Loading...
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 <linux/debugfs.h>
21#include <asm/cacheflush.h>
22#include <asm/tlbflush.h>
23#include <asm/page.h>
24#include <linux/memcontrol.h>
25
26#define CREATE_TRACE_POINTS
27#include <trace/events/kmem.h>
28
29#include "slab.h"
30
31enum slab_state slab_state;
32LIST_HEAD(slab_caches);
33DEFINE_MUTEX(slab_mutex);
34struct kmem_cache *kmem_cache;
35
36#ifdef CONFIG_HARDENED_USERCOPY
37bool usercopy_fallback __ro_after_init =
38 IS_ENABLED(CONFIG_HARDENED_USERCOPY_FALLBACK);
39module_param(usercopy_fallback, bool, 0400);
40MODULE_PARM_DESC(usercopy_fallback,
41 "WARN instead of reject usercopy whitelist violations");
42#endif
43
44static LIST_HEAD(slab_caches_to_rcu_destroy);
45static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work);
46static DECLARE_WORK(slab_caches_to_rcu_destroy_work,
47 slab_caches_to_rcu_destroy_workfn);
48
49/*
50 * Set of flags that will prevent slab merging
51 */
52#define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
53 SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \
54 SLAB_FAILSLAB | SLAB_KASAN)
55
56#define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \
57 SLAB_CACHE_DMA32 | SLAB_ACCOUNT)
58
59/*
60 * Merge control. If this is set then no merging of slab caches will occur.
61 */
62static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT);
63
64static int __init setup_slab_nomerge(char *str)
65{
66 slab_nomerge = true;
67 return 1;
68}
69
70#ifdef CONFIG_SLUB
71__setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0);
72#endif
73
74__setup("slab_nomerge", setup_slab_nomerge);
75
76/*
77 * Determine the size of a slab object
78 */
79unsigned int kmem_cache_size(struct kmem_cache *s)
80{
81 return s->object_size;
82}
83EXPORT_SYMBOL(kmem_cache_size);
84
85#ifdef CONFIG_DEBUG_VM
86static int kmem_cache_sanity_check(const char *name, unsigned int size)
87{
88 if (!name || in_interrupt() || size < sizeof(void *) ||
89 size > KMALLOC_MAX_SIZE) {
90 pr_err("kmem_cache_create(%s) integrity check failed\n", name);
91 return -EINVAL;
92 }
93
94 WARN_ON(strchr(name, ' ')); /* It confuses parsers */
95 return 0;
96}
97#else
98static inline int kmem_cache_sanity_check(const char *name, unsigned int size)
99{
100 return 0;
101}
102#endif
103
104void __kmem_cache_free_bulk(struct kmem_cache *s, size_t nr, void **p)
105{
106 size_t i;
107
108 for (i = 0; i < nr; i++) {
109 if (s)
110 kmem_cache_free(s, p[i]);
111 else
112 kfree(p[i]);
113 }
114}
115
116int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t nr,
117 void **p)
118{
119 size_t i;
120
121 for (i = 0; i < nr; i++) {
122 void *x = p[i] = kmem_cache_alloc(s, flags);
123 if (!x) {
124 __kmem_cache_free_bulk(s, i, p);
125 return 0;
126 }
127 }
128 return i;
129}
130
131#ifdef CONFIG_MEMCG_KMEM
132
133LIST_HEAD(slab_root_caches);
134static DEFINE_SPINLOCK(memcg_kmem_wq_lock);
135
136static void kmemcg_cache_shutdown(struct percpu_ref *percpu_ref);
137
138void slab_init_memcg_params(struct kmem_cache *s)
139{
140 s->memcg_params.root_cache = NULL;
141 RCU_INIT_POINTER(s->memcg_params.memcg_caches, NULL);
142 INIT_LIST_HEAD(&s->memcg_params.children);
143 s->memcg_params.dying = false;
144}
145
146static int init_memcg_params(struct kmem_cache *s,
147 struct kmem_cache *root_cache)
148{
149 struct memcg_cache_array *arr;
150
151 if (root_cache) {
152 int ret = percpu_ref_init(&s->memcg_params.refcnt,
153 kmemcg_cache_shutdown,
154 0, GFP_KERNEL);
155 if (ret)
156 return ret;
157
158 s->memcg_params.root_cache = root_cache;
159 INIT_LIST_HEAD(&s->memcg_params.children_node);
160 INIT_LIST_HEAD(&s->memcg_params.kmem_caches_node);
161 return 0;
162 }
163
164 slab_init_memcg_params(s);
165
166 if (!memcg_nr_cache_ids)
167 return 0;
168
169 arr = kvzalloc(sizeof(struct memcg_cache_array) +
170 memcg_nr_cache_ids * sizeof(void *),
171 GFP_KERNEL);
172 if (!arr)
173 return -ENOMEM;
174
175 RCU_INIT_POINTER(s->memcg_params.memcg_caches, arr);
176 return 0;
177}
178
179static void destroy_memcg_params(struct kmem_cache *s)
180{
181 if (is_root_cache(s)) {
182 kvfree(rcu_access_pointer(s->memcg_params.memcg_caches));
183 } else {
184 mem_cgroup_put(s->memcg_params.memcg);
185 WRITE_ONCE(s->memcg_params.memcg, NULL);
186 percpu_ref_exit(&s->memcg_params.refcnt);
187 }
188}
189
190static void free_memcg_params(struct rcu_head *rcu)
191{
192 struct memcg_cache_array *old;
193
194 old = container_of(rcu, struct memcg_cache_array, rcu);
195 kvfree(old);
196}
197
198static int update_memcg_params(struct kmem_cache *s, int new_array_size)
199{
200 struct memcg_cache_array *old, *new;
201
202 new = kvzalloc(sizeof(struct memcg_cache_array) +
203 new_array_size * sizeof(void *), GFP_KERNEL);
204 if (!new)
205 return -ENOMEM;
206
207 old = rcu_dereference_protected(s->memcg_params.memcg_caches,
208 lockdep_is_held(&slab_mutex));
209 if (old)
210 memcpy(new->entries, old->entries,
211 memcg_nr_cache_ids * sizeof(void *));
212
213 rcu_assign_pointer(s->memcg_params.memcg_caches, new);
214 if (old)
215 call_rcu(&old->rcu, free_memcg_params);
216 return 0;
217}
218
219int memcg_update_all_caches(int num_memcgs)
220{
221 struct kmem_cache *s;
222 int ret = 0;
223
224 mutex_lock(&slab_mutex);
225 list_for_each_entry(s, &slab_root_caches, root_caches_node) {
226 ret = update_memcg_params(s, num_memcgs);
227 /*
228 * Instead of freeing the memory, we'll just leave the caches
229 * up to this point in an updated state.
230 */
231 if (ret)
232 break;
233 }
234 mutex_unlock(&slab_mutex);
235 return ret;
236}
237
238void memcg_link_cache(struct kmem_cache *s, struct mem_cgroup *memcg)
239{
240 if (is_root_cache(s)) {
241 list_add(&s->root_caches_node, &slab_root_caches);
242 } else {
243 css_get(&memcg->css);
244 s->memcg_params.memcg = memcg;
245 list_add(&s->memcg_params.children_node,
246 &s->memcg_params.root_cache->memcg_params.children);
247 list_add(&s->memcg_params.kmem_caches_node,
248 &s->memcg_params.memcg->kmem_caches);
249 }
250}
251
252static void memcg_unlink_cache(struct kmem_cache *s)
253{
254 if (is_root_cache(s)) {
255 list_del(&s->root_caches_node);
256 } else {
257 list_del(&s->memcg_params.children_node);
258 list_del(&s->memcg_params.kmem_caches_node);
259 }
260}
261#else
262static inline int init_memcg_params(struct kmem_cache *s,
263 struct kmem_cache *root_cache)
264{
265 return 0;
266}
267
268static inline void destroy_memcg_params(struct kmem_cache *s)
269{
270}
271
272static inline void memcg_unlink_cache(struct kmem_cache *s)
273{
274}
275#endif /* CONFIG_MEMCG_KMEM */
276
277/*
278 * Figure out what the alignment of the objects will be given a set of
279 * flags, a user specified alignment and the size of the objects.
280 */
281static unsigned int calculate_alignment(slab_flags_t flags,
282 unsigned int align, unsigned int size)
283{
284 /*
285 * If the user wants hardware cache aligned objects then follow that
286 * suggestion if the object is sufficiently large.
287 *
288 * The hardware cache alignment cannot override the specified
289 * alignment though. If that is greater then use it.
290 */
291 if (flags & SLAB_HWCACHE_ALIGN) {
292 unsigned int ralign;
293
294 ralign = cache_line_size();
295 while (size <= ralign / 2)
296 ralign /= 2;
297 align = max(align, ralign);
298 }
299
300 if (align < ARCH_SLAB_MINALIGN)
301 align = ARCH_SLAB_MINALIGN;
302
303 return ALIGN(align, sizeof(void *));
304}
305
306/*
307 * Find a mergeable slab cache
308 */
309int slab_unmergeable(struct kmem_cache *s)
310{
311 if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
312 return 1;
313
314 if (!is_root_cache(s))
315 return 1;
316
317 if (s->ctor)
318 return 1;
319
320 if (s->usersize)
321 return 1;
322
323 /*
324 * We may have set a slab to be unmergeable during bootstrap.
325 */
326 if (s->refcount < 0)
327 return 1;
328
329 return 0;
330}
331
332struct kmem_cache *find_mergeable(unsigned int size, unsigned int align,
333 slab_flags_t flags, const char *name, void (*ctor)(void *))
334{
335 struct kmem_cache *s;
336
337 if (slab_nomerge)
338 return NULL;
339
340 if (ctor)
341 return NULL;
342
343 size = ALIGN(size, sizeof(void *));
344 align = calculate_alignment(flags, align, size);
345 size = ALIGN(size, align);
346 flags = kmem_cache_flags(size, flags, name, NULL);
347
348 if (flags & SLAB_NEVER_MERGE)
349 return NULL;
350
351 list_for_each_entry_reverse(s, &slab_root_caches, root_caches_node) {
352 if (slab_unmergeable(s))
353 continue;
354
355 if (size > s->size)
356 continue;
357
358 if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
359 continue;
360 /*
361 * Check if alignment is compatible.
362 * Courtesy of Adrian Drzewiecki
363 */
364 if ((s->size & ~(align - 1)) != s->size)
365 continue;
366
367 if (s->size - size >= sizeof(void *))
368 continue;
369
370 if (IS_ENABLED(CONFIG_SLAB) && align &&
371 (align > s->align || s->align % align))
372 continue;
373
374 return s;
375 }
376 return NULL;
377}
378
379static struct kmem_cache *create_cache(const char *name,
380 unsigned int object_size, unsigned int align,
381 slab_flags_t flags, unsigned int useroffset,
382 unsigned int usersize, void (*ctor)(void *),
383 struct mem_cgroup *memcg, struct kmem_cache *root_cache)
384{
385 struct kmem_cache *s;
386 int err;
387
388 if (WARN_ON(useroffset + usersize > object_size))
389 useroffset = usersize = 0;
390
391 err = -ENOMEM;
392 s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
393 if (!s)
394 goto out;
395
396 s->name = name;
397 s->size = s->object_size = object_size;
398 s->align = align;
399 s->ctor = ctor;
400 s->useroffset = useroffset;
401 s->usersize = usersize;
402
403 err = init_memcg_params(s, root_cache);
404 if (err)
405 goto out_free_cache;
406
407 err = __kmem_cache_create(s, flags);
408 if (err)
409 goto out_free_cache;
410
411 s->refcount = 1;
412 list_add(&s->list, &slab_caches);
413 memcg_link_cache(s, memcg);
414out:
415 if (err)
416 return ERR_PTR(err);
417 return s;
418
419out_free_cache:
420 destroy_memcg_params(s);
421 kmem_cache_free(kmem_cache, s);
422 goto out;
423}
424
425/**
426 * kmem_cache_create_usercopy - Create a cache with a region suitable
427 * for copying to userspace
428 * @name: A string which is used in /proc/slabinfo to identify this cache.
429 * @size: The size of objects to be created in this cache.
430 * @align: The required alignment for the objects.
431 * @flags: SLAB flags
432 * @useroffset: Usercopy region offset
433 * @usersize: Usercopy region size
434 * @ctor: A constructor for the objects.
435 *
436 * Cannot be called within a interrupt, but can be interrupted.
437 * The @ctor is run when new pages are allocated by the cache.
438 *
439 * The flags are
440 *
441 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
442 * to catch references to uninitialised memory.
443 *
444 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
445 * for buffer overruns.
446 *
447 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
448 * cacheline. This can be beneficial if you're counting cycles as closely
449 * as davem.
450 *
451 * Return: a pointer to the cache on success, NULL on failure.
