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 "internal.h"
30
31#include "slab.h"
32
33enum slab_state slab_state;
34LIST_HEAD(slab_caches);
35DEFINE_MUTEX(slab_mutex);
36struct kmem_cache *kmem_cache;
37
38#ifdef CONFIG_HARDENED_USERCOPY
39bool usercopy_fallback __ro_after_init =
40 IS_ENABLED(CONFIG_HARDENED_USERCOPY_FALLBACK);
41module_param(usercopy_fallback, bool, 0400);
42MODULE_PARM_DESC(usercopy_fallback,
43 "WARN instead of reject usercopy whitelist violations");
44#endif
45
46static LIST_HEAD(slab_caches_to_rcu_destroy);
47static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work);
48static DECLARE_WORK(slab_caches_to_rcu_destroy_work,
49 slab_caches_to_rcu_destroy_workfn);
50
51/*
52 * Set of flags that will prevent slab merging
53 */
54#define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
55 SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \
56 SLAB_FAILSLAB | SLAB_KASAN)
57
58#define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \
59 SLAB_CACHE_DMA32 | SLAB_ACCOUNT)
60
61/*
62 * Merge control. If this is set then no merging of slab caches will occur.
63 */
64static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT);
65
66static int __init setup_slab_nomerge(char *str)
67{
68 slab_nomerge = true;
69 return 1;
70}
71
72#ifdef CONFIG_SLUB
73__setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0);
74#endif
75
76__setup("slab_nomerge", setup_slab_nomerge);
77
78/*
79 * Determine the size of a slab object
80 */
81unsigned int kmem_cache_size(struct kmem_cache *s)
82{
83 return s->object_size;
84}
85EXPORT_SYMBOL(kmem_cache_size);
86
87#ifdef CONFIG_DEBUG_VM
88static int kmem_cache_sanity_check(const char *name, unsigned int size)
89{
90 if (!name || in_interrupt() || size < sizeof(void *) ||
91 size > KMALLOC_MAX_SIZE) {
92 pr_err("kmem_cache_create(%s) integrity check failed\n", name);
93 return -EINVAL;
94 }
95
96 WARN_ON(strchr(name, ' ')); /* It confuses parsers */
97 return 0;
98}
99#else
100static inline int kmem_cache_sanity_check(const char *name, unsigned int size)
101{
102 return 0;
103}
104#endif
105
106void __kmem_cache_free_bulk(struct kmem_cache *s, size_t nr, void **p)
107{
108 size_t i;
109
110 for (i = 0; i < nr; i++) {
111 if (s)
112 kmem_cache_free(s, p[i]);
113 else
114 kfree(p[i]);
115 }
116}
117
118int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t nr,
119 void **p)
120{
121 size_t i;
122
123 for (i = 0; i < nr; i++) {
124 void *x = p[i] = kmem_cache_alloc(s, flags);
125 if (!x) {
126 __kmem_cache_free_bulk(s, i, p);
127 return 0;
128 }
129 }
130 return i;
131}
132
133/*
134 * Figure out what the alignment of the objects will be given a set of
135 * flags, a user specified alignment and the size of the objects.
136 */
137static unsigned int calculate_alignment(slab_flags_t flags,
138 unsigned int align, unsigned int size)
139{
140 /*
141 * If the user wants hardware cache aligned objects then follow that
142 * suggestion if the object is sufficiently large.
143 *
144 * The hardware cache alignment cannot override the specified
145 * alignment though. If that is greater then use it.
146 */
147 if (flags & SLAB_HWCACHE_ALIGN) {
148 unsigned int ralign;
149
150 ralign = cache_line_size();
151 while (size <= ralign / 2)
152 ralign /= 2;
153 align = max(align, ralign);
154 }
155
156 if (align < ARCH_SLAB_MINALIGN)
157 align = ARCH_SLAB_MINALIGN;
158
159 return ALIGN(align, sizeof(void *));
160}
161
162/*
163 * Find a mergeable slab cache
164 */
165int slab_unmergeable(struct kmem_cache *s)
166{
167 if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
168 return 1;
169
170 if (s->ctor)
171 return 1;
172
173 if (s->usersize)
174 return 1;
175
176 /*
177 * We may have set a slab to be unmergeable during bootstrap.
178 */
179 if (s->refcount < 0)
180 return 1;
181
182 return 0;
183}
184
185struct kmem_cache *find_mergeable(unsigned int size, unsigned int align,
186 slab_flags_t flags, const char *name, void (*ctor)(void *))
187{
188 struct kmem_cache *s;
189
190 if (slab_nomerge)
191 return NULL;
192
193 if (ctor)
194 return NULL;
195
196 size = ALIGN(size, sizeof(void *));
197 align = calculate_alignment(flags, align, size);
198 size = ALIGN(size, align);
199 flags = kmem_cache_flags(size, flags, name, NULL);
200
201 if (flags & SLAB_NEVER_MERGE)
202 return NULL;
203
204 list_for_each_entry_reverse(s, &slab_caches, list) {
205 if (slab_unmergeable(s))
206 continue;
207
208 if (size > s->size)
209 continue;
210
211 if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
212 continue;
213 /*
214 * Check if alignment is compatible.
215 * Courtesy of Adrian Drzewiecki
216 */
217 if ((s->size & ~(align - 1)) != s->size)
218 continue;
219
220 if (s->size - size >= sizeof(void *))
221 continue;
222
223 if (IS_ENABLED(CONFIG_SLAB) && align &&
224 (align > s->align || s->align % align))
225 continue;
226
227 return s;
228 }
229 return NULL;
230}
231
232static struct kmem_cache *create_cache(const char *name,
233 unsigned int object_size, unsigned int align,
234 slab_flags_t flags, unsigned int useroffset,
235 unsigned int usersize, void (*ctor)(void *),
236 struct kmem_cache *root_cache)
237{
238 struct kmem_cache *s;
239 int err;
240
241 if (WARN_ON(useroffset + usersize > object_size))
242 useroffset = usersize = 0;
243
244 err = -ENOMEM;
245 s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
246 if (!s)
247 goto out;
248
249 s->name = name;
250 s->size = s->object_size = object_size;
251 s->align = align;
252 s->ctor = ctor;
253 s->useroffset = useroffset;
254 s->usersize = usersize;
255
256 err = __kmem_cache_create(s, flags);
257 if (err)
258 goto out_free_cache;
259
260 s->refcount = 1;
261 list_add(&s->list, &slab_caches);
262out:
263 if (err)
264 return ERR_PTR(err);
265 return s;
266
267out_free_cache:
268 kmem_cache_free(kmem_cache, s);
269 goto out;
270}
271
272/**
273 * kmem_cache_create_usercopy - Create a cache with a region suitable
274 * for copying to userspace
275 * @name: A string which is used in /proc/slabinfo to identify this cache.
276 * @size: The size of objects to be created in this cache.
277 * @align: The required alignment for the objects.
278 * @flags: SLAB flags
279 * @useroffset: Usercopy region offset
280 * @usersize: Usercopy region size
281 * @ctor: A constructor for the objects.
282 *
283 * Cannot be called within a interrupt, but can be interrupted.
284 * The @ctor is run when new pages are allocated by the cache.
285 *
286 * The flags are
287 *
288 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
289 * to catch references to uninitialised memory.
290 *
291 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
292 * for buffer overruns.
293 *
294 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
295 * cacheline. This can be beneficial if you're counting cycles as closely
296 * as davem.
297 *
298 * Return: a pointer to the cache on success, NULL on failure.
299 */
300struct kmem_cache *
301kmem_cache_create_usercopy(const char *name,
302 unsigned int size, unsigned int align,
303 slab_flags_t flags,
304 unsigned int useroffset, unsigned int usersize,
305 void (*ctor)(void *))
306{
307 struct kmem_cache *s = NULL;
308 const char *cache_name;
309 int err;
310
311 get_online_cpus();
312 get_online_mems();
313
314 mutex_lock(&slab_mutex);
315
316 err = kmem_cache_sanity_check(name, size);
317 if (err) {
318 goto out_unlock;
319 }
320
321 /* Refuse requests with allocator specific flags */
322 if (flags & ~SLAB_FLAGS_PERMITTED) {
323 err = -EINVAL;
324 goto out_unlock;
325 }
326
327 /*
328 * Some allocators will constraint the set of valid flags to a subset
329 * of all flags. We expect them to define CACHE_CREATE_MASK in this
330 * case, and we'll just provide them with a sanitized version of the
331 * passed flags.
332 */
333 flags &= CACHE_CREATE_MASK;
334
335 /* Fail closed on bad usersize of useroffset values. */
336 if (WARN_ON(!usersize && useroffset) ||
337 WARN_ON(size < usersize || size - usersize < useroffset))
338 usersize = useroffset = 0;
339
340 if (!usersize)
341 s = __kmem_cache_alias(name, size, align, flags, ctor);
342 if (s)
343 goto out_unlock;
344
345 cache_name = kstrdup_const(name, GFP_KERNEL);
346 if (!cache_name) {
347 err = -ENOMEM;
348 goto out_unlock;
349 }
350
351 s = create_cache(cache_name, size,
352 calculate_alignment(flags, align, size),
353 flags, useroffset, usersize, ctor, NULL);
354 if (IS_ERR(s)) {
355 err = PTR_ERR(s);
356 kfree_const(cache_name);
357 }
358
359out_unlock:
360 mutex_unlock(&slab_mutex);
361
362 put_online_mems();
363 put_online_cpus();
364
365 if (err) {
366 if (flags & SLAB_PANIC)
367 panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n",
368 name, err);
369 else {
370 pr_warn("kmem_cache_create(%s) failed with error %d\n",
371 name, err);
372 dump_stack();
373 }
374 return NULL;
375 }
376 return s;
377}
378EXPORT_SYMBOL(kmem_cache_create_usercopy);
379
380/**
381 * kmem_cache_create - Create a cache.
382 * @name: A string which is used in /proc/slabinfo to identify this cache.
383 * @size: The size of objects to be created in this cache.
384 * @align: The required alignment for the objects.
385 * @flags: SLAB flags
386 * @ctor: A constructor for the objects.
387 *
388 * Cannot be called within a interrupt, but can be interrupted.
389 * The @ctor is run when new pages are allocated by the cache.
390 *
391 * The flags are
392 *
393 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
394 * to catch references to uninitialised memory.
395 *
396 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
397 * for buffer overruns.
