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