]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - mm/slab_common.c
Merge tag 'io_uring-5.11-2020-12-23' of git://git.kernel.dk/linux-block
[mirror_ubuntu-jammy-kernel.git] / mm / slab_common.c
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 <linux/kasan.h>
22 #include <asm/cacheflush.h>
23 #include <asm/tlbflush.h>
24 #include <asm/page.h>
25 #include <linux/memcontrol.h>
26
27 #define CREATE_TRACE_POINTS
28 #include <trace/events/kmem.h>
29
30 #include "internal.h"
31
32 #include "slab.h"
33
34 enum slab_state slab_state;
35 LIST_HEAD(slab_caches);
36 DEFINE_MUTEX(slab_mutex);
37 struct kmem_cache *kmem_cache;
38
39 #ifdef CONFIG_HARDENED_USERCOPY
40 bool usercopy_fallback __ro_after_init =
41 IS_ENABLED(CONFIG_HARDENED_USERCOPY_FALLBACK);
42 module_param(usercopy_fallback, bool, 0400);
43 MODULE_PARM_DESC(usercopy_fallback,
44 "WARN instead of reject usercopy whitelist violations");
45 #endif
46
47 static LIST_HEAD(slab_caches_to_rcu_destroy);
48 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work);
49 static DECLARE_WORK(slab_caches_to_rcu_destroy_work,
50 slab_caches_to_rcu_destroy_workfn);
51
52 /*
53 * Set of flags that will prevent slab merging
54 */
55 #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
56 SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \
57 SLAB_FAILSLAB | kasan_never_merge())
58
59 #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \
60 SLAB_CACHE_DMA32 | SLAB_ACCOUNT)
61
62 /*
63 * Merge control. If this is set then no merging of slab caches will occur.
64 */
65 static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT);
66
67 static int __init setup_slab_nomerge(char *str)
68 {
69 slab_nomerge = true;
70 return 1;
71 }
72
73 #ifdef CONFIG_SLUB
74 __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0);
75 #endif
76
77 __setup("slab_nomerge", setup_slab_nomerge);
78
79 /*
80 * Determine the size of a slab object
81 */
82 unsigned int kmem_cache_size(struct kmem_cache *s)
83 {
84 return s->object_size;
85 }
86 EXPORT_SYMBOL(kmem_cache_size);
87
88 #ifdef CONFIG_DEBUG_VM
89 static int kmem_cache_sanity_check(const char *name, unsigned int size)
90 {
91 if (!name || in_interrupt() || size < sizeof(void *) ||
92 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
101 static inline int kmem_cache_sanity_check(const char *name, unsigned int size)
102 {
103 return 0;
104 }
105 #endif
106
107 void __kmem_cache_free_bulk(struct kmem_cache *s, size_t nr, void **p)
108 {
109 size_t i;
110
111 for (i = 0; i < nr; i++) {
112 if (s)
113 kmem_cache_free(s, p[i]);
114 else
115 kfree(p[i]);
116 }
117 }
118
119 int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t nr,
120 void **p)
121 {
122 size_t i;
123
124 for (i = 0; i < nr; i++) {
125 void *x = p[i] = kmem_cache_alloc(s, flags);
126 if (!x) {
127 __kmem_cache_free_bulk(s, i, p);
128 return 0;
129 }
130 }
131 return i;
132 }
133
134 /*
135 * Figure out what the alignment of the objects will be given a set of
136 * flags, a user specified alignment and the size of the objects.
137 */
138 static unsigned int calculate_alignment(slab_flags_t flags,
139 unsigned int align, unsigned int size)
140 {
141 /*
142 * If the user wants hardware cache aligned objects then follow that
143 * suggestion if the object is sufficiently large.
144 *
145 * The hardware cache alignment cannot override the specified
146 * alignment though. If that is greater then use it.
147 */
148 if (flags & SLAB_HWCACHE_ALIGN) {
149 unsigned int ralign;
150
151 ralign = cache_line_size();
152 while (size <= ralign / 2)
153 ralign /= 2;
154 align = max(align, ralign);
155 }
156
157 if (align < ARCH_SLAB_MINALIGN)
158 align = ARCH_SLAB_MINALIGN;
159
160 return ALIGN(align, sizeof(void *));
161 }
162
163 /*
164 * Find a mergeable slab cache
165 */
166 int slab_unmergeable(struct kmem_cache *s)
167 {
168 if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
169 return 1;
170
171 if (s->ctor)
172 return 1;
173
174 if (s->usersize)
175 return 1;
176
177 /*
178 * We may have set a slab to be unmergeable during bootstrap.