398 *
399 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
400 * cacheline. This can be beneficial if you're counting cycles as closely
401 * as davem.
402 *
403 * Return: a pointer to the cache on success, NULL on failure.
404 */
405struct kmem_cache *
406kmem_cache_create(const char *name, unsigned int size, unsigned int align,
407 slab_flags_t flags, void (*ctor)(void *))
408{
409 return kmem_cache_create_usercopy(name, size, align, flags, 0, 0,
410 ctor);
411}
412EXPORT_SYMBOL(kmem_cache_create);
413
414static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work)
415{
416 LIST_HEAD(to_destroy);
417 struct kmem_cache *s, *s2;
418
419 /*
420 * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the
421 * @slab_caches_to_rcu_destroy list. The slab pages are freed
422 * through RCU and the associated kmem_cache are dereferenced
423 * while freeing the pages, so the kmem_caches should be freed only
424 * after the pending RCU operations are finished. As rcu_barrier()
425 * is a pretty slow operation, we batch all pending destructions
426 * asynchronously.
427 */
428 mutex_lock(&slab_mutex);
429 list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy);
430 mutex_unlock(&slab_mutex);
431
432 if (list_empty(&to_destroy))
433 return;
434
435 rcu_barrier();
436
437 list_for_each_entry_safe(s, s2, &to_destroy, list) {
438#ifdef SLAB_SUPPORTS_SYSFS
439 sysfs_slab_release(s);
440#else
441 slab_kmem_cache_release(s);
442#endif
443 }
444}
445
446static int shutdown_cache(struct kmem_cache *s)
447{
448 /* free asan quarantined objects */
449 kasan_cache_shutdown(s);
450
451 if (__kmem_cache_shutdown(s) != 0)
452 return -EBUSY;
453
454 list_del(&s->list);
455
456 if (s->flags & SLAB_TYPESAFE_BY_RCU) {
457#ifdef SLAB_SUPPORTS_SYSFS
458 sysfs_slab_unlink(s);
459#endif
460 list_add_tail(&s->list, &slab_caches_to_rcu_destroy);
461 schedule_work(&slab_caches_to_rcu_destroy_work);
462 } else {
463#ifdef SLAB_SUPPORTS_SYSFS
464 sysfs_slab_unlink(s);
465 sysfs_slab_release(s);
466#else
467 slab_kmem_cache_release(s);
468#endif
469 }
470
471 return 0;
472}
473
474void slab_kmem_cache_release(struct kmem_cache *s)
475{
476 __kmem_cache_release(s);
477 kfree_const(s->name);
478 kmem_cache_free(kmem_cache, s);
479}
480
481void kmem_cache_destroy(struct kmem_cache *s)
482{
483 int err;
484
485 if (unlikely(!s))
486 return;
487
488 get_online_cpus();
489 get_online_mems();
490
491 mutex_lock(&slab_mutex);
492
493 s->refcount--;
494 if (s->refcount)
495 goto out_unlock;
496
497 err = shutdown_cache(s);
498 if (err) {
499 pr_err("kmem_cache_destroy %s: Slab cache still has objects\n",
500 s->name);
501 dump_stack();
502 }
503out_unlock:
504 mutex_unlock(&slab_mutex);
505
506 put_online_mems();
507 put_online_cpus();
508}
509EXPORT_SYMBOL(kmem_cache_destroy);
510
511/**
512 * kmem_cache_shrink - Shrink a cache.
513 * @cachep: The cache to shrink.
514 *
515 * Releases as many slabs as possible for a cache.
516 * To help debugging, a zero exit status indicates all slabs were released.
517 *
518 * Return: %0 if all slabs were released, non-zero otherwise
519 */
520int kmem_cache_shrink(struct kmem_cache *cachep)
521{
522 int ret;
523
524 get_online_cpus();
525 get_online_mems();
526 kasan_cache_shrink(cachep);
527 ret = __kmem_cache_shrink(cachep);
528 put_online_mems();
529 put_online_cpus();
530 return ret;
531}
532EXPORT_SYMBOL(kmem_cache_shrink);
533
534bool slab_is_available(void)
535{
536 return slab_state >= UP;
537}
538
539#ifndef CONFIG_SLOB
540/* Create a cache during boot when no slab services are available yet */
541void __init create_boot_cache(struct kmem_cache *s, const char *name,
542 unsigned int size, slab_flags_t flags,
543 unsigned int useroffset, unsigned int usersize)
544{
545 int err;
546 unsigned int align = ARCH_KMALLOC_MINALIGN;
547
548 s->name = name;
549 s->size = s->object_size = size;
550
551 /*
552 * For power of two sizes, guarantee natural alignment for kmalloc
553 * caches, regardless of SL*B debugging options.
554 */
555 if (is_power_of_2(size))
556 align = max(align, size);
557 s->align = calculate_alignment(flags, align, size);
558
559 s->useroffset = useroffset;
560 s->usersize = usersize;
561
562 err = __kmem_cache_create(s, flags);
563
564 if (err)
565 panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n",
566 name, size, err);
567
568 s->refcount = -1; /* Exempt from merging for now */
569}
570
571struct kmem_cache *__init create_kmalloc_cache(const char *name,
572 unsigned int size, slab_flags_t flags,
573 unsigned int useroffset, unsigned int usersize)
574{
575 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
576
577 if (!s)
578 panic("Out of memory when creating slab %s\n", name);
579
580 create_boot_cache(s, name, size, flags, useroffset, usersize);
581 list_add(&s->list, &slab_caches);
582 s->refcount = 1;
583 return s;
584}
585
586struct kmem_cache *
587kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1] __ro_after_init =
588{ /* initialization for https://bugs.llvm.org/show_bug.cgi?id=42570 */ };
589EXPORT_SYMBOL(kmalloc_caches);
590
591/*
592 * Conversion table for small slabs sizes / 8 to the index in the
593 * kmalloc array. This is necessary for slabs < 192 since we have non power
594 * of two cache sizes there. The size of larger slabs can be determined using
595 * fls.
596 */
597static u8 size_index[24] __ro_after_init = {
598 3, /* 8 */
599 4, /* 16 */
600 5, /* 24 */
601 5, /* 32 */
602 6, /* 40 */
603 6, /* 48 */
604 6, /* 56 */
605 6, /* 64 */
606 1, /* 72 */
607 1, /* 80 */
608 1, /* 88 */
609 1, /* 96 */
610 7, /* 104 */
611 7, /* 112 */
612 7, /* 120 */
613 7, /* 128 */
614 2, /* 136 */
615 2, /* 144 */
616 2, /* 152 */
617 2, /* 160 */
618 2, /* 168 */
619 2, /* 176 */
620 2, /* 184 */
621 2 /* 192 */
622};
623
624static inline unsigned int size_index_elem(unsigned int bytes)
625{
626 return (bytes - 1) / 8;
627}
628
629/*
630 * Find the kmem_cache structure that serves a given size of
631 * allocation
632 */
633struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
634{
635 unsigned int index;
636
637 if (size <= 192) {
638 if (!size)
639 return ZERO_SIZE_PTR;
640
641 index = size_index[size_index_elem(size)];
642 } else {
643 if (WARN_ON_ONCE(size > KMALLOC_MAX_CACHE_SIZE))
644 return NULL;
645 index = fls(size - 1);
646 }
647
648 return kmalloc_caches[kmalloc_type(flags)][index];
649}
650
651#ifdef CONFIG_ZONE_DMA
652#define INIT_KMALLOC_INFO(__size, __short_size) \
653{ \
654 .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \
655 .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #__short_size, \
656 .name[KMALLOC_DMA] = "dma-kmalloc-" #__short_size, \
657 .size = __size, \
658}
659#else
660#define INIT_KMALLOC_INFO(__size, __short_size) \
661{ \
662 .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \
663 .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #__short_size, \
664 .size = __size, \
665}
666#endif
667
668/*
669 * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
670 * kmalloc_index() supports up to 2^26=64MB, so the final entry of the table is
671 * kmalloc-67108864.
672 */
673const struct kmalloc_info_struct kmalloc_info[] __initconst = {
674 INIT_KMALLOC_INFO(0, 0),
675 INIT_KMALLOC_INFO(96, 96),
676 INIT_KMALLOC_INFO(192, 192),
677 INIT_KMALLOC_INFO(8, 8),
678 INIT_KMALLOC_INFO(16, 16),
679 INIT_KMALLOC_INFO(32, 32),
680 INIT_KMALLOC_INFO(64, 64),
681 INIT_KMALLOC_INFO(128, 128),
682 INIT_KMALLOC_INFO(256, 256),
683 INIT_KMALLOC_INFO(512, 512),
684 INIT_KMALLOC_INFO(1024, 1k),
685 INIT_KMALLOC_INFO(2048, 2k),
686 INIT_KMALLOC_INFO(4096, 4k),
687 INIT_KMALLOC_INFO(8192, 8k),
688 INIT_KMALLOC_INFO(16384, 16k),
689 INIT_KMALLOC_INFO(32768, 32k),
690 INIT_KMALLOC_INFO(65536, 64k),
691 INIT_KMALLOC_INFO(131072, 128k),
692 INIT_KMALLOC_INFO(262144, 256k),
693 INIT_KMALLOC_INFO(524288, 512k),
694 INIT_KMALLOC_INFO(1048576, 1M),
695 INIT_KMALLOC_INFO(2097152, 2M),
696 INIT_KMALLOC_INFO(4194304, 4M),
697 INIT_KMALLOC_INFO(8388608, 8M),
698 INIT_KMALLOC_INFO(16777216, 16M),
699 INIT_KMALLOC_INFO(33554432, 32M),
700 INIT_KMALLOC_INFO(67108864, 64M)
701};
702
703/*
704 * Patch up the size_index table if we have strange large alignment
705 * requirements for the kmalloc array. This is only the case for
706 * MIPS it seems. The standard arches will not generate any code here.
707 *
708 * Largest permitted alignment is 256 bytes due to the way we
709 * handle the index determination for the smaller caches.
710 *
711 * Make sure that nothing crazy happens if someone starts tinkering
712 * around with ARCH_KMALLOC_MINALIGN
713 */
714void __init setup_kmalloc_cache_index_table(void)
715{
716 unsigned int i;
717
718 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
719 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
720
721 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
722 unsigned int elem = size_index_elem(i);
723
724 if (elem >= ARRAY_SIZE(size_index))
725 break;
726 size_index[elem] = KMALLOC_SHIFT_LOW;
727 }
728
729 if (KMALLOC_MIN_SIZE >= 64) {
730 /*
731 * The 96 byte size cache is not used if the alignment
732 * is 64 byte.
