg cache, do not complain, because
545 	 * that's not critical at all as we can always proceed with the root
546 	 * cache.
547 	 */
548 	if (IS_ERR(s)) {
549 		kfree(cache_name);
550 		goto out_unlock;
551 	}
552 
553 	list_add(&s->memcg_params.list, &root_cache->memcg_params.list);
554 
555 	/*
556 	 * Since readers won't lock (see cache_from_memcg_idx()), we need a
557 	 * barrier here to ensure nobody will see the kmem_cache partially
558 	 * initialized.
559 	 */
560 	smp_wmb();
561 	arr->entries[idx] = s;
562 
563 out_unlock:
564 	mutex_unlock(&slab_mutex);
565 
566 	put_online_mems();
567 	put_online_cpus();
568 }
569 
570 void memcg_deactivate_kmem_caches(struct mem_cgroup *memcg)
571 {
572 	int idx;
573 	struct memcg_cache_array *arr;
574 	struct kmem_cache *s, *c;
575 
576 	idx = memcg_cache_id(memcg);
577 
578 	get_online_cpus();
579 	get_online_mems();
580 
581 	mutex_lock(&slab_mutex);
582 	list_for_each_entry(s, &slab_caches, list) {
583 		if (!is_root_cache(s))
584 			continue;
585 
586 		arr = rcu_dereference_protected(s->memcg_params.memcg_caches,
587 						lockdep_is_held(&slab_mutex));
588 		c = arr->entries[idx];
589 		if (!c)
590 			continue;
591 
592 		__kmem_cache_shrink(c, true);
593 		arr->entries[idx] = NULL;
594 	}
595 	mutex_unlock(&slab_mutex);
596 
597 	put_online_mems();
598 	put_online_cpus();
599 }
600 
601 void memcg_destroy_kmem_caches(struct mem_cgroup *memcg)
602 {
603 	LIST_HEAD(release);
604 	bool need_rcu_barrier = false;
605 	struct kmem_cache *s, *s2;
606 
607 	get_online_cpus();
608 	get_online_mems();
609 
610 	mutex_lock(&slab_mutex);
611 	list_for_each_entry_safe(s, s2, &slab_caches, list) {
612 		if (is_root_cache(s) || s->memcg_params.memcg != memcg)
613 			continue;
614 		/*
615 		 * The cgroup is about to be freed and therefore has no charges
616 		 * left. Hence, all its caches must be empty by now.
617 		 */
618 		BUG_ON(do_kmem_cache_shutdown(s, &release, &need_rcu_barrier));
619 	}
620 	mutex_unlock(&slab_mutex);
621 
622 	put_online_mems();
623 	put_online_cpus();
624 
625 	do_kmem_cache_release(&release, need_rcu_barrier);
626 }
627 #endif /* CONFIG_MEMCG_KMEM */
628 
629 void slab_kmem_cache_release(struct kmem_cache *s)
630 {
631 	destroy_memcg_params(s);
632 	kfree_const(s->name);
633 	kmem_cache_free(kmem_cache, s);
634 }
635 
636 void kmem_cache_destroy(struct kmem_cache *s)
637 {
638 	struct kmem_cache *c, *c2;
639 	LIST_HEAD(release);
640 	bool need_rcu_barrier = false;
641 	bool busy = false;
642 
643 	if (unlikely(!s))
644 		return;
645 
646 	BUG_ON(!is_root_cache(s));
647 
648 	get_online_cpus();
649 	get_online_mems();
650 
651 	mutex_lock(&slab_mutex);
652 
653 	s->refcount--;
654 	if (s->refcount)
655 		goto out_unlock;
656 
657 	for_each_memcg_cache_safe(c, c2, s) {
658 		if (do_kmem_cache_shutdown(c, &release, &need_rcu_barrier))
659 			busy = true;
660 	}
661 
662 	if (!busy)
663 		do_kmem_cache_shutdown(s, &release, &need_rcu_barrier);
664 
665 out_unlock:
666 	mutex_unlock(&slab_mutex);
667 
668 	put_online_mems();
669 	put_online_cpus();
670 
671 	do_kmem_cache_release(&release, need_rcu_barrier);
672 }
673 EXPORT_SYMBOL(kmem_cache_destroy);
674 
675 /**
676  * kmem_cache_shrink - Shrink a cache.
677  * @cachep: The cache to shrink.
678  *
679  * Releases as many slabs as possible for a cache.
680  * To help debugging, a zero exit status indicates all slabs were released.