179 */
180 if (s->refcount < 0)
181 return 1;
182
183 return 0;
184 }
185
186 struct kmem_cache *find_mergeable(unsigned int size, unsigned int align,
187 slab_flags_t flags, const char *name, void (*ctor)(void *))
188 {
189 struct kmem_cache *s;
190
191 if (slab_nomerge)
192 return NULL;
193
194 if (ctor)
195 return NULL;
196
197 size = ALIGN(size, sizeof(void *));
198 align = calculate_alignment(flags, align, size);
199 size = ALIGN(size, align);
200 flags = kmem_cache_flags(size, flags, name, NULL);
201
202 if (flags & SLAB_NEVER_MERGE)
203 return NULL;
204
205 list_for_each_entry_reverse(s, &slab_caches, list) {
206 if (slab_unmergeable(s))
207 continue;
208
209 if (size > s->size)
210 continue;
211
212 if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
213 continue;
214 /*
215 * Check if alignment is compatible.
216 * Courtesy of Adrian Drzewiecki
217 */
218 if ((s->size & ~(align - 1)) != s->size)
219 continue;
220
221 if (s->size - size >= sizeof(void *))
222 continue;
223
224 if (IS_ENABLED(CONFIG_SLAB) && align &&
225 (align > s->align || s->align % align))
226 continue;
227
228 return s;
229 }
230 return NULL;
231 }
232
233 static struct kmem_cache *create_cache(const char *name,
234 unsigned int object_size, unsigned int align,
235 slab_flags_t flags, unsigned int useroffset,
236 unsigned int usersize, void (*ctor)(void *),
237 struct kmem_cache *root_cache)
238 {
239 struct kmem_cache *s;
240 int err;
241
242 if (WARN_ON(useroffset + usersize > object_size))
243 useroffset = usersize = 0;
244
245 err = -ENOMEM;
246 s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
247 if (!s)
248 goto out;
249
250 s->name = name;
251 s->size = s->object_size = object_size;
252 s->align = align;
253 s->ctor = ctor;
254 s->useroffset = useroffset;
255 s->usersize = usersize;
256
257 err = __kmem_cache_create(s, flags);
258 if (err)
259 goto out_free_cache;
260
261 s->refcount = 1;
262 list_add(&s->list, &slab_caches);
263 out:
264 if (err)
265 return ERR_PTR(err);
266 return s;
267
268 out_free_cache:
269 kmem_cache_free(kmem_cache, s);
270 goto out;
271 }
272
273 /**
274 * kmem_cache_create_usercopy - Create a cache with a region suitable
275 * for copying to userspace
276 * @name: A string which is used in /proc/slabinfo to identify this cache.
277 * @size: The size of objects to be created in this cache.
278 * @align: The required alignment for the objects.
279 * @flags: SLAB flags
280 * @useroffset: Usercopy region offset
281 * @usersize: Usercopy region size
282 * @ctor: A constructor for the objects.
283 *
284 * Cannot be called within a interrupt, but can be interrupted.
285 * The @ctor is run when new pages are allocated by the cache.
286 *
287 * The flags are
288 *
289 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
290 * to catch references to uninitialised memory.
291 *
292 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
293 * for buffer overruns.
294 *
295 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
296 * cacheline. This can be beneficial if you're counting cycles as closely
297 * as davem.
298 *
299 * Return: a pointer to the cache on success, NULL on failure.
300 */
301 struct kmem_cache *
302 kmem_cache_create_usercopy(const char *name,
303 unsigned int size, unsigned int align,
304 slab_flags_t flags,
305 unsigned int useroffset, unsigned int usersize,
306 void (*ctor)(void *))
307 {
308 struct kmem_cache *s = NULL;
309 const char *cache_name;
310 int err;
311
312 get_online_cpus();
313 get_online_mems();
314
315 mutex_lock(&slab_mutex);
316
317 err = kmem_cache_sanity_check(name, size);
318 if (err) {
319 goto out_unlock;
320 }
321
322 /* Refuse requests with allocator specific flags */
323 if (flags & ~SLAB_FLAGS_PERMITTED) {
324 err = -EINVAL;
325 goto out_unlock;
326 }
327
328 /*
329 * Some allocators will constraint the set of valid flags to a subset
330 * of all flags. We expect them to define CACHE_CREATE_MASK in this
331 * case, and we'll just provide them with a sanitized version of the
332 * passed flags.