733 */
734 for (i = 64 + 8; i <= 96; i += 8)
735 size_index[size_index_elem(i)] = 7;
736
737 }
738
739 if (KMALLOC_MIN_SIZE >= 128) {
740 /*
741 * The 192 byte sized cache is not used if the alignment
742 * is 128 byte. Redirect kmalloc to use the 256 byte cache
743 * instead.
744 */
745 for (i = 128 + 8; i <= 192; i += 8)
746 size_index[size_index_elem(i)] = 8;
747 }
748}
749
750static void __init
751new_kmalloc_cache(int idx, enum kmalloc_cache_type type, slab_flags_t flags)
752{
753 if (type == KMALLOC_RECLAIM)
754 flags |= SLAB_RECLAIM_ACCOUNT;
755
756 kmalloc_caches[type][idx] = create_kmalloc_cache(
757 kmalloc_info[idx].name[type],
758 kmalloc_info[idx].size, flags, 0,
759 kmalloc_info[idx].size);
760}
761
762/*
763 * Create the kmalloc array. Some of the regular kmalloc arrays
764 * may already have been created because they were needed to
765 * enable allocations for slab creation.
766 */
767void __init create_kmalloc_caches(slab_flags_t flags)
768{
769 int i;
770 enum kmalloc_cache_type type;
771
772 for (type = KMALLOC_NORMAL; type <= KMALLOC_RECLAIM; type++) {
773 for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
774 if (!kmalloc_caches[type][i])
775 new_kmalloc_cache(i, type, flags);
776
777 /*
778 * Caches that are not of the two-to-the-power-of size.
779 * These have to be created immediately after the
780 * earlier power of two caches
781 */
782 if (KMALLOC_MIN_SIZE <= 32 && i == 6 &&
783 !kmalloc_caches[type][1])
784 new_kmalloc_cache(1, type, flags);
785 if (KMALLOC_MIN_SIZE <= 64 && i == 7 &&
786 !kmalloc_caches[type][2])
787 new_kmalloc_cache(2, type, flags);
788 }
789 }
790
791 /* Kmalloc array is now usable */
792 slab_state = UP;
793
794#ifdef CONFIG_ZONE_DMA
795 for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
796 struct kmem_cache *s = kmalloc_caches[KMALLOC_NORMAL][i];
797
798 if (s) {
799 kmalloc_caches[KMALLOC_DMA][i] = create_kmalloc_cache(
800 kmalloc_info[i].name[KMALLOC_DMA],
801 kmalloc_info[i].size,
802 SLAB_CACHE_DMA | flags, 0,
803 kmalloc_info[i].size);
804 }
805 }
806#endif
807}
808#endif /* !CONFIG_SLOB */
809
810gfp_t kmalloc_fix_flags(gfp_t flags)
811{
812 gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
813
814 flags &= ~GFP_SLAB_BUG_MASK;
815 pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
816 invalid_mask, &invalid_mask, flags, &flags);
817 dump_stack();
818
819 return flags;
820}
821
822/*
823 * To avoid unnecessary overhead, we pass through large allocation requests
824 * directly to the page allocator. We use __GFP_COMP, because we will need to
825 * know the allocation order to free the pages properly in kfree.
826 */
827void *kmalloc_order(size_t size, gfp_t flags, unsigned int order)
828{
829 void *ret = NULL;
830 struct page *page;
831
832 if (unlikely(flags & GFP_SLAB_BUG_MASK))
833 flags = kmalloc_fix_flags(flags);
834
835 flags |= __GFP_COMP;
836 page = alloc_pages(flags, order);
837 if (likely(page)) {
838 ret = page_address(page);
839 mod_node_page_state(page_pgdat(page), NR_SLAB_UNRECLAIMABLE_B,
840 PAGE_SIZE << order);
841 }
842 ret = kasan_kmalloc_large(ret, size, flags);
843 /* As ret might get tagged, call kmemleak hook after KASAN. */
844 kmemleak_alloc(ret, size, 1, flags);
845 return ret;
846}
847EXPORT_SYMBOL(kmalloc_order);
848
849#ifdef CONFIG_TRACING
850void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
851{
852 void *ret = kmalloc_order(size, flags, order);
853 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
854 return ret;
855}
856EXPORT_SYMBOL(kmalloc_order_trace);
857#endif
858
859#ifdef CONFIG_SLAB_FREELIST_RANDOM
860/* Randomize a generic freelist */
861static void freelist_randomize(struct rnd_state *state, unsigned int *list,
862 unsigned int count)
863{
864 unsigned int rand;
865 unsigned int i;
866
867 for (i = 0; i < count; i++)
868 list[i] = i;
869
870 /* Fisher-Yates shuffle */
871 for (i = count - 1; i > 0; i--) {
872 rand = prandom_u32_state(state);
873 rand %= (i + 1);
874 swap(list[i], list[rand]);
875 }
876}
877
878/* Create a random sequence per cache */
879int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
880 gfp_t gfp)
881{
882 struct rnd_state state;
883
884 if (count < 2 || cachep->random_seq)
885 return 0;
886
887 cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp);
888 if (!cachep->random_seq)
889 return -ENOMEM;
890
891 /* Get best entropy at this stage of boot */
892 prandom_seed_state(&state, get_random_long());
893
894 freelist_randomize(&state, cachep->random_seq, count);
895 return 0;
896}
897
898/* Destroy the per-cache random freelist sequence */
899void cache_random_seq_destroy(struct kmem_cache *cachep)
900{
901 kfree(cachep->random_seq);
902 cachep->random_seq = NULL;
903}
904#endif /* CONFIG_SLAB_FREELIST_RANDOM */
905
906#if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
907#ifdef CONFIG_SLAB
908#define SLABINFO_RIGHTS (0600)
909#else
910#define SLABINFO_RIGHTS (0400)
911#endif
912
913static void print_slabinfo_header(struct seq_file *m)
914{
915 /*
916 * Output format version, so at least we can change it
917 * without _too_ many complaints.
918 */
919#ifdef CONFIG_DEBUG_SLAB
920 seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
921#else
922 seq_puts(m, "slabinfo - version: 2.1\n");
923#endif
924 seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
925 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
926 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
927#ifdef CONFIG_DEBUG_SLAB
928 seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
929 seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
930#endif
931 seq_putc(m, '\n');
932}
933
934void *slab_start(struct seq_file *m, loff_t *pos)
935{
936 mutex_lock(&slab_mutex);
937 return seq_list_start(&slab_caches, *pos);
938}
939
940void *slab_next(struct seq_file *m, void *p, loff_t *pos)
941{
942 return seq_list_next(p, &slab_caches, pos);
943}
944
945void slab_stop(struct seq_file *m, void *p)
946{
947 mutex_unlock(&slab_mutex);
948}
949
950static void cache_show(struct kmem_cache *s, struct seq_file *m)
951{
952 struct slabinfo sinfo;
953
954 memset(&sinfo, 0, sizeof(sinfo));
955 get_slabinfo(s, &sinfo);
956
957 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
958 s->name, sinfo.active_objs, sinfo.num_objs, s->size,
959 sinfo.objects_per_slab, (1 << sinfo.cache_order));
960
961 seq_printf(m, " : tunables %4u %4u %4u",
962 sinfo.limit, sinfo.batchcount, sinfo.shared);
963 seq_printf(m, " : slabdata %6lu %6lu %6lu",
964 sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
965 slabinfo_show_stats(m, s);
966 seq_putc(m, '\n');
967}
968
969static int slab_show(struct seq_file *m, void *p)
970{
971 struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
972
973 if (p == slab_caches.next)
974 print_slabinfo_header(m);
975 cache_show(s, m);
976 return 0;
977}
978
979void dump_unreclaimable_slab(void)
980{
981 struct kmem_cache *s, *s2;
982 struct slabinfo sinfo;
983
984 /*
985 * Here acquiring slab_mutex is risky since we don't prefer to get
986 * sleep in oom path. But, without mutex hold, it may introduce a
987 * risk of crash.
988 * Use mutex_trylock to protect the list traverse, dump nothing
989 * without acquiring the mutex.
990 */
991 if (!mutex_trylock(&slab_mutex)) {
992 pr_warn("excessive unreclaimable slab but cannot dump stats\n");
993 return;
994 }
995
996 pr_info("Unreclaimable slab info:\n");
997 pr_info("Name Used Total\n");
998
999 list_for_each_entry_safe(s, s2, &slab_caches, list) {
1000 if (s->flags & SLAB_RECLAIM_ACCOUNT)
1001 continue;
1002
1003 get_slabinfo(s, &sinfo);
1004
1005 if (sinfo.num_objs > 0)
1006 pr_info("%-17s %10luKB %10luKB\n", s->name,
1007 (sinfo.active_objs * s->size) / 1024,
1008 (sinfo.num_objs * s->size) / 1024);
1009 }
1010 mutex_unlock(&slab_mutex);
1011}
1012
1013#if defined(CONFIG_MEMCG_KMEM)
1014int memcg_slab_show(struct seq_file *m, void *p)
1015{
1016 /*
1017 * Deprecated.
1018 * Please, take a look at tools/cgroup/slabinfo.py .
1019 */
1020 return 0;
1021}
1022#endif
1023
1024/*
1025 * slabinfo_op - iterator that generates /proc/slabinfo
1026 *
1027 * Output layout:
1028 * cache-name
1029 * num-active-objs
1030 * total-objs
1031 * object size
1032 * num-active-slabs
1033 * total-slabs
1034 * num-pages-per-slab
1035 * + further values on SMP and with statistics enabled
1036 */
1037static const struct seq_operations slabinfo_op = {
1038 .start = slab_start,
1039 .next = slab_next,
1040 .stop = slab_stop,
1041 .show = slab_show,
1042};
1043
1044static int slabinfo_open(struct inode *inode, struct file *file)
1045{
1046 return seq_open(file, &slabinfo_op);
1047}
1048
1049static const struct proc_ops slabinfo_proc_ops = {
1050 .proc_flags = PROC_ENTRY_PERMANENT,
1051 .proc_open = slabinfo_open,
1052 .proc_read = seq_read,
1053 .proc_write = slabinfo_write,
1054 .proc_lseek = seq_lseek,
1055 .proc_release = seq_release,
1056};
1057
1058static int __init slab_proc_init(void)
1059{
1060 proc_create("slabinfo", SLABINFO_RIGHTS, NULL, &slabinfo_proc_ops);
1061 return 0;
1062}
1063module_init(slab_proc_init);
1064
1065#endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */
1066
1067static __always_inline void *__do_krealloc(const void *p, size_t new_size,
1068 gfp_t flags)
1069{
1070 void *ret;
1071 size_t ks;
1072
1073 ks = ksize(p);
1074
1075 if (ks >= new_size) {
1076 p = kasan_krealloc((void *)p, new_size, flags);
1077 return (void *)p;
1078 }
1079
1080 ret = kmalloc_track_caller(new_size, flags);
1081 if (ret && p)
1082 memcpy(ret, p, ks);
1083
1084 return ret;
1085}
1086
1087/**
1088 * krealloc - reallocate memory. The contents will remain unchanged.