452 */
453struct kmem_cache *
454kmem_cache_create_usercopy(const char *name,
455 unsigned int size, unsigned int align,
456 slab_flags_t flags,
457 unsigned int useroffset, unsigned int usersize,
458 void (*ctor)(void *))
459{
460 struct kmem_cache *s = NULL;
461 const char *cache_name;
462 int err;
463
464 get_online_cpus();
465 get_online_mems();
466 memcg_get_cache_ids();
467
468 mutex_lock(&slab_mutex);
469
470 err = kmem_cache_sanity_check(name, size);
471 if (err) {
472 goto out_unlock;
473 }
474
475 /* Refuse requests with allocator specific flags */
476 if (flags & ~SLAB_FLAGS_PERMITTED) {
477 err = -EINVAL;
478 goto out_unlock;
479 }
480
481 /*
482 * Some allocators will constraint the set of valid flags to a subset
483 * of all flags. We expect them to define CACHE_CREATE_MASK in this
484 * case, and we'll just provide them with a sanitized version of the
485 * passed flags.
486 */
487 flags &= CACHE_CREATE_MASK;
488
489 /* Fail closed on bad usersize of useroffset values. */
490 if (WARN_ON(!usersize && useroffset) ||
491 WARN_ON(size < usersize || size - usersize < useroffset))
492 usersize = useroffset = 0;
493
494 if (!usersize)
495 s = __kmem_cache_alias(name, size, align, flags, ctor);
496 if (s)
497 goto out_unlock;
498
499 cache_name = kstrdup_const(name, GFP_KERNEL);
500 if (!cache_name) {
501 err = -ENOMEM;
502 goto out_unlock;
503 }
504
505 s = create_cache(cache_name, size,
506 calculate_alignment(flags, align, size),
507 flags, useroffset, usersize, ctor, NULL, NULL);
508 if (IS_ERR(s)) {
509 err = PTR_ERR(s);
510 kfree_const(cache_name);
511 }
512
513out_unlock:
514 mutex_unlock(&slab_mutex);
515
516 memcg_put_cache_ids();
517 put_online_mems();
518 put_online_cpus();
519
520 if (err) {
521 if (flags & SLAB_PANIC)
522 panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n",
523 name, err);
524 else {
525 pr_warn("kmem_cache_create(%s) failed with error %d\n",
526 name, err);
527 dump_stack();
528 }
529 return NULL;
530 }
531 return s;
532}
533EXPORT_SYMBOL(kmem_cache_create_usercopy);
534
535/**
536 * kmem_cache_create - Create a cache.
537 * @name: A string which is used in /proc/slabinfo to identify this cache.
538 * @size: The size of objects to be created in this cache.
539 * @align: The required alignment for the objects.
540 * @flags: SLAB flags
541 * @ctor: A constructor for the objects.
542 *
543 * Cannot be called within a interrupt, but can be interrupted.
544 * The @ctor is run when new pages are allocated by the cache.
545 *
546 * The flags are
547 *
548 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
549 * to catch references to uninitialised memory.
550 *
551 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
552 * for buffer overruns.
553 *
554 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
555 * cacheline. This can be beneficial if you're counting cycles as closely
556 * as davem.
557 *
558 * Return: a pointer to the cache on success, NULL on failure.
559 */
560struct kmem_cache *
561kmem_cache_create(const char *name, unsigned int size, unsigned int align,
562 slab_flags_t flags, void (*ctor)(void *))
563{
564 return kmem_cache_create_usercopy(name, size, align, flags, 0, 0,
565 ctor);
566}
567EXPORT_SYMBOL(kmem_cache_create);
568
569static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work)
570{
571 LIST_HEAD(to_destroy);
572 struct kmem_cache *s, *s2;
573
574 /*
575 * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the
576 * @slab_caches_to_rcu_destroy list. The slab pages are freed
577 * through RCU and and the associated kmem_cache are dereferenced
578 * while freeing the pages, so the kmem_caches should be freed only
579 * after the pending RCU operations are finished. As rcu_barrier()
580 * is a pretty slow operation, we batch all pending destructions
581 * asynchronously.
582 */
583 mutex_lock(&slab_mutex);
584 list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy);
585 mutex_unlock(&slab_mutex);
586
587 if (list_empty(&to_destroy))
588 return;
589
590 rcu_barrier();
591
592 list_for_each_entry_safe(s, s2, &to_destroy, list) {
593#ifdef SLAB_SUPPORTS_SYSFS
594 sysfs_slab_release(s);
595#else
596 slab_kmem_cache_release(s);
597#endif
598 }
599}
600
601static int shutdown_cache(struct kmem_cache *s)
602{
603 /* free asan quarantined objects */
604 kasan_cache_shutdown(s);
605
606 if (__kmem_cache_shutdown(s) != 0)
607 return -EBUSY;
608
609 memcg_unlink_cache(s);
610 list_del(&s->list);
611
612 if (s->flags & SLAB_TYPESAFE_BY_RCU) {
613#ifdef SLAB_SUPPORTS_SYSFS
614 sysfs_slab_unlink(s);
615#endif
616 list_add_tail(&s->list, &slab_caches_to_rcu_destroy);
617 schedule_work(&slab_caches_to_rcu_destroy_work);
618 } else {
619#ifdef SLAB_SUPPORTS_SYSFS
620 sysfs_slab_unlink(s);
621 sysfs_slab_release(s);
622#else
623 slab_kmem_cache_release(s);
624#endif
625 }
626
627 return 0;
628}
629
630#ifdef CONFIG_MEMCG_KMEM
631/*
632 * memcg_create_kmem_cache - Create a cache for a memory cgroup.
633 * @memcg: The memory cgroup the new cache is for.
634 * @root_cache: The parent of the new cache.
635 *
636 * This function attempts to create a kmem cache that will serve allocation
637 * requests going from @memcg to @root_cache. The new cache inherits properties
638 * from its parent.
639 */
640void memcg_create_kmem_cache(struct mem_cgroup *memcg,
641 struct kmem_cache *root_cache)
642{
643 static char memcg_name_buf[NAME_MAX + 1]; /* protected by slab_mutex */
644 struct cgroup_subsys_state *css = &memcg->css;
645 struct memcg_cache_array *arr;
646 struct kmem_cache *s = NULL;
647 char *cache_name;
648 int idx;
649
650 get_online_cpus();
651 get_online_mems();
652
653 mutex_lock(&slab_mutex);
654
655 /*
656 * The memory cgroup could have been offlined while the cache
657 * creation work was pending.
658 */
659 if (memcg->kmem_state != KMEM_ONLINE)
660 goto out_unlock;
661
662 idx = memcg_cache_id(memcg);
663 arr = rcu_dereference_protected(root_cache->memcg_params.memcg_caches,
664 lockdep_is_held(&slab_mutex));
665
666 /*
667 * Since per-memcg caches are created asynchronously on first
668 * allocation (see memcg_kmem_get_cache()), several threads can try to
669 * create the same cache, but only one of them may succeed.
670 */
671 if (arr->entries[idx])
672 goto out_unlock;
673
674 cgroup_name(css->cgroup, memcg_name_buf, sizeof(memcg_name_buf));
675 cache_name = kasprintf(GFP_KERNEL, "%s(%llu:%s)", root_cache->name,
676 css->serial_nr, memcg_name_buf);
677 if (!cache_name)
678 goto out_unlock;
679
680 s = create_cache(cache_name, root_cache->object_size,
681 root_cache->align,
682 root_cache->flags & CACHE_CREATE_MASK,
683 root_cache->useroffset, root_cache->usersize,
684 root_cache->ctor, memcg, root_cache);
685 /*
686 * If we could not create a memcg cache, do not complain, because
687 * that's not critical at all as we can always proceed with the root
688 * cache.
689 */
690 if (IS_ERR(s)) {
691 kfree(cache_name);
692 goto out_unlock;
693 }
694
695 /*
696 * Since readers won't lock (see memcg_kmem_get_cache()), we need a
697 * barrier here to ensure nobody will see the kmem_cache partially
698 * initialized.
699 */
700 smp_wmb();
701 arr->entries[idx] = s;
702
703out_unlock:
704 mutex_unlock(&slab_mutex);
705
706 put_online_mems();
707 put_online_cpus();
708}
709
710static void kmemcg_workfn(struct work_struct *work)
711{
712 struct kmem_cache *s = container_of(work, struct kmem_cache,
713 memcg_params.work);
714
715 get_online_cpus();
716 get_online_mems();
717
718 mutex_lock(&slab_mutex);
719 s->memcg_params.work_fn(s);
720 mutex_unlock(&slab_mutex);
721
722 put_online_mems();
723 put_online_cpus();
724}
725
726static void kmemcg_rcufn(struct rcu_head *head)
727{
728 struct kmem_cache *s = container_of(head, struct kmem_cache,
729 memcg_params.rcu_head);
730
731 /*
732 * We need to grab blocking locks. Bounce to ->work. The
733 * work item shares the space with the RCU head and can't be
734 * initialized eariler.
735 */
736 INIT_WORK(&s->memcg_params.work, kmemcg_workfn);
737 queue_work(memcg_kmem_cache_wq, &s->memcg_params.work);
738}
739
740static void kmemcg_cache_shutdown_fn(struct kmem_cache *s)
741{
742 WARN_ON(shutdown_cache(s));
743}
744
745static void kmemcg_cache_shutdown(struct percpu_ref *percpu_ref)
746{
747 struct kmem_cache *s = container_of(percpu_ref, struct kmem_cache,
748 memcg_params.refcnt);
749 unsigned long flags;
750
751 spin_lock_irqsave(&memcg_kmem_wq_lock, flags);
752 if (s->memcg_params.root_cache->memcg_params.dying)
753 goto unlock;
754
755 s->memcg_params.work_fn = kmemcg_cache_shutdown_fn;
756 INIT_WORK(&s->memcg_params.work, kmemcg_workfn);
757 queue_work(memcg_kmem_cache_wq, &s->memcg_params.work);
758
759unlock:
760 spin_unlock_irqrestore(&memcg_kmem_wq_lock, flags);
761}
762
763static void kmemcg_cache_deactivate_after_rcu(struct kmem_cache *s)
764{
765 __kmemcg_cache_deactivate_after_rcu(s);
766 percpu_ref_kill(&s->memcg_params.refcnt);
767}
768
769static void kmemcg_cache_deactivate(struct kmem_cache *s)
770{
771 if (WARN_ON_ONCE(is_root_cache(s)))
772 return;
773
774 __kmemcg_cache_deactivate(s);
775 s->flags |= SLAB_DEACTIVATED;
776
777 /*
778 * memcg_kmem_wq_lock is used to synchronize memcg_params.dying
779 * flag and make sure that no new kmem_cache deactivation tasks
780 * are queued (see flush_memcg_workqueue() ).
781 */
782 spin_lock_irq(&memcg_kmem_wq_lock);
783 if (s->memcg_params.root_cache->memcg_params.dying)
784 goto unlock;
785
786 s->memcg_params.work_fn = kmemcg_cache_deactivate_after_rcu;
787 call_rcu(&s->memcg_params.rcu_head, kmemcg_rcufn);
788unlock:
789 spin_unlock_irq(&memcg_kmem_wq_lock);
790}
791
792void memcg_deactivate_kmem_caches(struct mem_cgroup *memcg,
793 struct mem_cgroup *parent)
794{
795 int idx;
796 struct memcg_cache_array *arr;
797 struct kmem_cache *s, *c;
798 unsigned int nr_reparented;
799
800 idx = memcg_cache_id(memcg);
801
802 get_online_cpus();
803 get_online_mems();
804
805 mutex_lock(&slab_mutex);
806 list_for_each_entry(s, &slab_root_caches, root_caches_node) {
807 arr = rcu_dereference_protected(s->memcg_params.memcg_caches,
808 lockdep_is_held(&slab_mutex));
809 c = arr->entries[idx];
810 if (!c)
811 continue;
812
813 kmemcg_cache_deactivate(c);
814 arr->entries[idx] = NULL;
815 }
816 nr_reparented = 0;
817 list_for_each_entry(s, &memcg->kmem_caches,
818 memcg_params.kmem_caches_node) {
819 WRITE_ONCE(s->memcg_params.memcg, parent);
820 css_put(&memcg->css);
821 nr_reparented++;
822 }
823 if (nr_reparented) {
824 list_splice_init(&memcg->kmem_caches,
825 &parent->kmem_caches);
826 css_get_many(&parent->css, nr_reparented);
827 }
828 mutex_unlock(&slab_mutex);
829
830 put_online_mems();
831 put_online_cpus();
832}
833
834static int shutdown_memcg_caches(struct kmem_cache *s)
835{
836 struct memcg_cache_array *arr;
837 struct kmem_cache *c, *c2;
838 LIST_HEAD(busy);
839 int i;
840
841 BUG_ON(!is_root_cache(s));
842
843 /*
844 * First, shutdown active caches, i.e. caches that belong to online
845 * memory cgroups.
846 */
847 arr = rcu_dereference_protected(s->memcg_params.memcg_caches,
848 lockdep_is_held(&slab_mutex));
849 for_each_memcg_cache_index(i) {
850 c = arr->entries[i];
851 if (!c)
852 continue;
853 if (shutdown_cache(c))
854 /*
855 * The cache still has objects. Move it to a temporary
856 * list so as not to try to destroy it for a second
857 * time while iterating over inactive caches below.