333 */
334 flags &= CACHE_CREATE_MASK;
335
336 /* Fail closed on bad usersize of useroffset values. */
337 if (WARN_ON(!usersize && useroffset) ||
338 WARN_ON(size < usersize || size - usersize < useroffset))
339 usersize = useroffset = 0;
340
341 if (!usersize)
342 s = __kmem_cache_alias(name, size, align, flags, ctor);
343 if (s)
344 goto out_unlock;
345
346 cache_name = kstrdup_const(name, GFP_KERNEL);
347 if (!cache_name) {
348 err = -ENOMEM;
349 goto out_unlock;
350 }
351
352 s = create_cache(cache_name, size,
353 calculate_alignment(flags, align, size),
354 flags, useroffset, usersize, ctor, NULL);
355 if (IS_ERR(s)) {
356 err = PTR_ERR(s);
357 kfree_const(cache_name);
358 }
359
360 out_unlock:
361 mutex_unlock(&slab_mutex);
362
363 put_online_mems();
364 put_online_cpus();
365
366 if (err) {
367 if (flags & SLAB_PANIC)
368 panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n",
369 name, err);
370 else {
371 pr_warn("kmem_cache_create(%s) failed with error %d\n",
372 name, err);
373 dump_stack();
374 }
375 return NULL;
376 }
377 return s;
378 }
379 EXPORT_SYMBOL(kmem_cache_create_usercopy);
380
381 /**
382 * kmem_cache_create - Create a cache.
383 * @name: A string which is used in /proc/slabinfo to identify this cache.
384 * @size: The size of objects to be created in this cache.
385 * @align: The required alignment for the objects.
386 * @flags: SLAB flags
387 * @ctor: A constructor for the objects.
388 *
389 * Cannot be called within a interrupt, but can be interrupted.
390 * The @ctor is run when new pages are allocated by the cache.
391 *
392 * The flags are
393 *
394 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
395 * to catch references to uninitialised memory.
396 *
397 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
398 * for buffer overruns.
399 *
400 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
401 * cacheline. This can be beneficial if you're counting cycles as closely
402 * as davem.
403 *
404 * Return: a pointer to the cache on success, NULL on failure.
405 */
406 struct kmem_cache *
407 kmem_cache_create(const char *name, unsigned int size, unsigned int align,
408 slab_flags_t flags, void (*ctor)(void *))
409 {
410 return kmem_cache_create_usercopy(name, size, align, flags, 0, 0,
411 ctor);
412 }
413 EXPORT_SYMBOL(kmem_cache_create);
414
415 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work)
416 {
417 LIST_HEAD(to_destroy);
418 struct kmem_cache *s, *s2;
419
420 /*
421 * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the
422 * @slab_caches_to_rcu_destroy list. The slab pages are freed
423 * through RCU and the associated kmem_cache are dereferenced
424 * while freeing the pages, so the kmem_caches should be freed only
425 * after the pending RCU operations are finished. As rcu_barrier()
426 * is a pretty slow operation, we batch all pending destructions
427 * asynchronously.
428 */
429 mutex_lock(&slab_mutex);
430 list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy);
431 mutex_unlock(&slab_mutex);
432
433 if (list_empty(&to_destroy))
434 return;
435
436 rcu_barrier();
437
438 list_for_each_entry_safe(s, s2, &to_destroy, list) {
439 #ifdef SLAB_SUPPORTS_SYSFS
440 sysfs_slab_release(s);
441 #else
442 slab_kmem_cache_release(s);
443 #endif
444 }
445 }
446
447 static int shutdown_cache(struct kmem_cache *s)
448 {
449 /* free asan quarantined objects */
450 kasan_cache_shutdown(s);
451
452 if (__kmem_cache_shutdown(s) != 0)
453 return -EBUSY;
454
455 list_del(&s->list);
456
457 if (s->flags & SLAB_TYPESAFE_BY_RCU) {
458 #ifdef SLAB_SUPPORTS_SYSFS
459 sysfs_slab_unlink(s);
460 #endif
461 list_add_tail(&s->list, &slab_caches_to_rcu_destroy);
462 schedule_work(&slab_caches_to_rcu_destroy_work);
463 } else {
464 #ifdef SLAB_SUPPORTS_SYSFS
465 sysfs_slab_unlink(s);
466 sysfs_slab_release(s);
467 #else
468 slab_kmem_cache_release(s);
469 #endif
470 }
471
472 return 0;
473 }
474
475 void slab_kmem_cache_release(struct kmem_cache *s)
476 {
477 __kmem_cache_release(s);
478 kfree_const(s->name);
479 kmem_cache_free(kmem_cache, s);
480 }
481
482 void kmem_cache_destroy(struct kmem_cache *s)
483 {
484 int err;
485
486 if (unlikely(!s))
487 return;
488
489 get_online_cpus();
490 get_online_mems();
491
492 mutex_lock(&slab_mutex);
493
494 s->refcount--;
495 if (s->refcount)
496 goto out_unlock;
497
498 err = shutdown_cache(s);
499 if (err) {
500 pr_err("kmem_cache_destroy %s: Slab cache still has objects\n",
501 s->name);
502 dump_stack();
503 }
504 out_unlock:
505 mutex_unlock(&slab_mutex);
506
507 put_online_mems();
508 put_online_cpus();
509 }
510 EXPORT_SYMBOL(kmem_cache_destroy);
511
512 /**
513 * kmem_cache_shrink - Shrink a cache.