1089 * @p: object to reallocate memory for.
1090 * @new_size: how many bytes of memory are required.
1091 * @flags: the type of memory to allocate.
1092 *
1093 * The contents of the object pointed to are preserved up to the
1094 * lesser of the new and old sizes. If @p is %NULL, krealloc()
1095 * behaves exactly like kmalloc(). If @new_size is 0 and @p is not a
1096 * %NULL pointer, the object pointed to is freed.
1097 *
1098 * Return: pointer to the allocated memory or %NULL in case of error
1099 */
1100void *krealloc(const void *p, size_t new_size, gfp_t flags)
1101{
1102 void *ret;
1103
1104 if (unlikely(!new_size)) {
1105 kfree(p);
1106 return ZERO_SIZE_PTR;
1107 }
1108
1109 ret = __do_krealloc(p, new_size, flags);
1110 if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret))
1111 kfree(p);
1112
1113 return ret;
1114}
1115EXPORT_SYMBOL(krealloc);
1116
1117/**
1118 * kfree_sensitive - Clear sensitive information in memory before freeing
1119 * @p: object to free memory of
1120 *
1121 * The memory of the object @p points to is zeroed before freed.
1122 * If @p is %NULL, kfree_sensitive() does nothing.
1123 *
1124 * Note: this function zeroes the whole allocated buffer which can be a good
1125 * deal bigger than the requested buffer size passed to kmalloc(). So be
1126 * careful when using this function in performance sensitive code.
1127 */
1128void kfree_sensitive(const void *p)
1129{
1130 size_t ks;
1131 void *mem = (void *)p;
1132
1133 ks = ksize(mem);
1134 if (ks)
1135 memzero_explicit(mem, ks);
1136 kfree(mem);
1137}
1138EXPORT_SYMBOL(kfree_sensitive);
1139
1140/**
1141 * ksize - get the actual amount of memory allocated for a given object
1142 * @objp: Pointer to the object
1143 *
1144 * kmalloc may internally round up allocations and return more memory
1145 * than requested. ksize() can be used to determine the actual amount of
1146 * memory allocated. The caller may use this additional memory, even though
1147 * a smaller amount of memory was initially specified with the kmalloc call.
1148 * The caller must guarantee that objp points to a valid object previously
1149 * allocated with either kmalloc() or kmem_cache_alloc(). The object
1150 * must not be freed during the duration of the call.
1151 *
1152 * Return: size of the actual memory used by @objp in bytes
1153 */
1154size_t ksize(const void *objp)
1155{
1156 size_t size;
1157
1158 /*
1159 * We need to check that the pointed to object is valid, and only then
1160 * unpoison the shadow memory below. We use __kasan_check_read(), to
1161 * generate a more useful report at the time ksize() is called (rather
1162 * than later where behaviour is undefined due to potential
1163 * use-after-free or double-free).
1164 *
1165 * If the pointed to memory is invalid we return 0, to avoid users of
1166 * ksize() writing to and potentially corrupting the memory region.
1167 *
1168 * We want to perform the check before __ksize(), to avoid potentially
1169 * crashing in __ksize() due to accessing invalid metadata.
1170 */
1171 if (unlikely(ZERO_OR_NULL_PTR(objp)) || !__kasan_check_read(objp, 1))
1172 return 0;
1173
1174 size = __ksize(objp);
1175 /*
1176 * We assume that ksize callers could use whole allocated area,
1177 * so we need to unpoison this area.
1178 */
1179 kasan_unpoison_shadow(objp, size);
1180 return size;
1181}
1182EXPORT_SYMBOL(ksize);
1183
1184/* Tracepoints definitions. */
1185EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1186EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1187EXPORT_TRACEPOINT_SYMBOL(kmalloc_node);
1188EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node);
1189EXPORT_TRACEPOINT_SYMBOL(kfree);
1190EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);
1191
1192int should_failslab(struct kmem_cache *s, gfp_t gfpflags)
1193{
1194 if (__should_failslab(s, gfpflags))
1195 return -ENOMEM;
1196 return 0;
1197}
1198ALLOW_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/proc_fs.h>
21#include <linux/debugfs.h>
22#include <linux/kasan.h>
23#include <asm/cacheflush.h>
24#include <asm/tlbflush.h>
25#include <asm/page.h>
26#include <linux/memcontrol.h>
27#include <linux/stackdepot.h>
28
29#include "internal.h"
30#include "slab.h"
31
32#define CREATE_TRACE_POINTS
33#include <trace/events/kmem.h>
34
35enum slab_state slab_state;
36LIST_HEAD(slab_caches);
37DEFINE_MUTEX(slab_mutex);
38struct kmem_cache *kmem_cache;
39
40static LIST_HEAD(slab_caches_to_rcu_destroy);
41static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work);
42static DECLARE_WORK(slab_caches_to_rcu_destroy_work,
43 slab_caches_to_rcu_destroy_workfn);
44
45/*
46 * Set of flags that will prevent slab merging
47 */
48#define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
49 SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \
50 SLAB_FAILSLAB | kasan_never_merge())
51
52#define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \
53 SLAB_CACHE_DMA32 | SLAB_ACCOUNT)
54
55/*
56 * Merge control. If this is set then no merging of slab caches will occur.
57 */
58static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT);
59
60static int __init setup_slab_nomerge(char *str)
61{
62 slab_nomerge = true;
63 return 1;
64}
65
66static int __init setup_slab_merge(char *str)
67{
68 slab_nomerge = false;
69 return 1;
70}
71
72#ifdef CONFIG_SLUB
73__setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0);
74__setup_param("slub_merge", slub_merge, setup_slab_merge, 0);
75#endif
76
77__setup("slab_nomerge", setup_slab_nomerge);
78__setup("slab_merge", setup_slab_merge);
79
80/*
81 * Determine the size of a slab object
82 */
83unsigned int kmem_cache_size(struct kmem_cache *s)
84{
85 return s->object_size;
86}
87EXPORT_SYMBOL(kmem_cache_size);
88
89#ifdef CONFIG_DEBUG_VM
90static int kmem_cache_sanity_check(const char *name, unsigned int size)
91{
92 if (!name || in_interrupt() || size > KMALLOC_MAX_SIZE) {
93 pr_err("kmem_cache_create(%s) integrity check failed\n", name);
94 return -EINVAL;
95 }
96
97 WARN_ON(strchr(name, ' ')); /* It confuses parsers */
98 return 0;
99}
100#else
101static inline int kmem_cache_sanity_check(const char *name, unsigned int size)
102{
103 return 0;
104}
105#endif
106
107/*
108 * Figure out what the alignment of the objects will be given a set of
109 * flags, a user specified alignment and the size of the objects.
110 */
111static unsigned int calculate_alignment(slab_flags_t flags,
112 unsigned int align, unsigned int size)
113{
114 /*
115 * If the user wants hardware cache aligned objects then follow that
116 * suggestion if the object is sufficiently large.
117 *
118 * The hardware cache alignment cannot override the specified
119 * alignment though. If that is greater then use it.
120 */
121 if (flags & SLAB_HWCACHE_ALIGN) {
122 unsigned int ralign;
123
124 ralign = cache_line_size();
125 while (size <= ralign / 2)
126 ralign /= 2;
127 align = max(align, ralign);
128 }
129
130 align = max(align, arch_slab_minalign());
131
132 return ALIGN(align, sizeof(void *));
133}
134
135/*
136 * Find a mergeable slab cache
137 */
138int slab_unmergeable(struct kmem_cache *s)
139{
140 if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
141 return 1;
142
143 if (s->ctor)
144 return 1;
145
146#ifdef CONFIG_HARDENED_USERCOPY
147 if (s->usersize)
148 return 1;
149#endif
150
151 /*
152 * We may have set a slab to be unmergeable during bootstrap.
153 */
154 if (s->refcount < 0)
155 return 1;
156
157 return 0;
158}
159
160struct kmem_cache *find_mergeable(unsigned int size, unsigned int align,
161 slab_flags_t flags, const char *name, void (*ctor)(void *))
162{
163 struct kmem_cache *s;
164
165 if (slab_nomerge)
166 return NULL;
167
168 if (ctor)
169 return NULL;
170
171 size = ALIGN(size, sizeof(void *));
172 align = calculate_alignment(flags, align, size);
173 size = ALIGN(size, align);
174 flags = kmem_cache_flags(size, flags, name);
175
176 if (flags & SLAB_NEVER_MERGE)
177 return NULL;
178
179 list_for_each_entry_reverse(s, &slab_caches, list) {
180 if (slab_unmergeable(s))
181 continue;
182
183 if (size > s->size)
184 continue;
185
186 if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
187 continue;
188 /*
189 * Check if alignment is compatible.
190 * Courtesy of Adrian Drzewiecki
191 */
192 if ((s->size & ~(align - 1)) != s->size)
193 continue;
194
195 if (s->size - size >= sizeof(void *))
196 continue;
197
198 if (IS_ENABLED(CONFIG_SLAB) && align &&
199 (align > s->align || s->align % align))
200 continue;
201
202 return s;
203 }
204 return NULL;
205}
206
207static struct kmem_cache *create_cache(const char *name,
208 unsigned int object_size, unsigned int align,
209 slab_flags_t flags, unsigned int useroffset,
210 unsigned int usersize, void (*ctor)(void *),
211 struct kmem_cache *root_cache)
212{
213 struct kmem_cache *s;
214 int err;
215
216 if (WARN_ON(useroffset + usersize > object_size))
217 useroffset = usersize = 0;
218
219 err = -ENOMEM;
220 s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
221 if (!s)
222 goto out;
223
224 s->name = name;
225 s->size = s->object_size = object_size;
226 s->align = align;
227 s->ctor = ctor;
228#ifdef CONFIG_HARDENED_USERCOPY
229 s->useroffset = useroffset;
230 s->usersize = usersize;
231#endif
232
233 err = __kmem_cache_create(s, flags);
234 if (err)
235 goto out_free_cache;
236
237 s->refcount = 1;
238 list_add(&s->list, &slab_caches);
239out:
240 if (err)
241 return ERR_PTR(err);
242 return s;
243
244out_free_cache:
245 kmem_cache_free(kmem_cache, s);
246 goto out;
247}
248
249/**
250 * kmem_cache_create_usercopy - Create a cache with a region suitable
251 * for copying to userspace
252 * @name: A string which is used in /proc/slabinfo to identify this cache.