681  */
682 int kmem_cache_shrink(struct kmem_cache *cachep)
683 {
684 	int ret;
685 
686 	get_online_cpus();
687 	get_online_mems();
688 	ret = __kmem_cache_shrink(cachep, false);
689 	put_online_mems();
690 	put_online_cpus();
691 	return ret;
692 }
693 EXPORT_SYMBOL(kmem_cache_shrink);
694 
695 int slab_is_available(void)
696 {
697 	return slab_state >= UP;
698 }
699 
700 #ifndef CONFIG_SLOB
701 /* Create a cache during boot when no slab services are available yet */
702 void __init create_boot_cache(struct kmem_cache *s, const char *name, size_t size,
703 		unsigned long flags)
704 {
705 	int err;
706 
707 	s->name = name;
708 	s->size = s->object_size = size;
709 	s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size);
710 
711 	slab_init_memcg_params(s);
712 
713 	err = __kmem_cache_create(s, flags);
714 
715 	if (err)
716 		panic("Creation of kmalloc slab %s size=%zu failed. Reason %d\n",
717 					name, size, err);
718 
719 	s->refcount = -1;	/* Exempt from merging for now */
720 }
721 
722 struct kmem_cache *__init create_kmalloc_cache(const char *name, size_t size,
723 				unsigned long flags)
724 {
725 	struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
726 
727 	if (!s)
728 		panic("Out of memory when creating slab %s\n", name);
729 
730 	create_boot_cache(s, name, size, flags);
731 	list_add(&s->list, &slab_caches);
732 	s->refcount = 1;
733 	return s;
734 }
735 
736 struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1];
737 EXPORT_SYMBOL(kmalloc_caches);
738 
739 #ifdef CONFIG_ZONE_DMA
740 struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1];
741 EXPORT_SYMBOL(kmalloc_dma_caches);
742 #endif
743 
744 /*
745  * Conversion table for small slabs sizes / 8 to the index in the
746  * kmalloc array. This is necessary for slabs < 192 since we have non power
747  * of two cache sizes there. The size of larger slabs can be determined using
748  * fls.
749  */
750 static s8 size_index[24] = {
751 	3,	/* 8 */
752 	4,	/* 16 */
753 	5,	/* 24 */
754 	5,	/* 32 */
755 	6,	/* 40 */
756 	6,	/* 48 */
757 	6,	/* 56 */
758 	6,	/* 64 */
759 	1,	/* 72 */
760 	1,	/* 80 */
761 	1,	/* 88 */
762 	1,	/* 96 */
763 	7,	/* 104 */
764 	7,	/* 112 */
765 	7,	/* 120 */
766 	7,	/* 128 */
767 	2,	/* 136 */
768 	2,	/* 144 */
769 	2,	/* 152 */
770 	2,	/* 160 */
771 	2,	/* 168 */
772 	2,	/* 176 */
773 	2,	/* 184 */
774 	2	/* 192 */
775 };
776 
777 static inline int size_index_elem(size_t bytes)
778 {
779 	return (bytes - 1) / 8;
780 }
781 
782 /*
783  * Find the kmem_cache structure that serves a given size of
784  * allocation
785  */
786 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
787 {
788 	int index;
789 
790 	if (unlikely(size > KMALLOC_MAX_SIZE)) {
791 		WARN_ON_ONCE(!(flags & __GFP_NOWARN));
792 		return NULL;
793 	}
794 
795 	if (size <= 192) {
796 		if (!size)
797 			return ZERO_SIZE_PTR;
798 
799 		index = size_index[size_index_elem(size)];
800 	} else
801 		index = fls(size - 1);
802 
803 #ifdef CONFIG_ZONE_DMA
804 	if (unlikely((flags & GFP_DMA)))
805 		return kmalloc_dma_caches[index];
806 
807 #endif
808 	return kmalloc_caches[index];
809 }
810 
811 /*
812  * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
813  * kmalloc_index() supports up to 2^26=64MB, so the final entry of the table is
814  * kmalloc-67108864.