514 * @cachep: The cache to shrink.
515 *
516 * Releases as many slabs as possible for a cache.
517 * To help debugging, a zero exit status indicates all slabs were released.
518 *
519 * Return: %0 if all slabs were released, non-zero otherwise
520 */
521 int kmem_cache_shrink(struct kmem_cache *cachep)
522 {
523 int ret;
524
525 get_online_cpus();
526 get_online_mems();
527 kasan_cache_shrink(cachep);
528 ret = __kmem_cache_shrink(cachep);
529 put_online_mems();
530 put_online_cpus();
531 return ret;
532 }
533 EXPORT_SYMBOL(kmem_cache_shrink);
534
535 bool slab_is_available(void)
536 {
537 return slab_state >= UP;
538 }
539
540 #ifndef CONFIG_SLOB
541 /* Create a cache during boot when no slab services are available yet */
542 void __init create_boot_cache(struct kmem_cache *s, const char *name,
543 unsigned int size, slab_flags_t flags,
544 unsigned int useroffset, unsigned int usersize)
545 {
546 int err;
547 unsigned int align = ARCH_KMALLOC_MINALIGN;
548
549 s->name = name;
550 s->size = s->object_size = size;
551
552 /*
553 * For power of two sizes, guarantee natural alignment for kmalloc
554 * caches, regardless of SL*B debugging options.
555 */
556 if (is_power_of_2(size))
557 align = max(align, size);
558 s->align = calculate_alignment(flags, align, size);
559
560 s->useroffset = useroffset;
561 s->usersize = usersize;
562
563 err = __kmem_cache_create(s, flags);
564
565 if (err)
566 panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n",
567 name, size, err);
568
569 s->refcount = -1; /* Exempt from merging for now */
570 }
571
572 struct kmem_cache *__init create_kmalloc_cache(const char *name,
573 unsigned int size, slab_flags_t flags,
574 unsigned int useroffset, unsigned int usersize)
575 {
576 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
577
578 if (!s)
579 panic("Out of memory when creating slab %s\n", name);
580
581 create_boot_cache(s, name, size, flags, useroffset, usersize);
582 list_add(&s->list, &slab_caches);
583 s->refcount = 1;
584 return s;
585 }
586
587 struct kmem_cache *
588 kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1] __ro_after_init =
589 { /* initialization for https://bugs.llvm.org/show_bug.cgi?id=42570 */ };
590 EXPORT_SYMBOL(kmalloc_caches);
591
592 /*
593 * Conversion table for small slabs sizes / 8 to the index in the
594 * kmalloc array. This is necessary for slabs < 192 since we have non power
595 * of two cache sizes there. The size of larger slabs can be determined using
596 * fls.
597 */
598 static u8 size_index[24] __ro_after_init = {
599 3, /* 8 */
600 4, /* 16 */
601 5, /* 24 */
602 5, /* 32 */
603 6, /* 40 */
604 6, /* 48 */
605 6, /* 56 */
606 6, /* 64 */
607 1, /* 72 */
608 1, /* 80 */
609 1, /* 88 */
610 1, /* 96 */
611 7, /* 104 */
612 7, /* 112 */
613 7, /* 120 */
614 7, /* 128 */
615 2, /* 136 */
616 2, /* 144 */
617 2, /* 152 */
618 2, /* 160 */
619 2, /* 168 */
620 2, /* 176 */
621 2, /* 184 */
622 2 /* 192 */
623 };
624
625 static inline unsigned int size_index_elem(unsigned int bytes)
626 {
627 return (bytes - 1) / 8;
628 }
629
630 /*
631 * Find the kmem_cache structure that serves a given size of
632 * allocation
633 */
634 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
635 {
636 unsigned int index;
637
638 if (size <= 192) {
639 if (!size)
640 return ZERO_SIZE_PTR;
641
642 index = size_index[size_index_elem(size)];
643 } else {
644 if (WARN_ON_ONCE(size > KMALLOC_MAX_CACHE_SIZE))
645 return NULL;
646 index = fls(size - 1);
647 }
648
649 return kmalloc_caches[kmalloc_type(flags)][index];
650 }
651
652 #ifdef CONFIG_ZONE_DMA
653 #define INIT_KMALLOC_INFO(__size, __short_size) \
654 { \
655 .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \
656 .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #__short_size, \
657 .name[KMALLOC_DMA] = "dma-kmalloc-" #__short_size, \
658 .size = __size, \
659 }
660 #else
661 #define INIT_KMALLOC_INFO(__size, __short_size) \
662 { \
663 .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \
664 .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #__short_size, \
665 .size = __size, \
666 }
667 #endif
668
669 /*
670 * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
671 * kmalloc_index() supports up to 2^26=64MB, so the final entry of the table is
672 * kmalloc-67108864.