253 * @size: The size of objects to be created in this cache.
254 * @align: The required alignment for the objects.
255 * @flags: SLAB flags
256 * @useroffset: Usercopy region offset
257 * @usersize: Usercopy region size
258 * @ctor: A constructor for the objects.
259 *
260 * Cannot be called within a interrupt, but can be interrupted.
261 * The @ctor is run when new pages are allocated by the cache.
262 *
263 * The flags are
264 *
265 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
266 * to catch references to uninitialised memory.
267 *
268 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
269 * for buffer overruns.
270 *
271 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
272 * cacheline. This can be beneficial if you're counting cycles as closely
273 * as davem.
274 *
275 * Return: a pointer to the cache on success, NULL on failure.
276 */
277struct kmem_cache *
278kmem_cache_create_usercopy(const char *name,
279 unsigned int size, unsigned int align,
280 slab_flags_t flags,
281 unsigned int useroffset, unsigned int usersize,
282 void (*ctor)(void *))
283{
284 struct kmem_cache *s = NULL;
285 const char *cache_name;
286 int err;
287
288#ifdef CONFIG_SLUB_DEBUG
289 /*
290 * If no slub_debug was enabled globally, the static key is not yet
291 * enabled by setup_slub_debug(). Enable it if the cache is being
292 * created with any of the debugging flags passed explicitly.
293 * It's also possible that this is the first cache created with
294 * SLAB_STORE_USER and we should init stack_depot for it.
295 */
296 if (flags & SLAB_DEBUG_FLAGS)
297 static_branch_enable(&slub_debug_enabled);
298 if (flags & SLAB_STORE_USER)
299 stack_depot_init();
300#endif
301
302 mutex_lock(&slab_mutex);
303
304 err = kmem_cache_sanity_check(name, size);
305 if (err) {
306 goto out_unlock;
307 }
308
309 /* Refuse requests with allocator specific flags */
310 if (flags & ~SLAB_FLAGS_PERMITTED) {
311 err = -EINVAL;
312 goto out_unlock;
313 }
314
315 /*
316 * Some allocators will constraint the set of valid flags to a subset
317 * of all flags. We expect them to define CACHE_CREATE_MASK in this
318 * case, and we'll just provide them with a sanitized version of the
319 * passed flags.
320 */
321 flags &= CACHE_CREATE_MASK;
322
323 /* Fail closed on bad usersize of useroffset values. */
324 if (!IS_ENABLED(CONFIG_HARDENED_USERCOPY) ||
325 WARN_ON(!usersize && useroffset) ||
326 WARN_ON(size < usersize || size - usersize < useroffset))
327 usersize = useroffset = 0;
328
329 if (!usersize)
330 s = __kmem_cache_alias(name, size, align, flags, ctor);
331 if (s)
332 goto out_unlock;
333
334 cache_name = kstrdup_const(name, GFP_KERNEL);
335 if (!cache_name) {
336 err = -ENOMEM;
337 goto out_unlock;
338 }
339
340 s = create_cache(cache_name, size,
341 calculate_alignment(flags, align, size),
342 flags, useroffset, usersize, ctor, NULL);
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_usercopy);
365
366/**
367 * kmem_cache_create - Create a cache.
368 * @name: A string which is used in /proc/slabinfo to identify this cache.
369 * @size: The size of objects to be created in this cache.
370 * @align: The required alignment for the objects.
371 * @flags: SLAB flags
372 * @ctor: A constructor for the objects.
373 *
374 * Cannot be called within a interrupt, but can be interrupted.
375 * The @ctor is run when new pages are allocated by the cache.
376 *
377 * The flags are
378 *
379 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
380 * to catch references to uninitialised memory.
381 *
382 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
383 * for buffer overruns.
384 *
385 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
386 * cacheline. This can be beneficial if you're counting cycles as closely
387 * as davem.
388 *
389 * Return: a pointer to the cache on success, NULL on failure.
390 */
391struct kmem_cache *
392kmem_cache_create(const char *name, unsigned int size, unsigned int align,
393 slab_flags_t flags, void (*ctor)(void *))
394{
395 return kmem_cache_create_usercopy(name, size, align, flags, 0, 0,
396 ctor);
397}
398EXPORT_SYMBOL(kmem_cache_create);
399
400#ifdef SLAB_SUPPORTS_SYSFS
401/*
402 * For a given kmem_cache, kmem_cache_destroy() should only be called
403 * once or there will be a use-after-free problem. The actual deletion
404 * and release of the kobject does not need slab_mutex or cpu_hotplug_lock
405 * protection. So they are now done without holding those locks.
406 *
407 * Note that there will be a slight delay in the deletion of sysfs files
408 * if kmem_cache_release() is called indrectly from a work function.
409 */
410static void kmem_cache_release(struct kmem_cache *s)
411{
412 sysfs_slab_unlink(s);
413 sysfs_slab_release(s);
414}
415#else
416static void kmem_cache_release(struct kmem_cache *s)
417{
418 slab_kmem_cache_release(s);
419}
420#endif
421
422static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work)
423{
424 LIST_HEAD(to_destroy);
425 struct kmem_cache *s, *s2;
426
427 /*
428 * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the
429 * @slab_caches_to_rcu_destroy list. The slab pages are freed
430 * through RCU and the associated kmem_cache are dereferenced
431 * while freeing the pages, so the kmem_caches should be freed only
432 * after the pending RCU operations are finished. As rcu_barrier()
433 * is a pretty slow operation, we batch all pending destructions
434 * asynchronously.
435 */
436 mutex_lock(&slab_mutex);
437 list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy);
438 mutex_unlock(&slab_mutex);
439
440 if (list_empty(&to_destroy))
441 return;
442
443 rcu_barrier();
444
445 list_for_each_entry_safe(s, s2, &to_destroy, list) {
446 debugfs_slab_release(s);
447 kfence_shutdown_cache(s);
448 kmem_cache_release(s);
449 }
450}
451
452static int shutdown_cache(struct kmem_cache *s)
453{
454 /* free asan quarantined objects */
455 kasan_cache_shutdown(s);
456
457 if (__kmem_cache_shutdown(s) != 0)
458 return -EBUSY;
459
460 list_del(&s->list);
461
462 if (s->flags & SLAB_TYPESAFE_BY_RCU) {
463 list_add_tail(&s->list, &slab_caches_to_rcu_destroy);
464 schedule_work(&slab_caches_to_rcu_destroy_work);
465 } else {
466 kfence_shutdown_cache(s);
467 debugfs_slab_release(s);
468 }
469
470 return 0;
471}
472
473void slab_kmem_cache_release(struct kmem_cache *s)
474{
475 __kmem_cache_release(s);
476 kfree_const(s->name);
477 kmem_cache_free(kmem_cache, s);
478}
479
480void kmem_cache_destroy(struct kmem_cache *s)
481{
482 int refcnt;
483 bool rcu_set;
484
485 if (unlikely(!s) || !kasan_check_byte(s))
486 return;
487
488 cpus_read_lock();
489 mutex_lock(&slab_mutex);
490
491 rcu_set = s->flags & SLAB_TYPESAFE_BY_RCU;
492
493 refcnt = --s->refcount;
494 if (refcnt)
495 goto out_unlock;
496
497 WARN(shutdown_cache(s),
498 "%s %s: Slab cache still has objects when called from %pS",
499 __func__, s->name, (void *)_RET_IP_);
500out_unlock:
501 mutex_unlock(&slab_mutex);
502 cpus_read_unlock();
503 if (!refcnt && !rcu_set)
504 kmem_cache_release(s);
505}
506EXPORT_SYMBOL(kmem_cache_destroy);
507
508/**
509 * kmem_cache_shrink - Shrink a cache.
510 * @cachep: The cache to shrink.
511 *
512 * Releases as many slabs as possible for a cache.
513 * To help debugging, a zero exit status indicates all slabs were released.
514 *
515 * Return: %0 if all slabs were released, non-zero otherwise
516 */
517int kmem_cache_shrink(struct kmem_cache *cachep)
518{
519 kasan_cache_shrink(cachep);
520
521 return __kmem_cache_shrink(cachep);
522}
523EXPORT_SYMBOL(kmem_cache_shrink);
524
525bool slab_is_available(void)
526{
527 return slab_state >= UP;
528}
529
530#ifdef CONFIG_PRINTK
531/**
532 * kmem_valid_obj - does the pointer reference a valid slab object?
533 * @object: pointer to query.
534 *
535 * Return: %true if the pointer is to a not-yet-freed object from
536 * kmalloc() or kmem_cache_alloc(), either %true or %false if the pointer
537 * is to an already-freed object, and %false otherwise.
538 */
539bool kmem_valid_obj(void *object)
540{
541 struct folio *folio;
542
543 /* Some arches consider ZERO_SIZE_PTR to be a valid address. */
544 if (object < (void *)PAGE_SIZE || !virt_addr_valid(object))
545 return false;
546 folio = virt_to_folio(object);
547 return folio_test_slab(folio);
548}
549EXPORT_SYMBOL_GPL(kmem_valid_obj);
550
551static void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab)
552{
553 if (__kfence_obj_info(kpp, object, slab))
554 return;
555 __kmem_obj_info(kpp, object, slab);
556}
557
558/**
559 * kmem_dump_obj - Print available slab provenance information
560 * @object: slab object for which to find provenance information.
561 *
562 * This function uses pr_cont(), so that the caller is expected to have
563 * printed out whatever preamble is appropriate. The provenance information
564 * depends on the type of object and on how much debugging is enabled.
565 * For a slab-cache object, the fact that it is a slab object is printed,
566 * and, if available, the slab name, return address, and stack trace from
567 * the allocation and last free path of that object.