858 */
859 list_move(&c->memcg_params.children_node, &busy);
860 else
861 /*
862 * The cache is empty and will be destroyed soon. Clear
863 * the pointer to it in the memcg_caches array so that
864 * it will never be accessed even if the root cache
865 * stays alive.
866 */
867 arr->entries[i] = NULL;
868 }
869
870 /*
871 * Second, shutdown all caches left from memory cgroups that are now
872 * offline.
873 */
874 list_for_each_entry_safe(c, c2, &s->memcg_params.children,
875 memcg_params.children_node)
876 shutdown_cache(c);
877
878 list_splice(&busy, &s->memcg_params.children);
879
880 /*
881 * A cache being destroyed must be empty. In particular, this means
882 * that all per memcg caches attached to it must be empty too.
883 */
884 if (!list_empty(&s->memcg_params.children))
885 return -EBUSY;
886 return 0;
887}
888
889static void flush_memcg_workqueue(struct kmem_cache *s)
890{
891 spin_lock_irq(&memcg_kmem_wq_lock);
892 s->memcg_params.dying = true;
893 spin_unlock_irq(&memcg_kmem_wq_lock);
894
895 /*
896 * SLAB and SLUB deactivate the kmem_caches through call_rcu. Make
897 * sure all registered rcu callbacks have been invoked.
898 */
899 rcu_barrier();
900
901 /*
902 * SLAB and SLUB create memcg kmem_caches through workqueue and SLUB
903 * deactivates the memcg kmem_caches through workqueue. Make sure all
904 * previous workitems on workqueue are processed.
905 */
906 flush_workqueue(memcg_kmem_cache_wq);
907}
908#else
909static inline int shutdown_memcg_caches(struct kmem_cache *s)
910{
911 return 0;
912}
913
914static inline void flush_memcg_workqueue(struct kmem_cache *s)
915{
916}
917#endif /* CONFIG_MEMCG_KMEM */
918
919void slab_kmem_cache_release(struct kmem_cache *s)
920{
921 __kmem_cache_release(s);
922 destroy_memcg_params(s);
923 kfree_const(s->name);
924 kmem_cache_free(kmem_cache, s);
925}
926
927void kmem_cache_destroy(struct kmem_cache *s)
928{
929 int err;
930
931 if (unlikely(!s))
932 return;
933
934 flush_memcg_workqueue(s);
935
936 get_online_cpus();
937 get_online_mems();
938
939 mutex_lock(&slab_mutex);
940
941 s->refcount--;
942 if (s->refcount)
943 goto out_unlock;
944
945 err = shutdown_memcg_caches(s);
946 if (!err)
947 err = shutdown_cache(s);
948
949 if (err) {
950 pr_err("kmem_cache_destroy %s: Slab cache still has objects\n",
951 s->name);
952 dump_stack();
953 }
954out_unlock:
955 mutex_unlock(&slab_mutex);
956
957 put_online_mems();
958 put_online_cpus();
959}
960EXPORT_SYMBOL(kmem_cache_destroy);
961
962/**
963 * kmem_cache_shrink - Shrink a cache.
964 * @cachep: The cache to shrink.
965 *
966 * Releases as many slabs as possible for a cache.
967 * To help debugging, a zero exit status indicates all slabs were released.
968 *
969 * Return: %0 if all slabs were released, non-zero otherwise
970 */
971int kmem_cache_shrink(struct kmem_cache *cachep)
972{
973 int ret;
974
975 get_online_cpus();
976 get_online_mems();
977 kasan_cache_shrink(cachep);
978 ret = __kmem_cache_shrink(cachep);
979 put_online_mems();
980 put_online_cpus();
981 return ret;
982}
983EXPORT_SYMBOL(kmem_cache_shrink);
984
985/**
986 * kmem_cache_shrink_all - shrink a cache and all memcg caches for root cache
987 * @s: The cache pointer
988 */
989void kmem_cache_shrink_all(struct kmem_cache *s)
990{
991 struct kmem_cache *c;
992
993 if (!IS_ENABLED(CONFIG_MEMCG_KMEM) || !is_root_cache(s)) {
994 kmem_cache_shrink(s);
995 return;
996 }
997
998 get_online_cpus();
999 get_online_mems();
1000 kasan_cache_shrink(s);
1001 __kmem_cache_shrink(s);
1002
1003 /*
1004 * We have to take the slab_mutex to protect from the memcg list
1005 * modification.
1006 */
1007 mutex_lock(&slab_mutex);
1008 for_each_memcg_cache(c, s) {
1009 /*
1010 * Don't need to shrink deactivated memcg caches.
1011 */
1012 if (s->flags & SLAB_DEACTIVATED)
1013 continue;
1014 kasan_cache_shrink(c);
1015 __kmem_cache_shrink(c);
1016 }
1017 mutex_unlock(&slab_mutex);
1018 put_online_mems();
1019 put_online_cpus();
1020}
1021
1022bool slab_is_available(void)
1023{
1024 return slab_state >= UP;
1025}
1026
1027#ifndef CONFIG_SLOB
1028/* Create a cache during boot when no slab services are available yet */
1029void __init create_boot_cache(struct kmem_cache *s, const char *name,
1030 unsigned int size, slab_flags_t flags,
1031 unsigned int useroffset, unsigned int usersize)
1032{
1033 int err;
1034 unsigned int align = ARCH_KMALLOC_MINALIGN;
1035
1036 s->name = name;
1037 s->size = s->object_size = size;
1038
1039 /*
1040 * For power of two sizes, guarantee natural alignment for kmalloc
1041 * caches, regardless of SL*B debugging options.
1042 */
1043 if (is_power_of_2(size))
1044 align = max(align, size);
1045 s->align = calculate_alignment(flags, align, size);
1046
1047 s->useroffset = useroffset;
1048 s->usersize = usersize;
1049
1050 slab_init_memcg_params(s);
1051
1052 err = __kmem_cache_create(s, flags);
1053
1054 if (err)
1055 panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n",
1056 name, size, err);
1057
1058 s->refcount = -1; /* Exempt from merging for now */
1059}
1060
1061struct kmem_cache *__init create_kmalloc_cache(const char *name,
1062 unsigned int size, slab_flags_t flags,
1063 unsigned int useroffset, unsigned int usersize)
1064{
1065 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
1066
1067 if (!s)
1068 panic("Out of memory when creating slab %s\n", name);
1069
1070 create_boot_cache(s, name, size, flags, useroffset, usersize);
1071 list_add(&s->list, &slab_caches);
1072 memcg_link_cache(s, NULL);
1073 s->refcount = 1;
1074 return s;
1075}
1076
1077struct kmem_cache *
1078kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1] __ro_after_init =
1079{ /* initialization for https://bugs.llvm.org/show_bug.cgi?id=42570 */ };
1080EXPORT_SYMBOL(kmalloc_caches);
1081
1082/*
1083 * Conversion table for small slabs sizes / 8 to the index in the
1084 * kmalloc array. This is necessary for slabs < 192 since we have non power
1085 * of two cache sizes there. The size of larger slabs can be determined using
1086 * fls.
1087 */
1088static u8 size_index[24] __ro_after_init = {
1089 3, /* 8 */
1090 4, /* 16 */
1091 5, /* 24 */
1092 5, /* 32 */
1093 6, /* 40 */
1094 6, /* 48 */
1095 6, /* 56 */
1096 6, /* 64 */
1097 1, /* 72 */
1098 1, /* 80 */
1099 1, /* 88 */
1100 1, /* 96 */
1101 7, /* 104 */
1102 7, /* 112 */
1103 7, /* 120 */
1104 7, /* 128 */
1105 2, /* 136 */
1106 2, /* 144 */
1107 2, /* 152 */
1108 2, /* 160 */
1109 2, /* 168 */
1110 2, /* 176 */
1111 2, /* 184 */
1112 2 /* 192 */
1113};
1114
1115static inline unsigned int size_index_elem(unsigned int bytes)
1116{
1117 return (bytes - 1) / 8;
1118}
1119
1120/*
1121 * Find the kmem_cache structure that serves a given size of
1122 * allocation
1123 */
1124struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
1125{
1126 unsigned int index;
1127
1128 if (size <= 192) {
1129 if (!size)
1130 return ZERO_SIZE_PTR;
1131
1132 index = size_index[size_index_elem(size)];
1133 } else {
1134 if (WARN_ON_ONCE(size > KMALLOC_MAX_CACHE_SIZE))
1135 return NULL;
1136 index = fls(size - 1);
1137 }
1138
1139 return kmalloc_caches[kmalloc_type(flags)][index];
1140}
1141
1142/*
1143 * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
1144 * kmalloc_index() supports up to 2^26=64MB, so the final entry of the table is
1145 * kmalloc-67108864.
1146 */
1147const struct kmalloc_info_struct kmalloc_info[] __initconst = {
1148 {NULL, 0}, {"kmalloc-96", 96},
1149 {"kmalloc-192", 192}, {"kmalloc-8", 8},
1150 {"kmalloc-16", 16}, {"kmalloc-32", 32},
1151 {"kmalloc-64", 64}, {"kmalloc-128", 128},
1152 {"kmalloc-256", 256}, {"kmalloc-512", 512},
1153 {"kmalloc-1k", 1024}, {"kmalloc-2k", 2048},
1154 {"kmalloc-4k", 4096}, {"kmalloc-8k", 8192},
1155 {"kmalloc-16k", 16384}, {"kmalloc-32k", 32768},
1156 {"kmalloc-64k", 65536}, {"kmalloc-128k", 131072},
1157 {"kmalloc-256k", 262144}, {"kmalloc-512k", 524288},
1158 {"kmalloc-1M", 1048576}, {"kmalloc-2M", 2097152},
1159 {"kmalloc-4M", 4194304}, {"kmalloc-8M", 8388608},
1160 {"kmalloc-16M", 16777216}, {"kmalloc-32M", 33554432},
1161 {"kmalloc-64M", 67108864}
1162};
1163
1164/*
1165 * Patch up the size_index table if we have strange large alignment
1166 * requirements for the kmalloc array. This is only the case for
1167 * MIPS it seems. The standard arches will not generate any code here.
1168 *
1169 * Largest permitted alignment is 256 bytes due to the way we
1170 * handle the index determination for the smaller caches.
1171 *
1172 * Make sure that nothing crazy happens if someone starts tinkering
1173 * around with ARCH_KMALLOC_MINALIGN
1174 */
1175void __init setup_kmalloc_cache_index_table(void)
1176{
1177 unsigned int i;
1178
1179 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
1180 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
1181
1182 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
1183 unsigned int elem = size_index_elem(i);
1184
1185 if (elem >= ARRAY_SIZE(size_index))
1186 break;
1187 size_index[elem] = KMALLOC_SHIFT_LOW;
1188 }
1189
1190 if (KMALLOC_MIN_SIZE >= 64) {
1191 /*
1192 * The 96 byte size cache is not used if the alignment
1193 * is 64 byte.
1194 */
1195 for (i = 64 + 8; i <= 96; i += 8)
1196 size_index[size_index_elem(i)] = 7;
1197
1198 }
1199
1200 if (KMALLOC_MIN_SIZE >= 128) {
1201 /*
1202 * The 192 byte sized cache is not used if the alignment
1203 * is 128 byte. Redirect kmalloc to use the 256 byte cache
1204 * instead.
1205 */
1206 for (i = 128 + 8; i <= 192; i += 8)
1207 size_index[size_index_elem(i)] = 8;
1208 }
1209}
1210
1211static const char *
1212kmalloc_cache_name(const char *prefix, unsigned int size)
1213{
1214
1215 static const char units[3] = "\0kM";
1216 int idx = 0;
1217
1218 while (size >= 1024 && (size % 1024 == 0)) {
1219 size /= 1024;
1220 idx++;
1221 }
1222
1223 return kasprintf(GFP_NOWAIT, "%s-%u%c", prefix, size, units[idx]);
1224}
1225
1226static void __init
1227new_kmalloc_cache(int idx, int type, slab_flags_t flags)
1228{
1229 const char *name;
1230
1231 if (type == KMALLOC_RECLAIM) {
1232 flags |= SLAB_RECLAIM_ACCOUNT;
1233 name = kmalloc_cache_name("kmalloc-rcl",
1234 kmalloc_info[idx].size);
1235 BUG_ON(!name);
1236 } else {
1237 name = kmalloc_info[idx].name;
1238 }
1239
1240 kmalloc_caches[type][idx] = create_kmalloc_cache(name,
1241 kmalloc_info[idx].size, flags, 0,
1242 kmalloc_info[idx].size);
1243}
1244
1245/*
1246 * Create the kmalloc array. Some of the regular kmalloc arrays
1247 * may already have been created because they were needed to
1248 * enable allocations for slab creation.