815  */
816 static struct {
817 	const char *name;
818 	unsigned long size;
819 } const kmalloc_info[] __initconst = {
820 	{NULL,                      0},		{"kmalloc-96",             96},
821 	{"kmalloc-192",           192},		{"kmalloc-8",               8},
822 	{"kmalloc-16",             16},		{"kmalloc-32",             32},
823 	{"kmalloc-64",             64},		{"kmalloc-128",           128},
824 	{"kmalloc-256",           256},		{"kmalloc-512",           512},
825 	{"kmalloc-1024",         1024},		{"kmalloc-2048",         2048},
826 	{"kmalloc-4096",         4096},		{"kmalloc-8192",         8192},
827 	{"kmalloc-16384",       16384},		{"kmalloc-32768",       32768},
828 	{"kmalloc-65536",       65536},		{"kmalloc-131072",     131072},
829 	{"kmalloc-262144",     262144},		{"kmalloc-524288",     524288},
830 	{"kmalloc-1048576",   1048576},		{"kmalloc-2097152",   2097152},
831 	{"kmalloc-4194304",   4194304},		{"kmalloc-8388608",   8388608},
832 	{"kmalloc-16777216", 16777216},		{"kmalloc-33554432", 33554432},
833 	{"kmalloc-67108864", 67108864}
834 };
835 
836 /*
837  * Patch up the size_index table if we have strange large alignment
838  * requirements for the kmalloc array. This is only the case for
839  * MIPS it seems. The standard arches will not generate any code here.
840  *
841  * Largest permitted alignment is 256 bytes due to the way we
842  * handle the index determination for the smaller caches.
843  *
844  * Make sure that nothing crazy happens if someone starts tinkering
845  * around with ARCH_KMALLOC_MINALIGN
846  */
847 void __init setup_kmalloc_cache_index_table(void)
848 {
849 	int i;
850 
851 	BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
852 		(KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
853 
854 	for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
855 		int elem = size_index_elem(i);
856 
857 		if (elem >= ARRAY_SIZE(size_index))
858 			break;
859 		size_index[elem] = KMALLOC_SHIFT_LOW;
860 	}
861 
862 	if (KMALLOC_MIN_SIZE >= 64) {
863 		/*
864 		 * The 96 byte size cache is not used if the alignment
865 		 * is 64 byte.
866 		 */
867 		for (i = 64 + 8; i <= 96; i += 8)
868 			size_index[size_index_elem(i)] = 7;
869 
870 	}
871 
872 	if (KMALLOC_MIN_SIZE >= 128) {
873 		/*
874 		 * The 192 byte sized cache is not used if the alignment
875 		 * is 128 byte. Redirect kmalloc to use the 256 byte cache
876 		 * instead.
877 		 */
878 		for (i = 128 + 8; i <= 192; i += 8)
879 			size_index[size_index_elem(i)] = 8;
880 	}
881 }
882 
883 static void __init new_kmalloc_cache(int idx, unsigned long flags)
884 {
885 	kmalloc_caches[idx] = create_kmalloc_cache(kmalloc_info[idx].name,
886 					kmalloc_info[idx].size, flags);
887 }
888 
889 /*
890  * Create the kmalloc array. Some of the regular kmalloc arrays
891  * may already have been created because they were needed to
892  * enable allocations for slab creation.
893  */
894 void __init create_kmalloc_caches(unsigned long flags)
895 {
896 	int i;
897 
898 	for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
899 		if (!kmalloc_caches[i])
900 			new_kmalloc_cache(i, flags);
901 
902 		/*
903 		 * Caches that are not of the two-to-the-power-of size.
904 		 * These have to be created immediately after the
905 		 * earlier power of two caches
906 		 */
907 		if (KMALLOC_MIN_SIZE <= 32 && !kmalloc_caches[1] && i == 6)
908 			new_kmalloc_cache(1, flags);
909 		if (KMALLOC_MIN_SIZE <= 64 && !kmalloc_caches[2] && i == 7)
910 			new_kmalloc_cache(2, flags);
911 	}
912 
913 	/* Kmalloc array is now usable */
914 	slab_state = UP;
915 
916 #ifdef CONFIG_ZONE_DMA
917 	for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
918 		struct kmem_cache *s = kmalloc_caches[i];
919 
920 		if (s) {
921 			int size = kmalloc_size(i);
922 			char *n = kasprintf(GFP_NOWAIT,
923 				 "dma-kmalloc-%d", size);
924 
925 			BUG_ON(!n);
926 			kmalloc_dma_caches[i] = create_kmalloc_cache(n,
927 				size, SLAB_CACHE_DMA | flags);
928 		}
929 	}
930 #endif
931 }
932 #endif /* !CONFIG_SLOB */
933 
934 /*
935  * To avoid unnecessary overhead, we pass through large allocation requests
936  * directly to the page allocator. We use __GFP_COMP, because we will need to
937  * know the allocation order to free the pages properly in kfree.