568 *
569 * This function will splat if passed a pointer to a non-slab object.
570 * If you are not sure what type of object you have, you should instead
571 * use mem_dump_obj().
572 */
573void kmem_dump_obj(void *object)
574{
575 char *cp = IS_ENABLED(CONFIG_MMU) ? "" : "/vmalloc";
576 int i;
577 struct slab *slab;
578 unsigned long ptroffset;
579 struct kmem_obj_info kp = { };
580
581 if (WARN_ON_ONCE(!virt_addr_valid(object)))
582 return;
583 slab = virt_to_slab(object);
584 if (WARN_ON_ONCE(!slab)) {
585 pr_cont(" non-slab memory.\n");
586 return;
587 }
588 kmem_obj_info(&kp, object, slab);
589 if (kp.kp_slab_cache)
590 pr_cont(" slab%s %s", cp, kp.kp_slab_cache->name);
591 else
592 pr_cont(" slab%s", cp);
593 if (is_kfence_address(object))
594 pr_cont(" (kfence)");
595 if (kp.kp_objp)
596 pr_cont(" start %px", kp.kp_objp);
597 if (kp.kp_data_offset)
598 pr_cont(" data offset %lu", kp.kp_data_offset);
599 if (kp.kp_objp) {
600 ptroffset = ((char *)object - (char *)kp.kp_objp) - kp.kp_data_offset;
601 pr_cont(" pointer offset %lu", ptroffset);
602 }
603 if (kp.kp_slab_cache && kp.kp_slab_cache->object_size)
604 pr_cont(" size %u", kp.kp_slab_cache->object_size);
605 if (kp.kp_ret)
606 pr_cont(" allocated at %pS\n", kp.kp_ret);
607 else
608 pr_cont("\n");
609 for (i = 0; i < ARRAY_SIZE(kp.kp_stack); i++) {
610 if (!kp.kp_stack[i])
611 break;
612 pr_info(" %pS\n", kp.kp_stack[i]);
613 }
614
615 if (kp.kp_free_stack[0])
616 pr_cont(" Free path:\n");
617
618 for (i = 0; i < ARRAY_SIZE(kp.kp_free_stack); i++) {
619 if (!kp.kp_free_stack[i])
620 break;
621 pr_info(" %pS\n", kp.kp_free_stack[i]);
622 }
623
624}
625EXPORT_SYMBOL_GPL(kmem_dump_obj);
626#endif
627
628#ifndef CONFIG_SLOB
629/* Create a cache during boot when no slab services are available yet */
630void __init create_boot_cache(struct kmem_cache *s, const char *name,
631 unsigned int size, slab_flags_t flags,
632 unsigned int useroffset, unsigned int usersize)
633{
634 int err;
635 unsigned int align = ARCH_KMALLOC_MINALIGN;
636
637 s->name = name;
638 s->size = s->object_size = size;
639
640 /*
641 * For power of two sizes, guarantee natural alignment for kmalloc
642 * caches, regardless of SL*B debugging options.
643 */
644 if (is_power_of_2(size))
645 align = max(align, size);
646 s->align = calculate_alignment(flags, align, size);
647
648#ifdef CONFIG_HARDENED_USERCOPY
649 s->useroffset = useroffset;
650 s->usersize = usersize;
651#endif
652
653 err = __kmem_cache_create(s, flags);
654
655 if (err)
656 panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n",
657 name, size, err);
658
659 s->refcount = -1; /* Exempt from merging for now */
660}
661
662struct kmem_cache *__init create_kmalloc_cache(const char *name,
663 unsigned int size, slab_flags_t flags,
664 unsigned int useroffset, unsigned int usersize)
665{
666 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
667
668 if (!s)
669 panic("Out of memory when creating slab %s\n", name);
670
671 create_boot_cache(s, name, size, flags | SLAB_KMALLOC, useroffset,
672 usersize);
673 kasan_cache_create_kmalloc(s);
674 list_add(&s->list, &slab_caches);
675 s->refcount = 1;
676 return s;
677}
678
679struct kmem_cache *
680kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1] __ro_after_init =
681{ /* initialization for https://bugs.llvm.org/show_bug.cgi?id=42570 */ };
682EXPORT_SYMBOL(kmalloc_caches);
683
684/*
685 * Conversion table for small slabs sizes / 8 to the index in the
686 * kmalloc array. This is necessary for slabs < 192 since we have non power
687 * of two cache sizes there. The size of larger slabs can be determined using
688 * fls.
689 */
690static u8 size_index[24] __ro_after_init = {
691 3, /* 8 */
692 4, /* 16 */
693 5, /* 24 */
694 5, /* 32 */
695 6, /* 40 */
696 6, /* 48 */
697 6, /* 56 */
698 6, /* 64 */
699 1, /* 72 */
700 1, /* 80 */
701 1, /* 88 */
702 1, /* 96 */
703 7, /* 104 */
704 7, /* 112 */
705 7, /* 120 */
706 7, /* 128 */
707 2, /* 136 */
708 2, /* 144 */
709 2, /* 152 */
710 2, /* 160 */
711 2, /* 168 */
712 2, /* 176 */
713 2, /* 184 */
714 2 /* 192 */
715};
716
717static inline unsigned int size_index_elem(unsigned int bytes)
718{
719 return (bytes - 1) / 8;
720}
721
722/*
723 * Find the kmem_cache structure that serves a given size of
724 * allocation
725 */
726struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
727{
728 unsigned int index;
729
730 if (size <= 192) {
731 if (!size)
732 return ZERO_SIZE_PTR;
733
734 index = size_index[size_index_elem(size)];
735 } else {
736 if (WARN_ON_ONCE(size > KMALLOC_MAX_CACHE_SIZE))
737 return NULL;
738 index = fls(size - 1);
739 }
740
741 return kmalloc_caches[kmalloc_type(flags)][index];
742}
743
744size_t kmalloc_size_roundup(size_t size)
745{
746 struct kmem_cache *c;
747
748 /* Short-circuit the 0 size case. */
749 if (unlikely(size == 0))
750 return 0;
751 /* Short-circuit saturated "too-large" case. */
752 if (unlikely(size == SIZE_MAX))
753 return SIZE_MAX;
754 /* Above the smaller buckets, size is a multiple of page size. */
755 if (size > KMALLOC_MAX_CACHE_SIZE)
756 return PAGE_SIZE << get_order(size);
757
758 /* The flags don't matter since size_index is common to all. */
759 c = kmalloc_slab(size, GFP_KERNEL);
760 return c ? c->object_size : 0;
761}
762EXPORT_SYMBOL(kmalloc_size_roundup);
763
764#ifdef CONFIG_ZONE_DMA
765#define KMALLOC_DMA_NAME(sz) .name[KMALLOC_DMA] = "dma-kmalloc-" #sz,
766#else
767#define KMALLOC_DMA_NAME(sz)
768#endif
769
770#ifdef CONFIG_MEMCG_KMEM
771#define KMALLOC_CGROUP_NAME(sz) .name[KMALLOC_CGROUP] = "kmalloc-cg-" #sz,
772#else
773#define KMALLOC_CGROUP_NAME(sz)
774#endif
775
776#ifndef CONFIG_SLUB_TINY
777#define KMALLOC_RCL_NAME(sz) .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #sz,
778#else
779#define KMALLOC_RCL_NAME(sz)
780#endif
781
782#define INIT_KMALLOC_INFO(__size, __short_size) \
783{ \
784 .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \
785 KMALLOC_RCL_NAME(__short_size) \
786 KMALLOC_CGROUP_NAME(__short_size) \
787 KMALLOC_DMA_NAME(__short_size) \
788 .size = __size, \
789}
790
791/*
792 * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
793 * kmalloc_index() supports up to 2^21=2MB, so the final entry of the table is
794 * kmalloc-2M.
795 */
796const struct kmalloc_info_struct kmalloc_info[] __initconst = {
797 INIT_KMALLOC_INFO(0, 0),
798 INIT_KMALLOC_INFO(96, 96),
799 INIT_KMALLOC_INFO(192, 192),
800 INIT_KMALLOC_INFO(8, 8),
801 INIT_KMALLOC_INFO(16, 16),
802 INIT_KMALLOC_INFO(32, 32),
803 INIT_KMALLOC_INFO(64, 64),
804 INIT_KMALLOC_INFO(128, 128),
805 INIT_KMALLOC_INFO(256, 256),
806 INIT_KMALLOC_INFO(512, 512),
807 INIT_KMALLOC_INFO(1024, 1k),
808 INIT_KMALLOC_INFO(2048, 2k),
809 INIT_KMALLOC_INFO(4096, 4k),
810 INIT_KMALLOC_INFO(8192, 8k),
811 INIT_KMALLOC_INFO(16384, 16k),
812 INIT_KMALLOC_INFO(32768, 32k),
813 INIT_KMALLOC_INFO(65536, 64k),
814 INIT_KMALLOC_INFO(131072, 128k),
815 INIT_KMALLOC_INFO(262144, 256k),
816 INIT_KMALLOC_INFO(524288, 512k),
817 INIT_KMALLOC_INFO(1048576, 1M),
818 INIT_KMALLOC_INFO(2097152, 2M)
819};
820
821/*
822 * Patch up the size_index table if we have strange large alignment
823 * requirements for the kmalloc array. This is only the case for
824 * MIPS it seems. The standard arches will not generate any code here.
825 *
826 * Largest permitted alignment is 256 bytes due to the way we
827 * handle the index determination for the smaller caches.
828 *
829 * Make sure that nothing crazy happens if someone starts tinkering
830 * around with ARCH_KMALLOC_MINALIGN
831 */
832void __init setup_kmalloc_cache_index_table(void)
833{
834 unsigned int i;
835
836 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
837 !is_power_of_2(KMALLOC_MIN_SIZE));
838
839 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
840 unsigned int elem = size_index_elem(i);
841
842 if (elem >= ARRAY_SIZE(size_index))
843 break;
844 size_index[elem] = KMALLOC_SHIFT_LOW;
845 }
846
847 if (KMALLOC_MIN_SIZE >= 64) {
848 /*
849 * The 96 byte sized cache is not used if the alignment
850 * is 64 byte.
851 */
852 for (i = 64 + 8; i <= 96; i += 8)
853 size_index[size_index_elem(i)] = 7;
854
855 }
856
857 if (KMALLOC_MIN_SIZE >= 128) {
858 /*
859 * The 192 byte sized cache is not used if the alignment
860 * is 128 byte. Redirect kmalloc to use the 256 byte cache
861 * instead.