1249 */
1250void __init create_kmalloc_caches(slab_flags_t flags)
1251{
1252 int i, type;
1253
1254 for (type = KMALLOC_NORMAL; type <= KMALLOC_RECLAIM; type++) {
1255 for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
1256 if (!kmalloc_caches[type][i])
1257 new_kmalloc_cache(i, type, flags);
1258
1259 /*
1260 * Caches that are not of the two-to-the-power-of size.
1261 * These have to be created immediately after the
1262 * earlier power of two caches
1263 */
1264 if (KMALLOC_MIN_SIZE <= 32 && i == 6 &&
1265 !kmalloc_caches[type][1])
1266 new_kmalloc_cache(1, type, flags);
1267 if (KMALLOC_MIN_SIZE <= 64 && i == 7 &&
1268 !kmalloc_caches[type][2])
1269 new_kmalloc_cache(2, type, flags);
1270 }
1271 }
1272
1273 /* Kmalloc array is now usable */
1274 slab_state = UP;
1275
1276#ifdef CONFIG_ZONE_DMA
1277 for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
1278 struct kmem_cache *s = kmalloc_caches[KMALLOC_NORMAL][i];
1279
1280 if (s) {
1281 unsigned int size = kmalloc_size(i);
1282 const char *n = kmalloc_cache_name("dma-kmalloc", size);
1283
1284 BUG_ON(!n);
1285 kmalloc_caches[KMALLOC_DMA][i] = create_kmalloc_cache(
1286 n, size, SLAB_CACHE_DMA | flags, 0, 0);
1287 }
1288 }
1289#endif
1290}
1291#endif /* !CONFIG_SLOB */
1292
1293/*
1294 * To avoid unnecessary overhead, we pass through large allocation requests
1295 * directly to the page allocator. We use __GFP_COMP, because we will need to
1296 * know the allocation order to free the pages properly in kfree.
1297 */
1298void *kmalloc_order(size_t size, gfp_t flags, unsigned int order)
1299{
1300 void *ret = NULL;
1301 struct page *page;
1302
1303 flags |= __GFP_COMP;
1304 page = alloc_pages(flags, order);
1305 if (likely(page)) {
1306 ret = page_address(page);
1307 mod_node_page_state(page_pgdat(page), NR_SLAB_UNRECLAIMABLE,
1308 1 << order);
1309 }
1310 ret = kasan_kmalloc_large(ret, size, flags);
1311 /* As ret might get tagged, call kmemleak hook after KASAN. */
1312 kmemleak_alloc(ret, size, 1, flags);
1313 return ret;
1314}
1315EXPORT_SYMBOL(kmalloc_order);
1316
1317#ifdef CONFIG_TRACING
1318void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
1319{
1320 void *ret = kmalloc_order(size, flags, order);
1321 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
1322 return ret;
1323}
1324EXPORT_SYMBOL(kmalloc_order_trace);
1325#endif
1326
1327#ifdef CONFIG_SLAB_FREELIST_RANDOM
1328/* Randomize a generic freelist */
1329static void freelist_randomize(struct rnd_state *state, unsigned int *list,
1330 unsigned int count)
1331{
1332 unsigned int rand;
1333 unsigned int i;
1334
1335 for (i = 0; i < count; i++)
1336 list[i] = i;
1337
1338 /* Fisher-Yates shuffle */
1339 for (i = count - 1; i > 0; i--) {
1340 rand = prandom_u32_state(state);
1341 rand %= (i + 1);
1342 swap(list[i], list[rand]);
1343 }
1344}
1345
1346/* Create a random sequence per cache */
1347int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
1348 gfp_t gfp)
1349{
1350 struct rnd_state state;
1351
1352 if (count < 2 || cachep->random_seq)
1353 return 0;
1354
1355 cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp);
1356 if (!cachep->random_seq)
1357 return -ENOMEM;
1358
1359 /* Get best entropy at this stage of boot */
1360 prandom_seed_state(&state, get_random_long());
1361
1362 freelist_randomize(&state, cachep->random_seq, count);
1363 return 0;
1364}
1365
1366/* Destroy the per-cache random freelist sequence */
1367void cache_random_seq_destroy(struct kmem_cache *cachep)
1368{
1369 kfree(cachep->random_seq);
1370 cachep->random_seq = NULL;
1371}
1372#endif /* CONFIG_SLAB_FREELIST_RANDOM */
1373
1374#if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
1375#ifdef CONFIG_SLAB
1376#define SLABINFO_RIGHTS (0600)
1377#else
1378#define SLABINFO_RIGHTS (0400)
1379#endif
1380
1381static void print_slabinfo_header(struct seq_file *m)
1382{
1383 /*
1384 * Output format version, so at least we can change it
1385 * without _too_ many complaints.
1386 */
1387#ifdef CONFIG_DEBUG_SLAB
1388 seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
1389#else
1390 seq_puts(m, "slabinfo - version: 2.1\n");
1391#endif
1392 seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
1393 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
1394 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
1395#ifdef CONFIG_DEBUG_SLAB
1396 seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
1397 seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
1398#endif
1399 seq_putc(m, '\n');
1400}
1401
1402void *slab_start(struct seq_file *m, loff_t *pos)
1403{
1404 mutex_lock(&slab_mutex);
1405 return seq_list_start(&slab_root_caches, *pos);
1406}
1407
1408void *slab_next(struct seq_file *m, void *p, loff_t *pos)
1409{
1410 return seq_list_next(p, &slab_root_caches, pos);
1411}
1412
1413void slab_stop(struct seq_file *m, void *p)
1414{
1415 mutex_unlock(&slab_mutex);
1416}
1417
1418static void
1419memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info)
1420{
1421 struct kmem_cache *c;
1422 struct slabinfo sinfo;
1423
1424 if (!is_root_cache(s))
1425 return;
1426
1427 for_each_memcg_cache(c, s) {
1428 memset(&sinfo, 0, sizeof(sinfo));
1429 get_slabinfo(c, &sinfo);
1430
1431 info->active_slabs += sinfo.active_slabs;
1432 info->num_slabs += sinfo.num_slabs;
1433 info->shared_avail += sinfo.shared_avail;
1434 info->active_objs += sinfo.active_objs;
1435 info->num_objs += sinfo.num_objs;
1436 }
1437}
1438
1439static void cache_show(struct kmem_cache *s, struct seq_file *m)
1440{
1441 struct slabinfo sinfo;
1442
1443 memset(&sinfo, 0, sizeof(sinfo));
1444 get_slabinfo(s, &sinfo);
1445
1446 memcg_accumulate_slabinfo(s, &sinfo);
1447
1448 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
1449 cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size,
1450 sinfo.objects_per_slab, (1 << sinfo.cache_order));
1451
1452 seq_printf(m, " : tunables %4u %4u %4u",
1453 sinfo.limit, sinfo.batchcount, sinfo.shared);
1454 seq_printf(m, " : slabdata %6lu %6lu %6lu",
1455 sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
1456 slabinfo_show_stats(m, s);
1457 seq_putc(m, '\n');
1458}
1459
1460static int slab_show(struct seq_file *m, void *p)
1461{
1462 struct kmem_cache *s = list_entry(p, struct kmem_cache, root_caches_node);
1463
1464 if (p == slab_root_caches.next)
1465 print_slabinfo_header(m);
1466 cache_show(s, m);
1467 return 0;
1468}
1469
1470void dump_unreclaimable_slab(void)
1471{
1472 struct kmem_cache *s, *s2;
1473 struct slabinfo sinfo;
1474
1475 /*
1476 * Here acquiring slab_mutex is risky since we don't prefer to get
1477 * sleep in oom path. But, without mutex hold, it may introduce a
1478 * risk of crash.
1479 * Use mutex_trylock to protect the list traverse, dump nothing
1480 * without acquiring the mutex.
1481 */
1482 if (!mutex_trylock(&slab_mutex)) {
1483 pr_warn("excessive unreclaimable slab but cannot dump stats\n");
1484 return;
1485 }
1486
1487 pr_info("Unreclaimable slab info:\n");
1488 pr_info("Name Used Total\n");
1489
1490 list_for_each_entry_safe(s, s2, &slab_caches, list) {
1491 if (!is_root_cache(s) || (s->flags & SLAB_RECLAIM_ACCOUNT))
1492 continue;
1493
1494 get_slabinfo(s, &sinfo);
1495
1496 if (sinfo.num_objs > 0)
1497 pr_info("%-17s %10luKB %10luKB\n", cache_name(s),
1498 (sinfo.active_objs * s->size) / 1024,
1499 (sinfo.num_objs * s->size) / 1024);
1500 }
1501 mutex_unlock(&slab_mutex);
1502}
1503
1504#if defined(CONFIG_MEMCG)
1505void *memcg_slab_start(struct seq_file *m, loff_t *pos)
1506{
1507 struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
1508
1509 mutex_lock(&slab_mutex);
1510 return seq_list_start(&memcg->kmem_caches, *pos);
1511}
1512
1513void *memcg_slab_next(struct seq_file *m, void *p, loff_t *pos)
1514{
1515 struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
1516
1517 return seq_list_next(p, &memcg->kmem_caches, pos);
1518}
1519
1520void memcg_slab_stop(struct seq_file *m, void *p)
1521{
1522 mutex_unlock(&slab_mutex);
1523}
1524
1525int memcg_slab_show(struct seq_file *m, void *p)
1526{
1527 struct kmem_cache *s = list_entry(p, struct kmem_cache,
1528 memcg_params.kmem_caches_node);
1529 struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
1530
1531 if (p == memcg->kmem_caches.next)
1532 print_slabinfo_header(m);
1533 cache_show(s, m);
1534 return 0;
1535}
1536#endif
1537
1538/*
1539 * slabinfo_op - iterator that generates /proc/slabinfo
1540 *
1541 * Output layout:
1542 * cache-name
1543 * num-active-objs
1544 * total-objs
1545 * object size
1546 * num-active-slabs
1547 * total-slabs
1548 * num-pages-per-slab
1549 * + further values on SMP and with statistics enabled
1550 */
1551static const struct seq_operations slabinfo_op = {
1552 .start = slab_start,
1553 .next = slab_next,
1554 .stop = slab_stop,
1555 .show = slab_show,
1556};
1557
1558static int slabinfo_open(struct inode *inode, struct file *file)
1559{
1560 return seq_open(file, &slabinfo_op);
1561}
1562
1563static const struct file_operations proc_slabinfo_operations = {
1564 .open = slabinfo_open,
1565 .read = seq_read,
1566 .write = slabinfo_write,
1567 .llseek = seq_lseek,
1568 .release = seq_release,
1569};
1570
1571static int __init slab_proc_init(void)
1572{
1573 proc_create("slabinfo", SLABINFO_RIGHTS, NULL,
1574 &proc_slabinfo_operations);
1575 return 0;
1576}
1577module_init(slab_proc_init);
1578
1579#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_MEMCG_KMEM)
1580/*
1581 * Display information about kmem caches that have child memcg caches.
1582 */
1583static int memcg_slabinfo_show(struct seq_file *m, void *unused)
1584{
1585 struct kmem_cache *s, *c;
1586 struct slabinfo sinfo;
1587
1588 mutex_lock(&slab_mutex);
1589 seq_puts(m, "# <name> <css_id[:dead|deact]> <active_objs> <num_objs>");
1590 seq_puts(m, " <active_slabs> <num_slabs>\n");
1591 list_for_each_entry(s, &slab_root_caches, root_caches_node) {
1592 /*
1593 * Skip kmem caches that don't have any memcg children.
1594 */
1595 if (list_empty(&s->memcg_params.children))
1596 continue;
1597
1598 memset(&sinfo, 0, sizeof(sinfo));
1599 get_slabinfo(s, &sinfo);
1600 seq_printf(m, "%-17s root %6lu %6lu %6lu %6lu\n",
1601 cache_name(s), sinfo.active_objs, sinfo.num_objs,
1602 sinfo.active_slabs, sinfo.num_slabs);
1603
1604 for_each_memcg_cache(c, s) {
1605 struct cgroup_subsys_state *css;
1606 char *status = "";
1607
1608 css = &c->memcg_params.memcg->css;
1609 if (!(css->flags & CSS_ONLINE))
1610 status = ":dead";
1611 else if (c->flags & SLAB_DEACTIVATED)
1612 status = ":deact";
1613
1614 memset(&sinfo, 0, sizeof(sinfo));
1615 get_slabinfo(c, &sinfo);
1616 seq_printf(m, "%-17s %4d%-6s %6lu %6lu %6lu %6lu\n",
1617 cache_name(c), css->id, status,
1618 sinfo.active_objs, sinfo.num_objs,
1619 sinfo.active_slabs, sinfo.num_slabs);
1620 }
1621 }
1622 mutex_unlock(&slab_mutex);
1623 return 0;
1624}
1625DEFINE_SHOW_ATTRIBUTE(memcg_slabinfo);
1626
1627static int __init memcg_slabinfo_init(void)
1628{
1629 debugfs_create_file("memcg_slabinfo", S_IFREG | S_IRUGO,
1630 NULL, NULL, &memcg_slabinfo_fops);
1631 return 0;
1632}
1633
1634late_initcall(memcg_slabinfo_init);
1635#endif /* CONFIG_DEBUG_FS && CONFIG_MEMCG_KMEM */
1636#endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */
1637
1638static __always_inline void *__do_krealloc(const void *p, size_t new_size,
1639 gfp_t flags)
1640{
1641 void *ret;
1642 size_t ks = 0;
1643
1644 if (p)
1645 ks = ksize(p);
1646
1647 if (ks >= new_size) {
1648 p = kasan_krealloc((void *)p, new_size, flags);
1649 return (void *)p;
1650 }
1651
1652 ret = kmalloc_track_caller(new_size, flags);
1653 if (ret && p)
1654 memcpy(ret, p, ks);
1655
1656 return ret;
1657}
1658
1659/**
1660 * __krealloc - like krealloc() but don't free @p.