938  */
939 void *kmalloc_order(size_t size, gfp_t flags, unsigned int order)
940 {
941 	void *ret;
942 	struct page *page;
943 
944 	flags |= __GFP_COMP;
945 	page = alloc_kmem_pages(flags, order);
946 	ret = page ? page_address(page) : NULL;
947 	kmemleak_alloc(ret, size, 1, flags);
948 	kasan_kmalloc_large(ret, size);
949 	return ret;
950 }
951 EXPORT_SYMBOL(kmalloc_order);
952 
953 #ifdef CONFIG_TRACING
954 void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
955 {
956 	void *ret = kmalloc_order(size, flags, order);
957 	trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
958 	return ret;
959 }
960 EXPORT_SYMBOL(kmalloc_order_trace);
961 #endif
962 
963 #ifdef CONFIG_SLABINFO
964 
965 #ifdef CONFIG_SLAB
966 #define SLABINFO_RIGHTS (S_IWUSR | S_IRUSR)
967 #else
968 #define SLABINFO_RIGHTS S_IRUSR
969 #endif
970 
971 static void print_slabinfo_header(struct seq_file *m)
972 {
973 	/*
974 	 * Output format version, so at least we can change it
975 	 * without _too_ many complaints.
976 	 */
977 #ifdef CONFIG_DEBUG_SLAB
978 	seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
979 #else
980 	seq_puts(m, "slabinfo - version: 2.1\n");
981 #endif
982 	seq_puts(m, "# name            <active_objs> <num_objs> <objsize> "
983 		 "<objperslab> <pagesperslab>");
984 	seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
985 	seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
986 #ifdef CONFIG_DEBUG_SLAB
987 	seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> "
988 		 "<error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
989 	seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
990 #endif
991 	seq_putc(m, '\n');
992 }
993 
994 void *slab_start(struct seq_file *m, loff_t *pos)
995 {
996 	mutex_lock(&slab_mutex);
997 	return seq_list_start(&slab_caches, *pos);
998 }
999 
1000 void *slab_next(struct seq_file *m, void *p, loff_t *pos)
1001 {
1002 	return seq_list_next(p, &slab_caches, pos);
1003 }
1004 
1005 void slab_stop(struct seq_file *m, void *p)
1006 {
1007 	mutex_unlock(&slab_mutex);
1008 }
1009 
1010 static void
1011 memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info)
1012 {
1013 	struct kmem_cache *c;
1014 	struct slabinfo sinfo;
1015 
1016 	if (!is_root_cache(s))
1017 		return;
1018 
1019 	for_each_memcg_cache(c, s) {
1020 		memset(&sinfo, 0, sizeof(sinfo));
1021 		get_slabinfo(c, &sinfo);
1022 
1023 		info->active_slabs += sinfo.active_slabs;
1024 		info->num_slabs += sinfo.num_slabs;
1025 		info->shared_avail += sinfo.shared_avail;
1026 		info->active_objs += sinfo.active_objs;
1027 		info->num_objs += sinfo.num_objs;
1028 	}
1029 }
1030 
1031 static void cache_show(struct kmem_cache *s, struct seq_file *m)
1032 {
1033 	struct slabinfo sinfo;
1034 
1035 	memset(&sinfo, 0, sizeof(sinfo));
1036 	get_slabinfo(s, &sinfo);
1037 
1038 	memcg_accumulate_slabinfo(s, &sinfo);
1039 
1040 	seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
1041 		   cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size,
1042 		   sinfo.objects_per_slab, (1 << sinfo.cache_order));
1043 
1044 	seq_printf(m, " : tunables %4u %4u %4u",
1045 		   sinfo.limit, sinfo.batchcount, sinfo.shared);
1046 	seq_printf(m, " : slabdata %6lu %6lu %6lu",
1047 		   sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
1048 	slabinfo_show_stats(m, s);
1049 	seq_putc(m, '\n');
1050 }
1051 
1052 static int slab_show(struct seq_file *m, void *p)
1053 {
1054 	struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
1055 
1056 	if (p == slab_caches.next)
1057 		print_slabinfo_header(m);
1058 	if (is_root_cache(s))
1059 		cache_show(s, m);
1060 	return 0;
1061 }
1062 
1063 #ifdef CONFIG_MEMCG_KMEM
1064 int memcg_slab_show(struct seq_file *m, void *p)
1065 {
1066 	struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
1067 	struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
1068 
1069 	if (p == slab_caches.next)
1070 		print_slabinfo_header(m);
1071 	if (!is_root_cache(s) && s->memcg_params.memcg == memcg)
1072 		cache_show(s, m);
1073 	return 0;
1074 }
1075 #endif
1076 
1077 /*
1078  * slabinfo_op - iterator that generates /proc/slabinfo