673 */
674 const struct kmalloc_info_struct kmalloc_info[] __initconst = {
675 INIT_KMALLOC_INFO(0, 0),
676 INIT_KMALLOC_INFO(96, 96),
677 INIT_KMALLOC_INFO(192, 192),
678 INIT_KMALLOC_INFO(8, 8),
679 INIT_KMALLOC_INFO(16, 16),
680 INIT_KMALLOC_INFO(32, 32),
681 INIT_KMALLOC_INFO(64, 64),
682 INIT_KMALLOC_INFO(128, 128),
683 INIT_KMALLOC_INFO(256, 256),
684 INIT_KMALLOC_INFO(512, 512),
685 INIT_KMALLOC_INFO(1024, 1k),
686 INIT_KMALLOC_INFO(2048, 2k),
687 INIT_KMALLOC_INFO(4096, 4k),
688 INIT_KMALLOC_INFO(8192, 8k),
689 INIT_KMALLOC_INFO(16384, 16k),
690 INIT_KMALLOC_INFO(32768, 32k),
691 INIT_KMALLOC_INFO(65536, 64k),
692 INIT_KMALLOC_INFO(131072, 128k),
693 INIT_KMALLOC_INFO(262144, 256k),
694 INIT_KMALLOC_INFO(524288, 512k),
695 INIT_KMALLOC_INFO(1048576, 1M),
696 INIT_KMALLOC_INFO(2097152, 2M),
697 INIT_KMALLOC_INFO(4194304, 4M),
698 INIT_KMALLOC_INFO(8388608, 8M),
699 INIT_KMALLOC_INFO(16777216, 16M),
700 INIT_KMALLOC_INFO(33554432, 32M),
701 INIT_KMALLOC_INFO(67108864, 64M)
702 };
703
704 /*
705 * Patch up the size_index table if we have strange large alignment
706 * requirements for the kmalloc array. This is only the case for
707 * MIPS it seems. The standard arches will not generate any code here.
708 *
709 * Largest permitted alignment is 256 bytes due to the way we
710 * handle the index determination for the smaller caches.
711 *
712 * Make sure that nothing crazy happens if someone starts tinkering
713 * around with ARCH_KMALLOC_MINALIGN
714 */
715 void __init setup_kmalloc_cache_index_table(void)
716 {
717 unsigned int i;
718
719 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
720 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
721
722 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
723 unsigned int elem = size_index_elem(i);
724
725 if (elem >= ARRAY_SIZE(size_index))
726 break;
727 size_index[elem] = KMALLOC_SHIFT_LOW;
728 }
729
730 if (KMALLOC_MIN_SIZE >= 64) {
731 /*
732 * The 96 byte size cache is not used if the alignment
733 * is 64 byte.
734 */
735 for (i = 64 + 8; i <= 96; i += 8)
736 size_index[size_index_elem(i)] = 7;
737
738 }
739
740 if (KMALLOC_MIN_SIZE >= 128) {
741 /*
742 * The 192 byte sized cache is not used if the alignment
743 * is 128 byte. Redirect kmalloc to use the 256 byte cache
744 * instead.
745 */
746 for (i = 128 + 8; i <= 192; i += 8)
747 size_index[size_index_elem(i)] = 8;
748 }
749 }
750
751 static void __init
752 new_kmalloc_cache(int idx, enum kmalloc_cache_type type, slab_flags_t flags)
753 {
754 if (type == KMALLOC_RECLAIM)
755 flags |= SLAB_RECLAIM_ACCOUNT;
756
757 kmalloc_caches[type][idx] = create_kmalloc_cache(
758 kmalloc_info[idx].name[type],
759 kmalloc_info[idx].size, flags, 0,
760 kmalloc_info[idx].size);
761 }
762
763 /*
764 * Create the kmalloc array. Some of the regular kmalloc arrays
765 * may already have been created because they were needed to
766 * enable allocations for slab creation.
767 */
768 void __init create_kmalloc_caches(slab_flags_t flags)
769 {
770 int i;
771 enum kmalloc_cache_type type;
772
773 for (type = KMALLOC_NORMAL; type <= KMALLOC_RECLAIM; type++) {
774 for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
775 if (!kmalloc_caches[type][i])
776 new_kmalloc_cache(i, type, flags);
777
778 /*
779 * Caches that are not of the two-to-the-power-of size.