1661 * @p: object to reallocate memory for.
1662 * @new_size: how many bytes of memory are required.
1663 * @flags: the type of memory to allocate.
1664 *
1665 * This function is like krealloc() except it never frees the originally
1666 * allocated buffer. Use this if you don't want to free the buffer immediately
1667 * like, for example, with RCU.
1668 *
1669 * Return: pointer to the allocated memory or %NULL in case of error
1670 */
1671void *__krealloc(const void *p, size_t new_size, gfp_t flags)
1672{
1673 if (unlikely(!new_size))
1674 return ZERO_SIZE_PTR;
1675
1676 return __do_krealloc(p, new_size, flags);
1677
1678}
1679EXPORT_SYMBOL(__krealloc);
1680
1681/**
1682 * krealloc - reallocate memory. The contents will remain unchanged.
1683 * @p: object to reallocate memory for.
1684 * @new_size: how many bytes of memory are required.
1685 * @flags: the type of memory to allocate.
1686 *
1687 * The contents of the object pointed to are preserved up to the
1688 * lesser of the new and old sizes. If @p is %NULL, krealloc()
1689 * behaves exactly like kmalloc(). If @new_size is 0 and @p is not a
1690 * %NULL pointer, the object pointed to is freed.
1691 *
1692 * Return: pointer to the allocated memory or %NULL in case of error
1693 */
1694void *krealloc(const void *p, size_t new_size, gfp_t flags)
1695{
1696 void *ret;
1697
1698 if (unlikely(!new_size)) {
1699 kfree(p);
1700 return ZERO_SIZE_PTR;
1701 }
1702
1703 ret = __do_krealloc(p, new_size, flags);
1704 if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret))
1705 kfree(p);
1706
1707 return ret;
1708}
1709EXPORT_SYMBOL(krealloc);
1710
1711/**
1712 * kzfree - like kfree but zero memory
1713 * @p: object to free memory of
1714 *
1715 * The memory of the object @p points to is zeroed before freed.
1716 * If @p is %NULL, kzfree() does nothing.
1717 *
1718 * Note: this function zeroes the whole allocated buffer which can be a good
1719 * deal bigger than the requested buffer size passed to kmalloc(). So be
1720 * careful when using this function in performance sensitive code.
1721 */
1722void kzfree(const void *p)
1723{
1724 size_t ks;
1725 void *mem = (void *)p;
1726
1727 if (unlikely(ZERO_OR_NULL_PTR(mem)))
1728 return;
1729 ks = ksize(mem);
1730 memset(mem, 0, ks);
1731 kfree(mem);
1732}
1733EXPORT_SYMBOL(kzfree);
1734
1735/**
1736 * ksize - get the actual amount of memory allocated for a given object
1737 * @objp: Pointer to the object
1738 *
1739 * kmalloc may internally round up allocations and return more memory
1740 * than requested. ksize() can be used to determine the actual amount of
1741 * memory allocated. The caller may use this additional memory, even though
1742 * a smaller amount of memory was initially specified with the kmalloc call.
1743 * The caller must guarantee that objp points to a valid object previously
1744 * allocated with either kmalloc() or kmem_cache_alloc(). The object
1745 * must not be freed during the duration of the call.
1746 *
1747 * Return: size of the actual memory used by @objp in bytes
1748 */
1749size_t ksize(const void *objp)
1750{
1751 size_t size;
1752
1753 if (WARN_ON_ONCE(!objp))
1754 return 0;
1755 /*
1756 * We need to check that the pointed to object is valid, and only then
1757 * unpoison the shadow memory below. We use __kasan_check_read(), to
1758 * generate a more useful report at the time ksize() is called (rather
1759 * than later where behaviour is undefined due to potential
1760 * use-after-free or double-free).
1761 *
1762 * If the pointed to memory is invalid we return 0, to avoid users of
1763 * ksize() writing to and potentially corrupting the memory region.
1764 *
1765 * We want to perform the check before __ksize(), to avoid potentially
1766 * crashing in __ksize() due to accessing invalid metadata.
1767 */
1768 if (unlikely(objp == ZERO_SIZE_PTR) || !__kasan_check_read(objp, 1))
1769 return 0;
1770
1771 size = __ksize(objp);
1772 /*
1773 * We assume that ksize callers could use whole allocated area,
1774 * so we need to unpoison this area.
1775 */
1776 kasan_unpoison_shadow(objp, size);
1777 return size;
1778}
1779EXPORT_SYMBOL(ksize);
1780
1781/* Tracepoints definitions. */
1782EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1783EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1784EXPORT_TRACEPOINT_SYMBOL(kmalloc_node);
1785EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node);
1786EXPORT_TRACEPOINT_SYMBOL(kfree);
1787EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);
1788
1789int should_failslab(struct kmem_cache *s, gfp_t gfpflags)
1790{
1791 if (__should_failslab(s, gfpflags))
1792 return -ENOMEM;
1793 return 0;
1794}
1795ALLOW_ERROR_INJECTION(should_failslab, ERRNO);
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/kfence.h>
16#include <linux/module.h>
17#include <linux/cpu.h>
18#include <linux/uaccess.h>
19#include <linux/seq_file.h>
20#include <linux/dma-mapping.h>
21#include <linux/swiotlb.h>
22#include <linux/proc_fs.h>
23#include <linux/debugfs.h>
24#include <linux/kmemleak.h>
25#include <linux/kasan.h>
26#include <asm/cacheflush.h>
27#include <asm/tlbflush.h>
28#include <asm/page.h>
29#include <linux/memcontrol.h>
30#include <linux/stackdepot.h>
31
32#include "internal.h"
33#include "slab.h"
34
35#define CREATE_TRACE_POINTS
36#include <trace/events/kmem.h>
37
38enum slab_state slab_state;
39LIST_HEAD(slab_caches);
40DEFINE_MUTEX(slab_mutex);
41struct kmem_cache *kmem_cache;
42
43/*
44 * Set of flags that will prevent slab merging
45 */
46#define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
47 SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \
48 SLAB_FAILSLAB | SLAB_NO_MERGE)
49
50#define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \
51 SLAB_CACHE_DMA32 | SLAB_ACCOUNT)
52
53/*
54 * Merge control. If this is set then no merging of slab caches will occur.
55 */
56static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT);
57
58static int __init setup_slab_nomerge(char *str)
59{
60 slab_nomerge = true;
61 return 1;
62}
63
64static int __init setup_slab_merge(char *str)
65{
66 slab_nomerge = false;
67 return 1;
68}
69
70__setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0);
71__setup_param("slub_merge", slub_merge, setup_slab_merge, 0);
72
73__setup("slab_nomerge", setup_slab_nomerge);
74__setup("slab_merge", setup_slab_merge);
75
76/*
77 * Determine the size of a slab object
78 */
79unsigned int kmem_cache_size(struct kmem_cache *s)
80{
81 return s->object_size;
82}
83EXPORT_SYMBOL(kmem_cache_size);
84
85#ifdef CONFIG_DEBUG_VM
86
87static bool kmem_cache_is_duplicate_name(const char *name)
88{
89 struct kmem_cache *s;
90
91 list_for_each_entry(s, &slab_caches, list) {
92 if (!strcmp(s->name, name))
93 return true;
94 }
95
96 return false;
97}
98
99static int kmem_cache_sanity_check(const char *name, unsigned int size)
100{
101 if (!name || in_interrupt() || size > KMALLOC_MAX_SIZE) {
102 pr_err("kmem_cache_create(%s) integrity check failed\n", name);
103 return -EINVAL;
104 }
105
106 /* Duplicate names will confuse slabtop, et al */
107 WARN(kmem_cache_is_duplicate_name(name),
108 "kmem_cache of name '%s' already exists\n", name);
109
110 WARN_ON(strchr(name, ' ')); /* It confuses parsers */
111 return 0;
112}
113#else
114static inline int kmem_cache_sanity_check(const char *name, unsigned int size)
115{
116 return 0;
117}
118#endif
119
120/*
121 * Figure out what the alignment of the objects will be given a set of
122 * flags, a user specified alignment and the size of the objects.
123 */
124static unsigned int calculate_alignment(slab_flags_t flags,
125 unsigned int align, unsigned int size)
126{
127 /*
128 * If the user wants hardware cache aligned objects then follow that
129 * suggestion if the object is sufficiently large.
130 *
131 * The hardware cache alignment cannot override the specified
132 * alignment though. If that is greater then use it.
133 */
134 if (flags & SLAB_HWCACHE_ALIGN) {
135 unsigned int ralign;
136
137 ralign = cache_line_size();
138 while (size <= ralign / 2)
139 ralign /= 2;
140 align = max(align, ralign);
141 }
142
143 align = max(align, arch_slab_minalign());
144
145 return ALIGN(align, sizeof(void *));
146}
147
148/*
149 * Find a mergeable slab cache
150 */
151int slab_unmergeable(struct kmem_cache *s)
152{
153 if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
154 return 1;
155
156 if (s->ctor)
157 return 1;
158
159#ifdef CONFIG_HARDENED_USERCOPY
160 if (s->usersize)
161 return 1;
162#endif
163
164 /*
165 * We may have set a slab to be unmergeable during bootstrap.
166 */
167 if (s->refcount < 0)
168 return 1;
169
170 return 0;
171}
172
173struct kmem_cache *find_mergeable(unsigned int size, unsigned int align,
174 slab_flags_t flags, const char *name, void (*ctor)(void *))
175{
176 struct kmem_cache *s;
177
178 if (slab_nomerge)
179 return NULL;
180
181 if (ctor)
182 return NULL;
183
184 flags = kmem_cache_flags(flags, name);
185
186 if (flags & SLAB_NEVER_MERGE)
187 return NULL;
188
189 size = ALIGN(size, sizeof(void *));
190 align = calculate_alignment(flags, align, size);
191 size = ALIGN(size, align);
192
193 list_for_each_entry_reverse(s, &slab_caches, list) {
194 if (slab_unmergeable(s))
195 continue;
196
197 if (size > s->size)
198 continue;
199
200 if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
201 continue;
202 /*
203 * Check if alignment is compatible.
204 * Courtesy of Adrian Drzewiecki
205 */
206 if ((s->size & ~(align - 1)) != s->size)
207 continue;
208
209 if (s->size - size >= sizeof(void *))
210 continue;
211
212 return s;
213 }
214 return NULL;
215}
216
217static struct kmem_cache *create_cache(const char *name,
218 unsigned int object_size,
219 struct kmem_cache_args *args,
220 slab_flags_t flags)
221{
222 struct kmem_cache *s;
223 int err;
224
225 /* If a custom freelist pointer is requested make sure it's sane. */
226 err = -EINVAL;
227 if (args->use_freeptr_offset &&
228 (args->freeptr_offset >= object_size ||
229 !(flags & SLAB_TYPESAFE_BY_RCU) ||
230 !IS_ALIGNED(args->freeptr_offset, __alignof__(freeptr_t))))
231 goto out;
232
233 err = -ENOMEM;
234 s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
235 if (!s)
236 goto out;
237 err = do_kmem_cache_create(s, name, object_size, args, flags);
238 if (err)
239 goto out_free_cache;
240
241 s->refcount = 1;
242 list_add(&s->list, &slab_caches);
243 return s;
244
245out_free_cache:
246 kmem_cache_free(kmem_cache, s);
247out:
248 return ERR_PTR(err);
249}
250
251/**
252 * __kmem_cache_create_args - Create a kmem cache.
253 * @name: A string which is used in /proc/slabinfo to identify this cache.
254 * @object_size: The size of objects to be created in this cache.
255 * @args: Additional arguments for the cache creation (see
256 * &struct kmem_cache_args).
257 * @flags: See the desriptions of individual flags. The common ones are listed
258 * in the description below.
259 *
260 * Not to be called directly, use the kmem_cache_create() wrapper with the same
261 * parameters.
262 *
263 * Commonly used @flags:
264 *
265 * &SLAB_ACCOUNT - Account allocations to memcg.
266 *
267 * &SLAB_HWCACHE_ALIGN - Align objects on cache line boundaries.
268 *
269 * &SLAB_RECLAIM_ACCOUNT - Objects are reclaimable.