862 */
863 for (i = 128 + 8; i <= 192; i += 8)
864 size_index[size_index_elem(i)] = 8;
865 }
866}
867
868static void __init
869new_kmalloc_cache(int idx, enum kmalloc_cache_type type, slab_flags_t flags)
870{
871 if ((KMALLOC_RECLAIM != KMALLOC_NORMAL) && (type == KMALLOC_RECLAIM)) {
872 flags |= SLAB_RECLAIM_ACCOUNT;
873 } else if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_CGROUP)) {
874 if (mem_cgroup_kmem_disabled()) {
875 kmalloc_caches[type][idx] = kmalloc_caches[KMALLOC_NORMAL][idx];
876 return;
877 }
878 flags |= SLAB_ACCOUNT;
879 } else if (IS_ENABLED(CONFIG_ZONE_DMA) && (type == KMALLOC_DMA)) {
880 flags |= SLAB_CACHE_DMA;
881 }
882
883 kmalloc_caches[type][idx] = create_kmalloc_cache(
884 kmalloc_info[idx].name[type],
885 kmalloc_info[idx].size, flags, 0,
886 kmalloc_info[idx].size);
887
888 /*
889 * If CONFIG_MEMCG_KMEM is enabled, disable cache merging for
890 * KMALLOC_NORMAL caches.
891 */
892 if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_NORMAL))
893 kmalloc_caches[type][idx]->refcount = -1;
894}
895
896/*
897 * Create the kmalloc array. Some of the regular kmalloc arrays
898 * may already have been created because they were needed to
899 * enable allocations for slab creation.
900 */
901void __init create_kmalloc_caches(slab_flags_t flags)
902{
903 int i;
904 enum kmalloc_cache_type type;
905
906 /*
907 * Including KMALLOC_CGROUP if CONFIG_MEMCG_KMEM defined
908 */
909 for (type = KMALLOC_NORMAL; type < NR_KMALLOC_TYPES; type++) {
910 for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
911 if (!kmalloc_caches[type][i])
912 new_kmalloc_cache(i, type, flags);
913
914 /*
915 * Caches that are not of the two-to-the-power-of size.
916 * These have to be created immediately after the
917 * earlier power of two caches
918 */
919 if (KMALLOC_MIN_SIZE <= 32 && i == 6 &&
920 !kmalloc_caches[type][1])
921 new_kmalloc_cache(1, type, flags);
922 if (KMALLOC_MIN_SIZE <= 64 && i == 7 &&
923 !kmalloc_caches[type][2])
924 new_kmalloc_cache(2, type, flags);
925 }
926 }
927
928 /* Kmalloc array is now usable */
929 slab_state = UP;
930}
931
932void free_large_kmalloc(struct folio *folio, void *object)
933{
934 unsigned int order = folio_order(folio);
935
936 if (WARN_ON_ONCE(order == 0))
937 pr_warn_once("object pointer: 0x%p\n", object);
938
939 kmemleak_free(object);
940 kasan_kfree_large(object);
941 kmsan_kfree_large(object);
942
943 mod_lruvec_page_state(folio_page(folio, 0), NR_SLAB_UNRECLAIMABLE_B,
944 -(PAGE_SIZE << order));
945 __free_pages(folio_page(folio, 0), order);
946}
947
948static void *__kmalloc_large_node(size_t size, gfp_t flags, int node);
949static __always_inline
950void *__do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller)
951{
952 struct kmem_cache *s;
953 void *ret;
954
955 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
956 ret = __kmalloc_large_node(size, flags, node);
957 trace_kmalloc(caller, ret, size,
958 PAGE_SIZE << get_order(size), flags, node);
959 return ret;
960 }
961
962 s = kmalloc_slab(size, flags);
963
964 if (unlikely(ZERO_OR_NULL_PTR(s)))
965 return s;
966
967 ret = __kmem_cache_alloc_node(s, flags, node, size, caller);
968 ret = kasan_kmalloc(s, ret, size, flags);
969 trace_kmalloc(caller, ret, size, s->size, flags, node);
970 return ret;
971}
972
973void *__kmalloc_node(size_t size, gfp_t flags, int node)
974{
975 return __do_kmalloc_node(size, flags, node, _RET_IP_);
976}
977EXPORT_SYMBOL(__kmalloc_node);
978
979void *__kmalloc(size_t size, gfp_t flags)
980{
981 return __do_kmalloc_node(size, flags, NUMA_NO_NODE, _RET_IP_);
982}
983EXPORT_SYMBOL(__kmalloc);
984
985void *__kmalloc_node_track_caller(size_t size, gfp_t flags,
986 int node, unsigned long caller)
987{
988 return __do_kmalloc_node(size, flags, node, caller);
989}
990EXPORT_SYMBOL(__kmalloc_node_track_caller);
991
992/**
993 * kfree - free previously allocated memory
994 * @object: pointer returned by kmalloc.
995 *
996 * If @object is NULL, no operation is performed.
997 *
998 * Don't free memory not originally allocated by kmalloc()
999 * or you will run into trouble.
1000 */
1001void kfree(const void *object)
1002{
1003 struct folio *folio;
1004 struct slab *slab;
1005 struct kmem_cache *s;
1006
1007 trace_kfree(_RET_IP_, object);
1008
1009 if (unlikely(ZERO_OR_NULL_PTR(object)))
1010 return;
1011
1012 folio = virt_to_folio(object);
1013 if (unlikely(!folio_test_slab(folio))) {
1014 free_large_kmalloc(folio, (void *)object);
1015 return;
1016 }
1017
1018 slab = folio_slab(folio);
1019 s = slab->slab_cache;
1020 __kmem_cache_free(s, (void *)object, _RET_IP_);
1021}
1022EXPORT_SYMBOL(kfree);
1023
1024/**
1025 * __ksize -- Report full size of underlying allocation
1026 * @object: pointer to the object
1027 *
1028 * This should only be used internally to query the true size of allocations.
1029 * It is not meant to be a way to discover the usable size of an allocation
1030 * after the fact. Instead, use kmalloc_size_roundup(). Using memory beyond
1031 * the originally requested allocation size may trigger KASAN, UBSAN_BOUNDS,
1032 * and/or FORTIFY_SOURCE.
1033 *
1034 * Return: size of the actual memory used by @object in bytes
1035 */
1036size_t __ksize(const void *object)
1037{
1038 struct folio *folio;
1039
1040 if (unlikely(object == ZERO_SIZE_PTR))
1041 return 0;
1042
1043 folio = virt_to_folio(object);
1044
1045 if (unlikely(!folio_test_slab(folio))) {
1046 if (WARN_ON(folio_size(folio) <= KMALLOC_MAX_CACHE_SIZE))
1047 return 0;
1048 if (WARN_ON(object != folio_address(folio)))
1049 return 0;
1050 return folio_size(folio);
1051 }
1052
1053#ifdef CONFIG_SLUB_DEBUG
1054 skip_orig_size_check(folio_slab(folio)->slab_cache, object);
1055#endif
1056
1057 return slab_ksize(folio_slab(folio)->slab_cache);
1058}
1059
1060void *kmalloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
1061{
1062 void *ret = __kmem_cache_alloc_node(s, gfpflags, NUMA_NO_NODE,
1063 size, _RET_IP_);
1064
1065 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags, NUMA_NO_NODE);
1066
1067 ret = kasan_kmalloc(s, ret, size, gfpflags);
1068 return ret;
1069}
1070EXPORT_SYMBOL(kmalloc_trace);
1071
1072void *kmalloc_node_trace(struct kmem_cache *s, gfp_t gfpflags,
1073 int node, size_t size)
1074{
1075 void *ret = __kmem_cache_alloc_node(s, gfpflags, node, size, _RET_IP_);
1076
1077 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags, node);
1078
1079 ret = kasan_kmalloc(s, ret, size, gfpflags);
1080 return ret;
1081}
1082EXPORT_SYMBOL(kmalloc_node_trace);
1083#endif /* !CONFIG_SLOB */
1084
1085gfp_t kmalloc_fix_flags(gfp_t flags)
1086{
1087 gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
1088
1089 flags &= ~GFP_SLAB_BUG_MASK;
1090 pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
1091 invalid_mask, &invalid_mask, flags, &flags);
1092 dump_stack();
1093
1094 return flags;
1095}
1096
1097/*
1098 * To avoid unnecessary overhead, we pass through large allocation requests
1099 * directly to the page allocator. We use __GFP_COMP, because we will need to
1100 * know the allocation order to free the pages properly in kfree.
1101 */
1102
1103static void *__kmalloc_large_node(size_t size, gfp_t flags, int node)
1104{
1105 struct page *page;
1106 void *ptr = NULL;
1107 unsigned int order = get_order(size);
1108
1109 if (unlikely(flags & GFP_SLAB_BUG_MASK))
1110 flags = kmalloc_fix_flags(flags);
1111
1112 flags |= __GFP_COMP;
1113 page = alloc_pages_node(node, flags, order);
1114 if (page) {
1115 ptr = page_address(page);
1116 mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B,
1117 PAGE_SIZE << order);
1118 }
1119
1120 ptr = kasan_kmalloc_large(ptr, size, flags);
1121 /* As ptr might get tagged, call kmemleak hook after KASAN. */
1122 kmemleak_alloc(ptr, size, 1, flags);
1123 kmsan_kmalloc_large(ptr, size, flags);
1124
1125 return ptr;
1126}
1127
1128void *kmalloc_large(size_t size, gfp_t flags)
1129{
1130 void *ret = __kmalloc_large_node(size, flags, NUMA_NO_NODE);
1131
1132 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << get_order(size),
1133 flags, NUMA_NO_NODE);
1134 return ret;
1135}
1136EXPORT_SYMBOL(kmalloc_large);
1137
1138void *kmalloc_large_node(size_t size, gfp_t flags, int node)
1139{
1140 void *ret = __kmalloc_large_node(size, flags, node);
1141
1142 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << get_order(size),
1143 flags, node);
1144 return ret;
1145}
1146EXPORT_SYMBOL(kmalloc_large_node);
1147
1148#ifdef CONFIG_SLAB_FREELIST_RANDOM
1149/* Randomize a generic freelist */
1150static void freelist_randomize(struct rnd_state *state, unsigned int *list,
1151 unsigned int count)
1152{
1153 unsigned int rand;
1154 unsigned int i;
1155
1156 for (i = 0; i < count; i++)
1157 list[i] = i;
1158
1159 /* Fisher-Yates shuffle */
1160 for (i = count - 1; i > 0; i--) {
1161 rand = prandom_u32_state(state);
1162 rand %= (i + 1);
1163 swap(list[i], list[rand]);
1164 }
1165}
1166
1167/* Create a random sequence per cache */
1168int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
1169 gfp_t gfp)
1170{
1171 struct rnd_state state;
1172
1173 if (count < 2 || cachep->random_seq)
1174 return 0;
1175
1176 cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp);
1177 if (!cachep->random_seq)
1178 return -ENOMEM;
1179
1180 /* Get best entropy at this stage of boot */
1181 prandom_seed_state(&state, get_random_long());
1182
1183 freelist_randomize(&state, cachep->random_seq, count);
1184 return 0;
1185}
1186
1187/* Destroy the per-cache random freelist sequence */
1188void cache_random_seq_destroy(struct kmem_cache *cachep)
1189{
1190 kfree(cachep->random_seq);
1191 cachep->random_seq = NULL;
1192}
1193#endif /* CONFIG_SLAB_FREELIST_RANDOM */
1194
1195#if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
1196#ifdef CONFIG_SLAB
1197#define SLABINFO_RIGHTS (0600)
1198#else
1199#define SLABINFO_RIGHTS (0400)
1200#endif
1201
1202static void print_slabinfo_header(struct seq_file *m)
1203{
1204 /*
1205 * Output format version, so at least we can change it
1206 * without _too_ many complaints.