780 * These have to be created immediately after the
781 * earlier power of two caches
782 */
783 if (KMALLOC_MIN_SIZE <= 32 && i == 6 &&
784 !kmalloc_caches[type][1])
785 new_kmalloc_cache(1, type, flags);
786 if (KMALLOC_MIN_SIZE <= 64 && i == 7 &&
787 !kmalloc_caches[type][2])
788 new_kmalloc_cache(2, type, flags);
789 }
790 }
791
792 /* Kmalloc array is now usable */
793 slab_state = UP;
794
795 #ifdef CONFIG_ZONE_DMA
796 for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
797 struct kmem_cache *s = kmalloc_caches[KMALLOC_NORMAL][i];
798
799 if (s) {
800 kmalloc_caches[KMALLOC_DMA][i] = create_kmalloc_cache(
801 kmalloc_info[i].name[KMALLOC_DMA],
802 kmalloc_info[i].size,
803 SLAB_CACHE_DMA | flags, 0,
804 kmalloc_info[i].size);
805 }
806 }
807 #endif
808 }
809 #endif /* !CONFIG_SLOB */
810
811 gfp_t kmalloc_fix_flags(gfp_t flags)
812 {
813 gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
814
815 flags &= ~GFP_SLAB_BUG_MASK;
816 pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
817 invalid_mask, &invalid_mask, flags, &flags);
818 dump_stack();
819
820 return flags;
821 }
822
823 /*
824 * To avoid unnecessary overhead, we pass through large allocation requests
825 * directly to the page allocator. We use __GFP_COMP, because we will need to
826 * know the allocation order to free the pages properly in kfree.
827 */
828 void *kmalloc_order(size_t size, gfp_t flags, unsigned int order)
829 {
830 void *ret = NULL;
831 struct page *page;
832
833 if (unlikely(flags & GFP_SLAB_BUG_MASK))
834 flags = kmalloc_fix_flags(flags);
835
836 flags |= __GFP_COMP;
837 page = alloc_pages(flags, order);
838 if (likely(page)) {
839 ret = page_address(page);
840 mod_node_page_state(page_pgdat(page), NR_SLAB_UNRECLAIMABLE_B,
841 PAGE_SIZE << order);
842 }
843 ret = kasan_kmalloc_large(ret, size, flags);
844 /* As ret might get tagged, call kmemleak hook after KASAN. */
845 kmemleak_alloc(ret, size, 1, flags);
846 return ret;
847 }
848 EXPORT_SYMBOL(kmalloc_order);
849
850 #ifdef CONFIG_TRACING
851 void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
852 {
853 void *ret = kmalloc_order(size, flags, order);
854 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
855 return ret;
856 }
857 EXPORT_SYMBOL(kmalloc_order_trace);
858 #endif
859
860 #ifdef CONFIG_SLAB_FREELIST_RANDOM
861 /* Randomize a generic freelist */
862 static void freelist_randomize(struct rnd_state *state, unsigned int *list,
863 unsigned int count)
864 {
865 unsigned int rand;
866 unsigned int i;
867
868 for (i = 0; i < count; i++)
869 list[i] = i;
870
871 /* Fisher-Yates shuffle */
872 for (i = count - 1; i > 0; i--) {
873 rand = prandom_u32_state(state);
874 rand %= (i + 1);
875 swap(list[i], list[rand]);
876 }
877 }
878
879 /* Create a random sequence per cache */
880 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
881 gfp_t gfp)
882 {
883 struct rnd_state state;
884
885 if (count < 2 || cachep->random_seq)
886 return 0;
887
888 cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp);
889 if (!cachep->random_seq)
890 return -ENOMEM;
891
892 /* Get best entropy at this stage of boot */
893 prandom_seed_state(&state, get_random_long());
894
895 freelist_randomize(&state, cachep->random_seq, count);
896 return 0;
897 }
898
899 /* Destroy the per-cache random freelist sequence */
900 void cache_random_seq_destroy(struct kmem_cache *cachep)
901 {
902 kfree(cachep->random_seq);
903 cachep->random_seq = NULL;
904 }
905 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
906
907 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
908 #ifdef CONFIG_SLAB
909 #define SLABINFO_RIGHTS (0600)
910 #else
911 #define SLABINFO_RIGHTS (0400)
912 #endif
913
914 static void print_slabinfo_header(struct seq_file *m)
915 {
916 /*
917 * Output format version, so at least we can change it
918 * without _too_ many complaints.