270 *
271 * &SLAB_TYPESAFE_BY_RCU - Slab page (not individual objects) freeing delayed
272 * by a grace period - see the full description before using.
273 *
274 * Context: Cannot be called within a interrupt, but can be interrupted.
275 *
276 * Return: a pointer to the cache on success, NULL on failure.
277 */
278struct kmem_cache *__kmem_cache_create_args(const char *name,
279 unsigned int object_size,
280 struct kmem_cache_args *args,
281 slab_flags_t flags)
282{
283 struct kmem_cache *s = NULL;
284 const char *cache_name;
285 int err;
286
287#ifdef CONFIG_SLUB_DEBUG
288 /*
289 * If no slab_debug was enabled globally, the static key is not yet
290 * enabled by setup_slub_debug(). Enable it if the cache is being
291 * created with any of the debugging flags passed explicitly.
292 * It's also possible that this is the first cache created with
293 * SLAB_STORE_USER and we should init stack_depot for it.
294 */
295 if (flags & SLAB_DEBUG_FLAGS)
296 static_branch_enable(&slub_debug_enabled);
297 if (flags & SLAB_STORE_USER)
298 stack_depot_init();
299#endif
300
301 mutex_lock(&slab_mutex);
302
303 err = kmem_cache_sanity_check(name, object_size);
304 if (err) {
305 goto out_unlock;
306 }
307
308 /* Refuse requests with allocator specific flags */
309 if (flags & ~SLAB_FLAGS_PERMITTED) {
310 err = -EINVAL;
311 goto out_unlock;
312 }
313
314 /*
315 * Some allocators will constraint the set of valid flags to a subset
316 * of all flags. We expect them to define CACHE_CREATE_MASK in this
317 * case, and we'll just provide them with a sanitized version of the
318 * passed flags.
319 */
320 flags &= CACHE_CREATE_MASK;
321
322 /* Fail closed on bad usersize of useroffset values. */
323 if (!IS_ENABLED(CONFIG_HARDENED_USERCOPY) ||
324 WARN_ON(!args->usersize && args->useroffset) ||
325 WARN_ON(object_size < args->usersize ||
326 object_size - args->usersize < args->useroffset))
327 args->usersize = args->useroffset = 0;
328
329 if (!args->usersize)
330 s = __kmem_cache_alias(name, object_size, args->align, flags,
331 args->ctor);
332 if (s)
333 goto out_unlock;
334
335 cache_name = kstrdup_const(name, GFP_KERNEL);
336 if (!cache_name) {
337 err = -ENOMEM;
338 goto out_unlock;
339 }
340
341 args->align = calculate_alignment(flags, args->align, object_size);
342 s = create_cache(cache_name, object_size, args, flags);
343 if (IS_ERR(s)) {
344 err = PTR_ERR(s);
345 kfree_const(cache_name);
346 }
347
348out_unlock:
349 mutex_unlock(&slab_mutex);
350
351 if (err) {
352 if (flags & SLAB_PANIC)
353 panic("%s: Failed to create slab '%s'. Error %d\n",
354 __func__, name, err);
355 else {
356 pr_warn("%s(%s) failed with error %d\n",
357 __func__, name, err);
358 dump_stack();
359 }
360 return NULL;
361 }
362 return s;
363}
364EXPORT_SYMBOL(__kmem_cache_create_args);
365
366static struct kmem_cache *kmem_buckets_cache __ro_after_init;
367
368/**
369 * kmem_buckets_create - Create a set of caches that handle dynamic sized
370 * allocations via kmem_buckets_alloc()
371 * @name: A prefix string which is used in /proc/slabinfo to identify this
372 * cache. The individual caches with have their sizes as the suffix.
373 * @flags: SLAB flags (see kmem_cache_create() for details).
374 * @useroffset: Starting offset within an allocation that may be copied
375 * to/from userspace.
376 * @usersize: How many bytes, starting at @useroffset, may be copied
377 * to/from userspace.
378 * @ctor: A constructor for the objects, run when new allocations are made.
379 *
380 * Cannot be called within an interrupt, but can be interrupted.
381 *
382 * Return: a pointer to the cache on success, NULL on failure. When
383 * CONFIG_SLAB_BUCKETS is not enabled, ZERO_SIZE_PTR is returned, and
384 * subsequent calls to kmem_buckets_alloc() will fall back to kmalloc().
385 * (i.e. callers only need to check for NULL on failure.)
386 */
387kmem_buckets *kmem_buckets_create(const char *name, slab_flags_t flags,
388 unsigned int useroffset,
389 unsigned int usersize,
390 void (*ctor)(void *))
391{
392 unsigned long mask = 0;
393 unsigned int idx;
394 kmem_buckets *b;
395
396 BUILD_BUG_ON(ARRAY_SIZE(kmalloc_caches[KMALLOC_NORMAL]) > BITS_PER_LONG);
397
398 /*
399 * When the separate buckets API is not built in, just return
400 * a non-NULL value for the kmem_buckets pointer, which will be
401 * unused when performing allocations.
402 */
403 if (!IS_ENABLED(CONFIG_SLAB_BUCKETS))
404 return ZERO_SIZE_PTR;
405
406 if (WARN_ON(!kmem_buckets_cache))
407 return NULL;
408
409 b = kmem_cache_alloc(kmem_buckets_cache, GFP_KERNEL|__GFP_ZERO);
410 if (WARN_ON(!b))
411 return NULL;
412
413 flags |= SLAB_NO_MERGE;
414
415 for (idx = 0; idx < ARRAY_SIZE(kmalloc_caches[KMALLOC_NORMAL]); idx++) {
416 char *short_size, *cache_name;
417 unsigned int cache_useroffset, cache_usersize;
418 unsigned int size, aligned_idx;
419
420 if (!kmalloc_caches[KMALLOC_NORMAL][idx])
421 continue;
422
423 size = kmalloc_caches[KMALLOC_NORMAL][idx]->object_size;
424 if (!size)
425 continue;
426
427 short_size = strchr(kmalloc_caches[KMALLOC_NORMAL][idx]->name, '-');
428 if (WARN_ON(!short_size))
429 goto fail;
430
431 if (useroffset >= size) {
432 cache_useroffset = 0;
433 cache_usersize = 0;
434 } else {
435 cache_useroffset = useroffset;
436 cache_usersize = min(size - cache_useroffset, usersize);
437 }
438
439 aligned_idx = __kmalloc_index(size, false);
440 if (!(*b)[aligned_idx]) {
441 cache_name = kasprintf(GFP_KERNEL, "%s-%s", name, short_size + 1);
442 if (WARN_ON(!cache_name))
443 goto fail;
444 (*b)[aligned_idx] = kmem_cache_create_usercopy(cache_name, size,
445 0, flags, cache_useroffset,
446 cache_usersize, ctor);
447 kfree(cache_name);
448 if (WARN_ON(!(*b)[aligned_idx]))
449 goto fail;
450 set_bit(aligned_idx, &mask);
451 }
452 if (idx != aligned_idx)
453 (*b)[idx] = (*b)[aligned_idx];
454 }
455
456 return b;
457
458fail:
459 for_each_set_bit(idx, &mask, ARRAY_SIZE(kmalloc_caches[KMALLOC_NORMAL]))
460 kmem_cache_destroy((*b)[idx]);
461 kmem_cache_free(kmem_buckets_cache, b);
462
463 return NULL;
464}
465EXPORT_SYMBOL(kmem_buckets_create);
466
467/*
468 * For a given kmem_cache, kmem_cache_destroy() should only be called
469 * once or there will be a use-after-free problem. The actual deletion
470 * and release of the kobject does not need slab_mutex or cpu_hotplug_lock
471 * protection. So they are now done without holding those locks.
472 */
473static void kmem_cache_release(struct kmem_cache *s)
474{
475 kfence_shutdown_cache(s);
476 if (__is_defined(SLAB_SUPPORTS_SYSFS) && slab_state >= FULL)
477 sysfs_slab_release(s);
478 else
479 slab_kmem_cache_release(s);
480}
481
482void slab_kmem_cache_release(struct kmem_cache *s)
483{
484 __kmem_cache_release(s);
485 kfree_const(s->name);
486 kmem_cache_free(kmem_cache, s);
487}
488
489void kmem_cache_destroy(struct kmem_cache *s)
490{
491 int err;
492
493 if (unlikely(!s) || !kasan_check_byte(s))
494 return;
495
496 /* in-flight kfree_rcu()'s may include objects from our cache */
497 kvfree_rcu_barrier();
498
499 if (IS_ENABLED(CONFIG_SLUB_RCU_DEBUG) &&
500 (s->flags & SLAB_TYPESAFE_BY_RCU)) {
501 /*
502 * Under CONFIG_SLUB_RCU_DEBUG, when objects in a
503 * SLAB_TYPESAFE_BY_RCU slab are freed, SLUB will internally
504 * defer their freeing with call_rcu().
505 * Wait for such call_rcu() invocations here before actually
506 * destroying the cache.
507 *
508 * It doesn't matter that we haven't looked at the slab refcount
509 * yet - slabs with SLAB_TYPESAFE_BY_RCU can't be merged, so
510 * the refcount should be 1 here.
511 */
512 rcu_barrier();
513 }
514
515 cpus_read_lock();
516 mutex_lock(&slab_mutex);
517
518 s->refcount--;
519 if (s->refcount) {
520 mutex_unlock(&slab_mutex);
521 cpus_read_unlock();
522 return;
523 }
524
525 /* free asan quarantined objects */
526 kasan_cache_shutdown(s);
527
528 err = __kmem_cache_shutdown(s);
529 if (!slab_in_kunit_test())
530 WARN(err, "%s %s: Slab cache still has objects when called from %pS",
531 __func__, s->name, (void *)_RET_IP_);
532
533 list_del(&s->list);
534
535 mutex_unlock(&slab_mutex);
536 cpus_read_unlock();
537
538 if (slab_state >= FULL)
539 sysfs_slab_unlink(s);
540 debugfs_slab_release(s);
541
542 if (err)
543 return;
544
545 if (s->flags & SLAB_TYPESAFE_BY_RCU)
546 rcu_barrier();
547
548 kmem_cache_release(s);
549}
550EXPORT_SYMBOL(kmem_cache_destroy);
551
552/**
553 * kmem_cache_shrink - Shrink a cache.
554 * @cachep: The cache to shrink.
555 *
556 * Releases as many slabs as possible for a cache.
557 * To help debugging, a zero exit status indicates all slabs were released.
558 *
559 * Return: %0 if all slabs were released, non-zero otherwise
560 */
561int kmem_cache_shrink(struct kmem_cache *cachep)
562{
563 kasan_cache_shrink(cachep);
564
565 return __kmem_cache_shrink(cachep);
566}
567EXPORT_SYMBOL(kmem_cache_shrink);
568
569bool slab_is_available(void)
570{
571 return slab_state >= UP;
572}
573
574#ifdef CONFIG_PRINTK
575static void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab)
576{
577 if (__kfence_obj_info(kpp, object, slab))
578 return;
579 __kmem_obj_info(kpp, object, slab);
580}
581
582/**
583 * kmem_dump_obj - Print available slab provenance information
584 * @object: slab object for which to find provenance information.
585 *
586 * This function uses pr_cont(), so that the caller is expected to have
587 * printed out whatever preamble is appropriate. The provenance information
588 * depends on the type of object and on how much debugging is enabled.
589 * For a slab-cache object, the fact that it is a slab object is printed,
590 * and, if available, the slab name, return address, and stack trace from
591 * the allocation and last free path of that object.
592 *
593 * Return: %true if the pointer is to a not-yet-freed object from
594 * kmalloc() or kmem_cache_alloc(), either %true or %false if the pointer
595 * is to an already-freed object, and %false otherwise.