1207 */
1208#ifdef CONFIG_DEBUG_SLAB
1209 seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
1210#else
1211 seq_puts(m, "slabinfo - version: 2.1\n");
1212#endif
1213 seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
1214 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
1215 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
1216#ifdef CONFIG_DEBUG_SLAB
1217 seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
1218 seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
1219#endif
1220 seq_putc(m, '\n');
1221}
1222
1223static void *slab_start(struct seq_file *m, loff_t *pos)
1224{
1225 mutex_lock(&slab_mutex);
1226 return seq_list_start(&slab_caches, *pos);
1227}
1228
1229static void *slab_next(struct seq_file *m, void *p, loff_t *pos)
1230{
1231 return seq_list_next(p, &slab_caches, pos);
1232}
1233
1234static void slab_stop(struct seq_file *m, void *p)
1235{
1236 mutex_unlock(&slab_mutex);
1237}
1238
1239static void cache_show(struct kmem_cache *s, struct seq_file *m)
1240{
1241 struct slabinfo sinfo;
1242
1243 memset(&sinfo, 0, sizeof(sinfo));
1244 get_slabinfo(s, &sinfo);
1245
1246 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
1247 s->name, sinfo.active_objs, sinfo.num_objs, s->size,
1248 sinfo.objects_per_slab, (1 << sinfo.cache_order));
1249
1250 seq_printf(m, " : tunables %4u %4u %4u",
1251 sinfo.limit, sinfo.batchcount, sinfo.shared);
1252 seq_printf(m, " : slabdata %6lu %6lu %6lu",
1253 sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
1254 slabinfo_show_stats(m, s);
1255 seq_putc(m, '\n');
1256}
1257
1258static int slab_show(struct seq_file *m, void *p)
1259{
1260 struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
1261
1262 if (p == slab_caches.next)
1263 print_slabinfo_header(m);
1264 cache_show(s, m);
1265 return 0;
1266}
1267
1268void dump_unreclaimable_slab(void)
1269{
1270 struct kmem_cache *s;
1271 struct slabinfo sinfo;
1272
1273 /*
1274 * Here acquiring slab_mutex is risky since we don't prefer to get
1275 * sleep in oom path. But, without mutex hold, it may introduce a
1276 * risk of crash.
1277 * Use mutex_trylock to protect the list traverse, dump nothing
1278 * without acquiring the mutex.
1279 */
1280 if (!mutex_trylock(&slab_mutex)) {
1281 pr_warn("excessive unreclaimable slab but cannot dump stats\n");
1282 return;
1283 }
1284
1285 pr_info("Unreclaimable slab info:\n");
1286 pr_info("Name Used Total\n");
1287
1288 list_for_each_entry(s, &slab_caches, list) {
1289 if (s->flags & SLAB_RECLAIM_ACCOUNT)
1290 continue;
1291
1292 get_slabinfo(s, &sinfo);
1293
1294 if (sinfo.num_objs > 0)
1295 pr_info("%-17s %10luKB %10luKB\n", s->name,
1296 (sinfo.active_objs * s->size) / 1024,
1297 (sinfo.num_objs * s->size) / 1024);
1298 }
1299 mutex_unlock(&slab_mutex);
1300}
1301
1302/*
1303 * slabinfo_op - iterator that generates /proc/slabinfo
1304 *
1305 * Output layout:
1306 * cache-name
1307 * num-active-objs
1308 * total-objs
1309 * object size
1310 * num-active-slabs
1311 * total-slabs
1312 * num-pages-per-slab
1313 * + further values on SMP and with statistics enabled
1314 */
1315static const struct seq_operations slabinfo_op = {
1316 .start = slab_start,
1317 .next = slab_next,
1318 .stop = slab_stop,
1319 .show = slab_show,
1320};
1321
1322static int slabinfo_open(struct inode *inode, struct file *file)
1323{
1324 return seq_open(file, &slabinfo_op);
1325}
1326
1327static const struct proc_ops slabinfo_proc_ops = {
1328 .proc_flags = PROC_ENTRY_PERMANENT,
1329 .proc_open = slabinfo_open,
1330 .proc_read = seq_read,
1331 .proc_write = slabinfo_write,
1332 .proc_lseek = seq_lseek,
1333 .proc_release = seq_release,
1334};
1335
1336static int __init slab_proc_init(void)
1337{
1338 proc_create("slabinfo", SLABINFO_RIGHTS, NULL, &slabinfo_proc_ops);
1339 return 0;
1340}
1341module_init(slab_proc_init);
1342
1343#endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */
1344
1345static __always_inline __realloc_size(2) void *
1346__do_krealloc(const void *p, size_t new_size, gfp_t flags)
1347{
1348 void *ret;
1349 size_t ks;
1350
1351 /* Check for double-free before calling ksize. */
1352 if (likely(!ZERO_OR_NULL_PTR(p))) {
1353 if (!kasan_check_byte(p))
1354 return NULL;
1355 ks = ksize(p);
1356 } else
1357 ks = 0;
1358
1359 /* If the object still fits, repoison it precisely. */
1360 if (ks >= new_size) {
1361 p = kasan_krealloc((void *)p, new_size, flags);
1362 return (void *)p;
1363 }
1364
1365 ret = kmalloc_track_caller(new_size, flags);
1366 if (ret && p) {
1367 /* Disable KASAN checks as the object's redzone is accessed. */
1368 kasan_disable_current();
1369 memcpy(ret, kasan_reset_tag(p), ks);
1370 kasan_enable_current();
1371 }
1372
1373 return ret;
1374}
1375
1376/**
1377 * krealloc - reallocate memory. The contents will remain unchanged.
1378 * @p: object to reallocate memory for.
1379 * @new_size: how many bytes of memory are required.
1380 * @flags: the type of memory to allocate.
1381 *
1382 * The contents of the object pointed to are preserved up to the
1383 * lesser of the new and old sizes (__GFP_ZERO flag is effectively ignored).
1384 * If @p is %NULL, krealloc() behaves exactly like kmalloc(). If @new_size
1385 * is 0 and @p is not a %NULL pointer, the object pointed to is freed.
1386 *
1387 * Return: pointer to the allocated memory or %NULL in case of error
1388 */
1389void *krealloc(const void *p, size_t new_size, gfp_t flags)
1390{
1391 void *ret;
1392
1393 if (unlikely(!new_size)) {
1394 kfree(p);
1395 return ZERO_SIZE_PTR;
1396 }
1397
1398 ret = __do_krealloc(p, new_size, flags);
1399 if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret))
1400 kfree(p);
1401
1402 return ret;
1403}
1404EXPORT_SYMBOL(krealloc);
1405
1406/**
1407 * kfree_sensitive - Clear sensitive information in memory before freeing
1408 * @p: object to free memory of
1409 *
1410 * The memory of the object @p points to is zeroed before freed.
1411 * If @p is %NULL, kfree_sensitive() does nothing.
1412 *
1413 * Note: this function zeroes the whole allocated buffer which can be a good
1414 * deal bigger than the requested buffer size passed to kmalloc(). So be
1415 * careful when using this function in performance sensitive code.
1416 */
1417void kfree_sensitive(const void *p)
1418{
1419 size_t ks;
1420 void *mem = (void *)p;
1421
1422 ks = ksize(mem);
1423 if (ks) {
1424 kasan_unpoison_range(mem, ks);
1425 memzero_explicit(mem, ks);
1426 }
1427 kfree(mem);
1428}
1429EXPORT_SYMBOL(kfree_sensitive);
1430
1431size_t ksize(const void *objp)
1432{
1433 /*
1434 * We need to first check that the pointer to the object is valid.
1435 * The KASAN report printed from ksize() is more useful, then when
1436 * it's printed later when the behaviour could be undefined due to
1437 * a potential use-after-free or double-free.
1438 *
1439 * We use kasan_check_byte(), which is supported for the hardware
1440 * tag-based KASAN mode, unlike kasan_check_read/write().
1441 *
1442 * If the pointed to memory is invalid, we return 0 to avoid users of
1443 * ksize() writing to and potentially corrupting the memory region.
1444 *
1445 * We want to perform the check before __ksize(), to avoid potentially
1446 * crashing in __ksize() due to accessing invalid metadata.
1447 */
1448 if (unlikely(ZERO_OR_NULL_PTR(objp)) || !kasan_check_byte(objp))
1449 return 0;
1450
1451 return kfence_ksize(objp) ?: __ksize(objp);
1452}
1453EXPORT_SYMBOL(ksize);
1454
1455/* Tracepoints definitions. */
1456EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1457EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1458EXPORT_TRACEPOINT_SYMBOL(kfree);
1459EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);
1460
1461int should_failslab(struct kmem_cache *s, gfp_t gfpflags)
1462{
1463 if (__should_failslab(s, gfpflags))
1464 return -ENOMEM;
1465 return 0;
1466}
1467ALLOW_ERROR_INJECTION(should_failslab, ERRNO);