1079  *
1080  * Output layout:
1081  * cache-name
1082  * num-active-objs
1083  * total-objs
1084  * object size
1085  * num-active-slabs
1086  * total-slabs
1087  * num-pages-per-slab
1088  * + further values on SMP and with statistics enabled
1089  */
1090 static const struct seq_operations slabinfo_op = {
1091 	.start = slab_start,
1092 	.next = slab_next,
1093 	.stop = slab_stop,
1094 	.show = slab_show,
1095 };
1096 
1097 static int slabinfo_open(struct inode *inode, struct file *file)
1098 {
1099 	return seq_open(file, &slabinfo_op);
1100 }
1101 
1102 static const struct file_operations proc_slabinfo_operations = {
1103 	.open		= slabinfo_open,
1104 	.read		= seq_read,
1105 	.write          = slabinfo_write,
1106 	.llseek		= seq_lseek,
1107 	.release	= seq_release,
1108 };
1109 
1110 static int __init slab_proc_init(void)
1111 {
1112 	proc_create("slabinfo", SLABINFO_RIGHTS, NULL,
1113 						&proc_slabinfo_operations);
1114 	return 0;
1115 }
1116 module_init(slab_proc_init);
1117 #endif /* CONFIG_SLABINFO */
1118 
1119 static __always_inline void *__do_krealloc(const void *p, size_t new_size,
1120 					   gfp_t flags)
1121 {
1122 	void *ret;
1123 	size_t ks = 0;
1124 
1125 	if (p)
1126 		ks = ksize(p);
1127 
1128 	if (ks >= new_size) {
1129 		kasan_krealloc((void *)p, new_size);
1130 		return (void *)p;
1131 	}
1132 
1133 	ret = kmalloc_track_caller(new_size, flags);
1134 	if (ret && p)
1135 		memcpy(ret, p, ks);
1136 
1137 	return ret;
1138 }
1139 
1140 /**
1141  * __krealloc - like krealloc() but don't free @p.
1142  * @p: object to reallocate memory for.
1143  * @new_size: how many bytes of memory are required.
1144  * @flags: the type of memory to allocate.
1145  *
1146  * This function is like krealloc() except it never frees the originally
1147  * allocated buffer. Use this if you don't want to free the buffer immediately
1148  * like, for example, with RCU.
1149  */
1150 void *__krealloc(const void *p, size_t new_size, gfp_t flags)
1151 {
1152 	if (unlikely(!new_size))
1153 		return ZERO_SIZE_PTR;
1154 
1155 	return __do_krealloc(p, new_size, flags);
1156 
1157 }
1158 EXPORT_SYMBOL(__krealloc);
1159 
1160 /**
1161  * krealloc - reallocate memory. The contents will remain unchanged.
1162  * @p: object to reallocate memory for.
1163  * @new_size: how many bytes of memory are required.
1164  * @flags: the type of memory to allocate.
1165  *
1166  * The contents of the object pointed to are preserved up to the
1167  * lesser of the new and old sizes.  If @p is %NULL, krealloc()
1168  * behaves exactly like kmalloc().  If @new_size is 0 and @p is not a
1169  * %NULL pointer, the object pointed to is freed.
1170  */
1171 void *krealloc(const void *p, size_t new_size, gfp_t flags)
1172 {
1173 	void *ret;
1174 
1175 	if (unlikely(!new_size)) {
1176 		kfree(p);
1177 		return ZERO_SIZE_PTR;
1178 	}
1179 
1180 	ret = __do_krealloc(p, new_size, flags);
1181 	if (ret && p != ret)
1182 		kfree(p);
1183 
1184 	return ret;
1185 }
1186 EXPORT_SYMBOL(krealloc);
1187 
1188 /**
1189  * kzfree - like kfree but zero memory
1190  * @p: object to free memory of
1191  *
1192  * The memory of the object @p points to is zeroed before freed.
1193  * If @p is %NULL, kzfree() does nothing.
1194  *
1195  * Note: this function zeroes the whole allocated buffer which can be a good
1196  * deal bigger than the requested buffer size passed to kmalloc(). So be
1197  * careful when using this function in performance sensitive code.
1198  */
1199 void kzfree(const void *p)
1200 {
1201 	size_t ks;
1202 	void *mem = (void *)p;
1203 
1204 	if (unlikely(ZERO_OR_NULL_PTR(mem)))
1205 		return;
1206 	ks = ksize(mem);
1207 	memset(mem, 0, ks);
1208 	kfree(mem);
1209 }
1210 EXPORT_SYMBOL(kzfree);
1211 
1212 /* Tracepoints definitions. */
1213 EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1214 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1215 EXPORT_TRACEPOINT_SYMBOL(kmalloc_node);
1216 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node);
1217 EXPORT_TRACEPOINT_SYMBOL(kfree);
1218 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);
1219