919 */
920 #ifdef CONFIG_DEBUG_SLAB
921 seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
922 #else
923 seq_puts(m, "slabinfo - version: 2.1\n");
924 #endif
925 seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
926 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
927 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
928 #ifdef CONFIG_DEBUG_SLAB
929 seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
930 seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
931 #endif
932 seq_putc(m, '\n');
933 }
934
935 void *slab_start(struct seq_file *m, loff_t *pos)
936 {
937 mutex_lock(&slab_mutex);
938 return seq_list_start(&slab_caches, *pos);
939 }
940
941 void *slab_next(struct seq_file *m, void *p, loff_t *pos)
942 {
943 return seq_list_next(p, &slab_caches, pos);
944 }
945
946 void slab_stop(struct seq_file *m, void *p)
947 {
948 mutex_unlock(&slab_mutex);
949 }
950
951 static void cache_show(struct kmem_cache *s, struct seq_file *m)
952 {
953 struct slabinfo sinfo;
954
955 memset(&sinfo, 0, sizeof(sinfo));
956 get_slabinfo(s, &sinfo);
957
958 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
959 s->name, sinfo.active_objs, sinfo.num_objs, s->size,
960 sinfo.objects_per_slab, (1 << sinfo.cache_order));
961
962 seq_printf(m, " : tunables %4u %4u %4u",
963 sinfo.limit, sinfo.batchcount, sinfo.shared);
964 seq_printf(m, " : slabdata %6lu %6lu %6lu",
965 sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
966 slabinfo_show_stats(m, s);
967 seq_putc(m, '\n');
968 }
969
970 static int slab_show(struct seq_file *m, void *p)
971 {
972 struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
973
974 if (p == slab_caches.next)
975 print_slabinfo_header(m);
976 cache_show(s, m);
977 return 0;
978 }
979
980 void dump_unreclaimable_slab(void)
981 {
982 struct kmem_cache *s;
983 struct slabinfo sinfo;
984
985 /*
986 * Here acquiring slab_mutex is risky since we don't prefer to get
987 * sleep in oom path. But, without mutex hold, it may introduce a
988 * risk of crash.
989 * Use mutex_trylock to protect the list traverse, dump nothing
990 * without acquiring the mutex.
991 */
992 if (!mutex_trylock(&slab_mutex)) {
993 pr_warn("excessive unreclaimable slab but cannot dump stats\n");
994 return;
995 }
996
997 pr_info("Unreclaimable slab info:\n");
998 pr_info("Name Used Total\n");
999
1000 list_for_each_entry(s, &slab_caches, list) {
1001 if (s->flags & SLAB_RECLAIM_ACCOUNT)
1002 continue;
1003
1004 get_slabinfo(s, &sinfo);
1005
1006 if (sinfo.num_objs > 0)
1007 pr_info("%-17s %10luKB %10luKB\n", s->name,
1008 (sinfo.active_objs * s->size) / 1024,
1009 (sinfo.num_objs * s->size) / 1024);
1010 }
1011 mutex_unlock(&slab_mutex);
1012 }
1013
1014 #if defined(CONFIG_MEMCG_KMEM)
1015 int memcg_slab_show(struct seq_file *m, void *p)
1016 {
1017 /*
1018 * Deprecated.
1019 * Please, take a look at tools/cgroup/slabinfo.py .
1020 */
1021 return 0;
1022 }
1023 #endif
1024
1025 /*
1026 * slabinfo_op - iterator that generates /proc/slabinfo
1027 *
1028 * Output layout:
1029 * cache-name
1030 * num-active-objs
1031 * total-objs
1032 * object size
1033 * num-active-slabs
1034 * total-slabs
1035 * num-pages-per-slab
1036 * + further values on SMP and with statistics enabled
1037 */
1038 static const struct seq_operations slabinfo_op = {
1039 .start = slab_start,
1040 .next = slab_next,
1041 .stop = slab_stop,
1042 .show = slab_show,
1043 };
1044
1045 static int slabinfo_open(struct inode *inode, struct file *file)
1046 {
1047 return seq_open(file, &slabinfo_op);
1048 }
1049
1050 static const struct proc_ops slabinfo_proc_ops = {
1051 .proc_flags = PROC_ENTRY_PERMANENT,
1052 .proc_open = slabinfo_open,
1053 .proc_read = seq_read,
1054 .proc_write = slabinfo_write,
1055 .proc_lseek = seq_lseek,
1056 .proc_release = seq_release,
1057 };
1058
1059 static int __init slab_proc_init(void)
1060 {
1061 proc_create("slabinfo", SLABINFO_RIGHTS, NULL, &slabinfo_proc_ops);
1062 return 0;
1063 }
1064 module_init(slab_proc_init);
1065
1066 #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */
1067
1068 static __always_inline void *__do_krealloc(const void *p, size_t new_size,
1069 gfp_t flags)
1070 {
1071 void *ret;
1072 size_t ks;
1073
1074 ks = ksize(p);
1075
1076 if (ks >= new_size) {
1077 p = kasan_krealloc((void *)p, new_size, flags);
1078 return (void *)p;
1079 }
1080
1081 ret = kmalloc_track_caller(new_size, flags);
1082 if (ret && p)
1083 memcpy(ret, p, ks);
1084
1085 return ret;
1086 }
1087
1088 /**
1089 * krealloc - reallocate memory. The contents will remain unchanged.