596 */
597bool kmem_dump_obj(void *object)
598{
599 char *cp = IS_ENABLED(CONFIG_MMU) ? "" : "/vmalloc";
600 int i;
601 struct slab *slab;
602 unsigned long ptroffset;
603 struct kmem_obj_info kp = { };
604
605 /* Some arches consider ZERO_SIZE_PTR to be a valid address. */
606 if (object < (void *)PAGE_SIZE || !virt_addr_valid(object))
607 return false;
608 slab = virt_to_slab(object);
609 if (!slab)
610 return false;
611
612 kmem_obj_info(&kp, object, slab);
613 if (kp.kp_slab_cache)
614 pr_cont(" slab%s %s", cp, kp.kp_slab_cache->name);
615 else
616 pr_cont(" slab%s", cp);
617 if (is_kfence_address(object))
618 pr_cont(" (kfence)");
619 if (kp.kp_objp)
620 pr_cont(" start %px", kp.kp_objp);
621 if (kp.kp_data_offset)
622 pr_cont(" data offset %lu", kp.kp_data_offset);
623 if (kp.kp_objp) {
624 ptroffset = ((char *)object - (char *)kp.kp_objp) - kp.kp_data_offset;
625 pr_cont(" pointer offset %lu", ptroffset);
626 }
627 if (kp.kp_slab_cache && kp.kp_slab_cache->object_size)
628 pr_cont(" size %u", kp.kp_slab_cache->object_size);
629 if (kp.kp_ret)
630 pr_cont(" allocated at %pS\n", kp.kp_ret);
631 else
632 pr_cont("\n");
633 for (i = 0; i < ARRAY_SIZE(kp.kp_stack); i++) {
634 if (!kp.kp_stack[i])
635 break;
636 pr_info(" %pS\n", kp.kp_stack[i]);
637 }
638
639 if (kp.kp_free_stack[0])
640 pr_cont(" Free path:\n");
641
642 for (i = 0; i < ARRAY_SIZE(kp.kp_free_stack); i++) {
643 if (!kp.kp_free_stack[i])
644 break;
645 pr_info(" %pS\n", kp.kp_free_stack[i]);
646 }
647
648 return true;
649}
650EXPORT_SYMBOL_GPL(kmem_dump_obj);
651#endif
652
653/* Create a cache during boot when no slab services are available yet */
654void __init create_boot_cache(struct kmem_cache *s, const char *name,
655 unsigned int size, slab_flags_t flags,
656 unsigned int useroffset, unsigned int usersize)
657{
658 int err;
659 unsigned int align = ARCH_KMALLOC_MINALIGN;
660 struct kmem_cache_args kmem_args = {};
661
662 /*
663 * kmalloc caches guarantee alignment of at least the largest
664 * power-of-two divisor of the size. For power-of-two sizes,
665 * it is the size itself.
666 */
667 if (flags & SLAB_KMALLOC)
668 align = max(align, 1U << (ffs(size) - 1));
669 kmem_args.align = calculate_alignment(flags, align, size);
670
671#ifdef CONFIG_HARDENED_USERCOPY
672 kmem_args.useroffset = useroffset;
673 kmem_args.usersize = usersize;
674#endif
675
676 err = do_kmem_cache_create(s, name, size, &kmem_args, flags);
677
678 if (err)
679 panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n",
680 name, size, err);
681
682 s->refcount = -1; /* Exempt from merging for now */
683}
684
685static struct kmem_cache *__init create_kmalloc_cache(const char *name,
686 unsigned int size,
687 slab_flags_t flags)
688{
689 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
690
691 if (!s)
692 panic("Out of memory when creating slab %s\n", name);
693
694 create_boot_cache(s, name, size, flags | SLAB_KMALLOC, 0, size);
695 list_add(&s->list, &slab_caches);
696 s->refcount = 1;
697 return s;
698}
699
700kmem_buckets kmalloc_caches[NR_KMALLOC_TYPES] __ro_after_init =
701{ /* initialization for https://llvm.org/pr42570 */ };
702EXPORT_SYMBOL(kmalloc_caches);
703
704#ifdef CONFIG_RANDOM_KMALLOC_CACHES
705unsigned long random_kmalloc_seed __ro_after_init;
706EXPORT_SYMBOL(random_kmalloc_seed);
707#endif
708
709/*
710 * Conversion table for small slabs sizes / 8 to the index in the
711 * kmalloc array. This is necessary for slabs < 192 since we have non power
712 * of two cache sizes there. The size of larger slabs can be determined using
713 * fls.
714 */
715u8 kmalloc_size_index[24] __ro_after_init = {
716 3, /* 8 */
717 4, /* 16 */
718 5, /* 24 */
719 5, /* 32 */
720 6, /* 40 */
721 6, /* 48 */
722 6, /* 56 */
723 6, /* 64 */
724 1, /* 72 */
725 1, /* 80 */
726 1, /* 88 */
727 1, /* 96 */
728 7, /* 104 */
729 7, /* 112 */
730 7, /* 120 */
731 7, /* 128 */
732 2, /* 136 */
733 2, /* 144 */
734 2, /* 152 */
735 2, /* 160 */
736 2, /* 168 */
737 2, /* 176 */
738 2, /* 184 */
739 2 /* 192 */
740};
741
742size_t kmalloc_size_roundup(size_t size)
743{
744 if (size && size <= KMALLOC_MAX_CACHE_SIZE) {
745 /*
746 * The flags don't matter since size_index is common to all.
747 * Neither does the caller for just getting ->object_size.
748 */
749 return kmalloc_slab(size, NULL, GFP_KERNEL, 0)->object_size;
750 }
751
752 /* Above the smaller buckets, size is a multiple of page size. */
753 if (size && size <= KMALLOC_MAX_SIZE)
754 return PAGE_SIZE << get_order(size);
755
756 /*
757 * Return 'size' for 0 - kmalloc() returns ZERO_SIZE_PTR
758 * and very large size - kmalloc() may fail.
759 */
760 return size;
761
762}
763EXPORT_SYMBOL(kmalloc_size_roundup);
764
765#ifdef CONFIG_ZONE_DMA
766#define KMALLOC_DMA_NAME(sz) .name[KMALLOC_DMA] = "dma-kmalloc-" #sz,
767#else
768#define KMALLOC_DMA_NAME(sz)
769#endif
770
771#ifdef CONFIG_MEMCG
772#define KMALLOC_CGROUP_NAME(sz) .name[KMALLOC_CGROUP] = "kmalloc-cg-" #sz,
773#else
774#define KMALLOC_CGROUP_NAME(sz)
775#endif
776
777#ifndef CONFIG_SLUB_TINY
778#define KMALLOC_RCL_NAME(sz) .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #sz,
779#else
780#define KMALLOC_RCL_NAME(sz)
781#endif
782
783#ifdef CONFIG_RANDOM_KMALLOC_CACHES
784#define __KMALLOC_RANDOM_CONCAT(a, b) a ## b
785#define KMALLOC_RANDOM_NAME(N, sz) __KMALLOC_RANDOM_CONCAT(KMA_RAND_, N)(sz)
786#define KMA_RAND_1(sz) .name[KMALLOC_RANDOM_START + 1] = "kmalloc-rnd-01-" #sz,
787#define KMA_RAND_2(sz) KMA_RAND_1(sz) .name[KMALLOC_RANDOM_START + 2] = "kmalloc-rnd-02-" #sz,
788#define KMA_RAND_3(sz) KMA_RAND_2(sz) .name[KMALLOC_RANDOM_START + 3] = "kmalloc-rnd-03-" #sz,
789#define KMA_RAND_4(sz) KMA_RAND_3(sz) .name[KMALLOC_RANDOM_START + 4] = "kmalloc-rnd-04-" #sz,
790#define KMA_RAND_5(sz) KMA_RAND_4(sz) .name[KMALLOC_RANDOM_START + 5] = "kmalloc-rnd-05-" #sz,
791#define KMA_RAND_6(sz) KMA_RAND_5(sz) .name[KMALLOC_RANDOM_START + 6] = "kmalloc-rnd-06-" #sz,
792#define KMA_RAND_7(sz) KMA_RAND_6(sz) .name[KMALLOC_RANDOM_START + 7] = "kmalloc-rnd-07-" #sz,
793#define KMA_RAND_8(sz) KMA_RAND_7(sz) .name[KMALLOC_RANDOM_START + 8] = "kmalloc-rnd-08-" #sz,
794#define KMA_RAND_9(sz) KMA_RAND_8(sz) .name[KMALLOC_RANDOM_START + 9] = "kmalloc-rnd-09-" #sz,
795#define KMA_RAND_10(sz) KMA_RAND_9(sz) .name[KMALLOC_RANDOM_START + 10] = "kmalloc-rnd-10-" #sz,
796#define KMA_RAND_11(sz) KMA_RAND_10(sz) .name[KMALLOC_RANDOM_START + 11] = "kmalloc-rnd-11-" #sz,
797#define KMA_RAND_12(sz) KMA_RAND_11(sz) .name[KMALLOC_RANDOM_START + 12] = "kmalloc-rnd-12-" #sz,
798#define KMA_RAND_13(sz) KMA_RAND_12(sz) .name[KMALLOC_RANDOM_START + 13] = "kmalloc-rnd-13-" #sz,
799#define KMA_RAND_14(sz) KMA_RAND_13(sz) .name[KMALLOC_RANDOM_START + 14] = "kmalloc-rnd-14-" #sz,
800#define KMA_RAND_15(sz) KMA_RAND_14(sz) .name[KMALLOC_RANDOM_START + 15] = "kmalloc-rnd-15-" #sz,
801#else // CONFIG_RANDOM_KMALLOC_CACHES
802#define KMALLOC_RANDOM_NAME(N, sz)
803#endif
804
805#define INIT_KMALLOC_INFO(__size, __short_size) \
806{ \
807 .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \
808 KMALLOC_RCL_NAME(__short_size) \
809 KMALLOC_CGROUP_NAME(__short_size) \
810 KMALLOC_DMA_NAME(__short_size) \
811 KMALLOC_RANDOM_NAME(RANDOM_KMALLOC_CACHES_NR, __short_size) \
812 .size = __size, \
813}
814
815/*
816 * kmalloc_info[] is to make slab_debug=,kmalloc-xx option work at boot time.
817 * kmalloc_index() supports up to 2^21=2MB, so the final entry of the table is
818 * kmalloc-2M.
819 */
820const struct kmalloc_info_struct kmalloc_info[] __initconst = {
821 INIT_KMALLOC_INFO(0, 0),
822 INIT_KMALLOC_INFO(96, 96),
823 INIT_KMALLOC_INFO(192, 192),
824 INIT_KMALLOC_INFO(8, 8),
825 INIT_KMALLOC_INFO(16, 16),
826 INIT_KMALLOC_INFO(32, 32),
827 INIT_KMALLOC_INFO(64, 64),
828 INIT_KMALLOC_INFO(128, 128),
829 INIT_KMALLOC_INFO(256, 256),
830 INIT_KMALLOC_INFO(512, 512),
831 INIT_KMALLOC_INFO(1024, 1k),
832 INIT_KMALLOC_INFO(2048, 2k),
833 INIT_KMALLOC_INFO(4096, 4k),
834 INIT_KMALLOC_INFO(8192, 8k),
835 INIT_KMALLOC_INFO(16384, 16k),
836 INIT_KMALLOC_INFO(32768, 32k),
837 INIT_KMALLOC_INFO(65536, 64k),
838 INIT_KMALLOC_INFO(131072, 128k),
839 INIT_KMALLOC_INFO(262144, 256k),
840 INIT_KMALLOC_INFO(524288, 512k),
841 INIT_KMALLOC_INFO(1048576, 1M),
842 INIT_KMALLOC_INFO(2097152, 2M)
843};
844
845/*
846 * Patch up the size_index table if we have strange large alignment
847 * requirements for the kmalloc array. This is only the case for
848 * MIPS it seems. The standard arches will not generate any code here.
849 *
850 * Largest permitted alignment is 256 bytes due to the way we
851 * handle the index determination for the smaller caches.
852 *
853 * Make sure that nothing crazy happens if someone starts tinkering
854 * around with ARCH_KMALLOC_MINALIGN
855 */
856void __init setup_kmalloc_cache_index_table(void)
857{
858 unsigned int i;
859
860 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
861 !is_power_of_2(KMALLOC_MIN_SIZE));
862
863 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
864 unsigned int elem = size_index_elem(i);
865
866 if (elem >= ARRAY_SIZE(kmalloc_size_index))
867 break;
868 kmalloc_size_index[elem] = KMALLOC_SHIFT_LOW;
869 }
870
871 if (KMALLOC_MIN_SIZE >= 64) {
872 /*
873 * The 96 byte sized cache is not used if the alignment
874 * is 64 byte.
875 */
876 for (i = 64 + 8; i <= 96; i += 8)
877 kmalloc_size_index[size_index_elem(i)] = 7;
878
879 }
880
881 if (KMALLOC_MIN_SIZE >= 128) {
882 /*
883 * The 192 byte sized cache is not used if the alignment
884 * is 128 byte. Redirect kmalloc to use the 256 byte cache
885 * instead.
886 */
887 for (i = 128 + 8; i <= 192; i += 8)
888 kmalloc_size_index[size_index_elem(i)] = 8;
889 }
890}
891
892static unsigned int __kmalloc_minalign(void)
893{
894 unsigned int minalign = dma_get_cache_alignment();
895
896 if (IS_ENABLED(CONFIG_DMA_BOUNCE_UNALIGNED_KMALLOC) &&
897 is_swiotlb_allocated())
898 minalign = ARCH_KMALLOC_MINALIGN;
899
900 return max(minalign, arch_slab_minalign());
901}
902
903static void __init
904new_kmalloc_cache(int idx, enum kmalloc_cache_type type)
905{
906 slab_flags_t flags = 0;
907 unsigned int minalign = __kmalloc_minalign();
908 unsigned int aligned_size = kmalloc_info[idx].size;
909 int aligned_idx = idx;
910
911 if ((KMALLOC_RECLAIM != KMALLOC_NORMAL) && (type == KMALLOC_RECLAIM)) {
912 flags |= SLAB_RECLAIM_ACCOUNT;
913 } else if (IS_ENABLED(CONFIG_MEMCG) && (type == KMALLOC_CGROUP)) {
914 if (mem_cgroup_kmem_disabled()) {
915 kmalloc_caches[type][idx] = kmalloc_caches[KMALLOC_NORMAL][idx];
916 return;
917 }
918 flags |= SLAB_ACCOUNT;
919 } else if (IS_ENABLED(CONFIG_ZONE_DMA) && (type == KMALLOC_DMA)) {
920 flags |= SLAB_CACHE_DMA;
921 }
922
923#ifdef CONFIG_RANDOM_KMALLOC_CACHES
924 if (type >= KMALLOC_RANDOM_START && type <= KMALLOC_RANDOM_END)
925 flags |= SLAB_NO_MERGE;
926#endif
927
928 /*
929 * If CONFIG_MEMCG is enabled, disable cache merging for
930 * KMALLOC_NORMAL caches.