1090 * @p: object to reallocate memory for.
1091 * @new_size: how many bytes of memory are required.
1092 * @flags: the type of memory to allocate.
1093 *
1094 * The contents of the object pointed to are preserved up to the
1095 * lesser of the new and old sizes (__GFP_ZERO flag is effectively ignored).
1096 * If @p is %NULL, krealloc() behaves exactly like kmalloc(). If @new_size
1097 * is 0 and @p is not a %NULL pointer, the object pointed to is freed.
1098 *
1099 * Return: pointer to the allocated memory or %NULL in case of error
1100 */
1101 void *krealloc(const void *p, size_t new_size, gfp_t flags)
1102 {
1103 void *ret;
1104
1105 if (unlikely(!new_size)) {
1106 kfree(p);
1107 return ZERO_SIZE_PTR;
1108 }
1109
1110 ret = __do_krealloc(p, new_size, flags);
1111 if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret))
1112 kfree(p);
1113
1114 return ret;
1115 }
1116 EXPORT_SYMBOL(krealloc);
1117
1118 /**
1119 * kfree_sensitive - Clear sensitive information in memory before freeing
1120 * @p: object to free memory of
1121 *
1122 * The memory of the object @p points to is zeroed before freed.
1123 * If @p is %NULL, kfree_sensitive() does nothing.
1124 *
1125 * Note: this function zeroes the whole allocated buffer which can be a good
1126 * deal bigger than the requested buffer size passed to kmalloc(). So be
1127 * careful when using this function in performance sensitive code.
1128 */
1129 void kfree_sensitive(const void *p)
1130 {
1131 size_t ks;
1132 void *mem = (void *)p;
1133
1134 ks = ksize(mem);
1135 if (ks)
1136 memzero_explicit(mem, ks);
1137 kfree(mem);
1138 }
1139 EXPORT_SYMBOL(kfree_sensitive);
1140
1141 /**
1142 * ksize - get the actual amount of memory allocated for a given object
1143 * @objp: Pointer to the object
1144 *
1145 * kmalloc may internally round up allocations and return more memory
1146 * than requested. ksize() can be used to determine the actual amount of
1147 * memory allocated. The caller may use this additional memory, even though
1148 * a smaller amount of memory was initially specified with the kmalloc call.
1149 * The caller must guarantee that objp points to a valid object previously
1150 * allocated with either kmalloc() or kmem_cache_alloc(). The object
1151 * must not be freed during the duration of the call.
1152 *
1153 * Return: size of the actual memory used by @objp in bytes
1154 */
1155 size_t ksize(const void *objp)
1156 {
1157 size_t size;
1158
1159 /*
1160 * We need to check that the pointed to object is valid, and only then
1161 * unpoison the shadow memory below. We use __kasan_check_read(), to
1162 * generate a more useful report at the time ksize() is called (rather
1163 * than later where behaviour is undefined due to potential
1164 * use-after-free or double-free).
1165 *
1166 * If the pointed to memory is invalid we return 0, to avoid users of
1167 * ksize() writing to and potentially corrupting the memory region.
1168 *
1169 * We want to perform the check before __ksize(), to avoid potentially
1170 * crashing in __ksize() due to accessing invalid metadata.
1171 */
1172 if (unlikely(ZERO_OR_NULL_PTR(objp)) || !__kasan_check_read(objp, 1))
1173 return 0;
1174
1175 size = __ksize(objp);
1176 /*
1177 * We assume that ksize callers could use whole allocated area,
1178 * so we need to unpoison this area.
1179 */
1180 kasan_unpoison_range(objp, size);
1181 return size;
1182 }
1183 EXPORT_SYMBOL(ksize);
1184
1185 /* Tracepoints definitions. */
1186 EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1187 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1188 EXPORT_TRACEPOINT_SYMBOL(kmalloc_node);
1189 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node);
1190 EXPORT_TRACEPOINT_SYMBOL(kfree);
1191 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);
1192
1193 int should_failslab(struct kmem_cache *s, gfp_t gfpflags)
1194 {
1195 if (__should_failslab(s, gfpflags))
1196 return -ENOMEM;
1197 return 0;
1198 }
1199 ALLOW_ERROR_INJECTION(should_failslab, ERRNO);