931 */
932 if (IS_ENABLED(CONFIG_MEMCG) && (type == KMALLOC_NORMAL))
933 flags |= SLAB_NO_MERGE;
934
935 if (minalign > ARCH_KMALLOC_MINALIGN) {
936 aligned_size = ALIGN(aligned_size, minalign);
937 aligned_idx = __kmalloc_index(aligned_size, false);
938 }
939
940 if (!kmalloc_caches[type][aligned_idx])
941 kmalloc_caches[type][aligned_idx] = create_kmalloc_cache(
942 kmalloc_info[aligned_idx].name[type],
943 aligned_size, flags);
944 if (idx != aligned_idx)
945 kmalloc_caches[type][idx] = kmalloc_caches[type][aligned_idx];
946}
947
948/*
949 * Create the kmalloc array. Some of the regular kmalloc arrays
950 * may already have been created because they were needed to
951 * enable allocations for slab creation.
952 */
953void __init create_kmalloc_caches(void)
954{
955 int i;
956 enum kmalloc_cache_type type;
957
958 /*
959 * Including KMALLOC_CGROUP if CONFIG_MEMCG defined
960 */
961 for (type = KMALLOC_NORMAL; type < NR_KMALLOC_TYPES; type++) {
962 /* Caches that are NOT of the two-to-the-power-of size. */
963 if (KMALLOC_MIN_SIZE <= 32)
964 new_kmalloc_cache(1, type);
965 if (KMALLOC_MIN_SIZE <= 64)
966 new_kmalloc_cache(2, type);
967
968 /* Caches that are of the two-to-the-power-of size. */
969 for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++)
970 new_kmalloc_cache(i, type);
971 }
972#ifdef CONFIG_RANDOM_KMALLOC_CACHES
973 random_kmalloc_seed = get_random_u64();
974#endif
975
976 /* Kmalloc array is now usable */
977 slab_state = UP;
978
979 if (IS_ENABLED(CONFIG_SLAB_BUCKETS))
980 kmem_buckets_cache = kmem_cache_create("kmalloc_buckets",
981 sizeof(kmem_buckets),
982 0, SLAB_NO_MERGE, NULL);
983}
984
985/**
986 * __ksize -- Report full size of underlying allocation
987 * @object: pointer to the object
988 *
989 * This should only be used internally to query the true size of allocations.
990 * It is not meant to be a way to discover the usable size of an allocation
991 * after the fact. Instead, use kmalloc_size_roundup(). Using memory beyond
992 * the originally requested allocation size may trigger KASAN, UBSAN_BOUNDS,
993 * and/or FORTIFY_SOURCE.
994 *
995 * Return: size of the actual memory used by @object in bytes
996 */
997size_t __ksize(const void *object)
998{
999 struct folio *folio;
1000
1001 if (unlikely(object == ZERO_SIZE_PTR))
1002 return 0;
1003
1004 folio = virt_to_folio(object);
1005
1006 if (unlikely(!folio_test_slab(folio))) {
1007 if (WARN_ON(folio_size(folio) <= KMALLOC_MAX_CACHE_SIZE))
1008 return 0;
1009 if (WARN_ON(object != folio_address(folio)))
1010 return 0;
1011 return folio_size(folio);
1012 }
1013
1014#ifdef CONFIG_SLUB_DEBUG
1015 skip_orig_size_check(folio_slab(folio)->slab_cache, object);
1016#endif
1017
1018 return slab_ksize(folio_slab(folio)->slab_cache);
1019}
1020
1021gfp_t kmalloc_fix_flags(gfp_t flags)
1022{
1023 gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
1024
1025 flags &= ~GFP_SLAB_BUG_MASK;
1026 pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
1027 invalid_mask, &invalid_mask, flags, &flags);
1028 dump_stack();
1029
1030 return flags;
1031}
1032
1033#ifdef CONFIG_SLAB_FREELIST_RANDOM
1034/* Randomize a generic freelist */
1035static void freelist_randomize(unsigned int *list,
1036 unsigned int count)
1037{
1038 unsigned int rand;
1039 unsigned int i;
1040
1041 for (i = 0; i < count; i++)
1042 list[i] = i;
1043
1044 /* Fisher-Yates shuffle */
1045 for (i = count - 1; i > 0; i--) {
1046 rand = get_random_u32_below(i + 1);
1047 swap(list[i], list[rand]);
1048 }
1049}
1050
1051/* Create a random sequence per cache */
1052int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
1053 gfp_t gfp)
1054{
1055
1056 if (count < 2 || cachep->random_seq)
1057 return 0;
1058
1059 cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp);
1060 if (!cachep->random_seq)
1061 return -ENOMEM;
1062
1063 freelist_randomize(cachep->random_seq, count);
1064 return 0;
1065}
1066
1067/* Destroy the per-cache random freelist sequence */
1068void cache_random_seq_destroy(struct kmem_cache *cachep)
1069{
1070 kfree(cachep->random_seq);
1071 cachep->random_seq = NULL;
1072}
1073#endif /* CONFIG_SLAB_FREELIST_RANDOM */
1074
1075#ifdef CONFIG_SLUB_DEBUG
1076#define SLABINFO_RIGHTS (0400)
1077
1078static void print_slabinfo_header(struct seq_file *m)
1079{
1080 /*
1081 * Output format version, so at least we can change it
1082 * without _too_ many complaints.
1083 */
1084 seq_puts(m, "slabinfo - version: 2.1\n");
1085 seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
1086 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
1087 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
1088 seq_putc(m, '\n');
1089}
1090
1091static void *slab_start(struct seq_file *m, loff_t *pos)
1092{
1093 mutex_lock(&slab_mutex);
1094 return seq_list_start(&slab_caches, *pos);
1095}
1096
1097static void *slab_next(struct seq_file *m, void *p, loff_t *pos)
1098{
1099 return seq_list_next(p, &slab_caches, pos);
1100}
1101
1102static void slab_stop(struct seq_file *m, void *p)
1103{
1104 mutex_unlock(&slab_mutex);
1105}
1106
1107static void cache_show(struct kmem_cache *s, struct seq_file *m)
1108{
1109 struct slabinfo sinfo;
1110
1111 memset(&sinfo, 0, sizeof(sinfo));
1112 get_slabinfo(s, &sinfo);
1113
1114 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
1115 s->name, sinfo.active_objs, sinfo.num_objs, s->size,
1116 sinfo.objects_per_slab, (1 << sinfo.cache_order));
1117
1118 seq_printf(m, " : tunables %4u %4u %4u",
1119 sinfo.limit, sinfo.batchcount, sinfo.shared);
1120 seq_printf(m, " : slabdata %6lu %6lu %6lu",
1121 sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
1122 seq_putc(m, '\n');
1123}
1124
1125static int slab_show(struct seq_file *m, void *p)
1126{
1127 struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
1128
1129 if (p == slab_caches.next)
1130 print_slabinfo_header(m);
1131 cache_show(s, m);
1132 return 0;
1133}
1134
1135void dump_unreclaimable_slab(void)
1136{
1137 struct kmem_cache *s;
1138 struct slabinfo sinfo;
1139
1140 /*
1141 * Here acquiring slab_mutex is risky since we don't prefer to get
1142 * sleep in oom path. But, without mutex hold, it may introduce a
1143 * risk of crash.
1144 * Use mutex_trylock to protect the list traverse, dump nothing
1145 * without acquiring the mutex.
1146 */
1147 if (!mutex_trylock(&slab_mutex)) {
1148 pr_warn("excessive unreclaimable slab but cannot dump stats\n");
1149 return;
1150 }
1151
1152 pr_info("Unreclaimable slab info:\n");
1153 pr_info("Name Used Total\n");
1154
1155 list_for_each_entry(s, &slab_caches, list) {
1156 if (s->flags & SLAB_RECLAIM_ACCOUNT)
1157 continue;
1158
1159 get_slabinfo(s, &sinfo);
1160
1161 if (sinfo.num_objs > 0)
1162 pr_info("%-17s %10luKB %10luKB\n", s->name,
1163 (sinfo.active_objs * s->size) / 1024,
1164 (sinfo.num_objs * s->size) / 1024);
1165 }
1166 mutex_unlock(&slab_mutex);
1167}
1168
1169/*
1170 * slabinfo_op - iterator that generates /proc/slabinfo
1171 *
1172 * Output layout:
1173 * cache-name
1174 * num-active-objs
1175 * total-objs
1176 * object size
1177 * num-active-slabs
1178 * total-slabs
1179 * num-pages-per-slab
1180 * + further values on SMP and with statistics enabled
1181 */
1182static const struct seq_operations slabinfo_op = {
1183 .start = slab_start,
1184 .next = slab_next,
1185 .stop = slab_stop,
1186 .show = slab_show,
1187};
1188
1189static int slabinfo_open(struct inode *inode, struct file *file)
1190{
1191 return seq_open(file, &slabinfo_op);
1192}
1193
1194static const struct proc_ops slabinfo_proc_ops = {
1195 .proc_flags = PROC_ENTRY_PERMANENT,
1196 .proc_open = slabinfo_open,
1197 .proc_read = seq_read,
1198 .proc_lseek = seq_lseek,
1199 .proc_release = seq_release,
1200};
1201
1202static int __init slab_proc_init(void)
1203{
1204 proc_create("slabinfo", SLABINFO_RIGHTS, NULL, &slabinfo_proc_ops);
1205 return 0;
1206}
1207module_init(slab_proc_init);
1208
1209#endif /* CONFIG_SLUB_DEBUG */
1210
1211/**
1212 * kfree_sensitive - Clear sensitive information in memory before freeing
1213 * @p: object to free memory of
1214 *
1215 * The memory of the object @p points to is zeroed before freed.
1216 * If @p is %NULL, kfree_sensitive() does nothing.
1217 *
1218 * Note: this function zeroes the whole allocated buffer which can be a good
1219 * deal bigger than the requested buffer size passed to kmalloc(). So be
1220 * careful when using this function in performance sensitive code.
1221 */
1222void kfree_sensitive(const void *p)
1223{
1224 size_t ks;
1225 void *mem = (void *)p;
1226
1227 ks = ksize(mem);
1228 if (ks) {
1229 kasan_unpoison_range(mem, ks);
1230 memzero_explicit(mem, ks);
1231 }
1232 kfree(mem);
1233}
1234EXPORT_SYMBOL(kfree_sensitive);
1235
1236size_t ksize(const void *objp)
1237{
1238 /*
1239 * We need to first check that the pointer to the object is valid.
1240 * The KASAN report printed from ksize() is more useful, then when
1241 * it's printed later when the behaviour could be undefined due to
1242 * a potential use-after-free or double-free.
1243 *
1244 * We use kasan_check_byte(), which is supported for the hardware
1245 * tag-based KASAN mode, unlike kasan_check_read/write().
1246 *
1247 * If the pointed to memory is invalid, we return 0 to avoid users of
1248 * ksize() writing to and potentially corrupting the memory region.
1249 *
1250 * We want to perform the check before __ksize(), to avoid potentially
1251 * crashing in __ksize() due to accessing invalid metadata.
1252 */
1253 if (unlikely(ZERO_OR_NULL_PTR(objp)) || !kasan_check_byte(objp))
1254 return 0;
1255
1256 return kfence_ksize(objp) ?: __ksize(objp);
1257}
1258EXPORT_SYMBOL(ksize);
1259
1260#ifdef CONFIG_BPF_SYSCALL
1261#include <linux/btf.h>
1262
1263__bpf_kfunc_start_defs();
1264
1265__bpf_kfunc struct kmem_cache *bpf_get_kmem_cache(u64 addr)
1266{
1267 struct slab *slab;
1268
1269 if (!virt_addr_valid((void *)(long)addr))
1270 return NULL;
1271
1272 slab = virt_to_slab((void *)(long)addr);
1273 return slab ? slab->slab_cache : NULL;
1274}
1275
1276__bpf_kfunc_end_defs();
1277#endif /* CONFIG_BPF_SYSCALL */
1278
1279/* Tracepoints definitions. */
1280EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1281EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1282EXPORT_TRACEPOINT_SYMBOL(kfree);
1283EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);
1284