1 /*
2 * Slab allocator functions that are independent of the allocator strategy
3 *
4 * (C) 2012 Christoph Lameter <cl@linux.com>
5 */
6 #include <linux/slab.h>
7
8 #include <linux/mm.h>
9 #include <linux/poison.h>
10 #include <linux/interrupt.h>
11 #include <linux/memory.h>
12 #include <linux/compiler.h>
13 #include <linux/module.h>
14 #include <linux/cpu.h>
15 #include <linux/uaccess.h>
16 #include <linux/seq_file.h>
17 #include <linux/proc_fs.h>
18 #include <asm/cacheflush.h>
19 #include <asm/tlbflush.h>
20 #include <asm/page.h>
21 #include <linux/memcontrol.h>
22
23 #define CREATE_TRACE_POINTS
24 #include <trace/events/kmem.h>
25
26 #include "slab.h"
27
28 enum slab_state slab_state;
29 LIST_HEAD(slab_caches);
30 DEFINE_MUTEX(slab_mutex);
31 struct kmem_cache *kmem_cache;
32
33 /*
34 * Set of flags that will prevent slab merging
35 */
36 #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
37 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \
38 SLAB_FAILSLAB)
39
40 #define SLAB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
41 SLAB_CACHE_DMA | SLAB_NOTRACK)
42
43 /*
44 * Merge control. If this is set then no merging of slab caches will occur.
45 * (Could be removed. This was introduced to pacify the merge skeptics.)
46 */
47 static int slab_nomerge;
48
setup_slab_nomerge(char * str)49 static int __init setup_slab_nomerge(char *str)
50 {
51 slab_nomerge = 1;
52 return 1;
53 }
54
55 #ifdef CONFIG_SLUB
56 __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0);
57 #endif
58
59 __setup("slab_nomerge", setup_slab_nomerge);
60
61 /*
62 * Determine the size of a slab object
63 */
kmem_cache_size(struct kmem_cache * s)64 unsigned int kmem_cache_size(struct kmem_cache *s)
65 {
66 return s->object_size;
67 }
68 EXPORT_SYMBOL(kmem_cache_size);
69
70 #ifdef CONFIG_DEBUG_VM
kmem_cache_sanity_check(const char * name,size_t size)71 static int kmem_cache_sanity_check(const char *name, size_t size)
72 {
73 struct kmem_cache *s = NULL;
74
75 if (!name || in_interrupt() || size < sizeof(void *) ||
76 size > KMALLOC_MAX_SIZE) {
77 pr_err("kmem_cache_create(%s) integrity check failed\n", name);
78 return -EINVAL;
79 }
80
81 list_for_each_entry(s, &slab_caches, list) {
82 char tmp;
83 int res;
84
85 /*
86 * This happens when the module gets unloaded and doesn't
87 * destroy its slab cache and no-one else reuses the vmalloc
88 * area of the module. Print a warning.
89 */
90 res = probe_kernel_address(s->name, tmp);
91 if (res) {
92 pr_err("Slab cache with size %d has lost its name\n",
93 s->object_size);
94 continue;
95 }
96 }
97
98 WARN_ON(strchr(name, ' ')); /* It confuses parsers */
99 return 0;
100 }
101 #else
kmem_cache_sanity_check(const char * name,size_t size)102 static inline int kmem_cache_sanity_check(const char *name, size_t size)
103 {
104 return 0;
105 }
106 #endif
107
108 #ifdef CONFIG_MEMCG_KMEM
memcg_alloc_cache_params(struct mem_cgroup * memcg,struct kmem_cache * s,struct kmem_cache * root_cache)109 static int memcg_alloc_cache_params(struct mem_cgroup *memcg,
110 struct kmem_cache *s, struct kmem_cache *root_cache)
111 {
112 size_t size;
113
114 if (!memcg_kmem_enabled())
115 return 0;
116
117 if (!memcg) {
118 size = offsetof(struct memcg_cache_params, memcg_caches);
119 size += memcg_limited_groups_array_size * sizeof(void *);
120 } else
121 size = sizeof(struct memcg_cache_params);
122
123 s->memcg_params = kzalloc(size, GFP_KERNEL);
124 if (!s->memcg_params)
125 return -ENOMEM;
126
127 if (memcg) {
128 s->memcg_params->memcg = memcg;
129 s->memcg_params->root_cache = root_cache;
130 } else
131 s->memcg_params->is_root_cache = true;
132
133 return 0;
134 }
135
memcg_free_cache_params(struct kmem_cache * s)136 static void memcg_free_cache_params(struct kmem_cache *s)
137 {
138 kfree(s->memcg_params);
139 }
140
memcg_update_cache_params(struct kmem_cache * s,int num_memcgs)141 static int memcg_update_cache_params(struct kmem_cache *s, int num_memcgs)
142 {
143 int size;
144 struct memcg_cache_params *new_params, *cur_params;
145
146 BUG_ON(!is_root_cache(s));
147
148 size = offsetof(struct memcg_cache_params, memcg_caches);
149 size += num_memcgs * sizeof(void *);
150
151 new_params = kzalloc(size, GFP_KERNEL);
152 if (!new_params)
153 return -ENOMEM;
154
155 cur_params = s->memcg_params;
156 memcpy(new_params->memcg_caches, cur_params->memcg_caches,
157 memcg_limited_groups_array_size * sizeof(void *));
158
159 new_params->is_root_cache = true;
160
161 rcu_assign_pointer(s->memcg_params, new_params);
162 if (cur_params)
163 kfree_rcu(cur_params, rcu_head);
164
165 return 0;
166 }
167
memcg_update_all_caches(int num_memcgs)168 int memcg_update_all_caches(int num_memcgs)
169 {
170 struct kmem_cache *s;
171 int ret = 0;
172 mutex_lock(&slab_mutex);
173
174 list_for_each_entry(s, &slab_caches, list) {
175 if (!is_root_cache(s))
176 continue;
177
178 ret = memcg_update_cache_params(s, num_memcgs);
179 /*
180 * Instead of freeing the memory, we'll just leave the caches
181 * up to this point in an updated state.
182 */
183 if (ret)
184 goto out;
185 }
186
187 memcg_update_array_size(num_memcgs);
188 out:
189 mutex_unlock(&slab_mutex);
190 return ret;
191 }
192 #else
memcg_alloc_cache_params(struct mem_cgroup * memcg,struct kmem_cache * s,struct kmem_cache * root_cache)193 static inline int memcg_alloc_cache_params(struct mem_cgroup *memcg,
194 struct kmem_cache *s, struct kmem_cache *root_cache)
195 {
196 return 0;
197 }
198
memcg_free_cache_params(struct kmem_cache * s)199 static inline void memcg_free_cache_params(struct kmem_cache *s)
200 {
201 }
202 #endif /* CONFIG_MEMCG_KMEM */
203
204 /*
205 * Find a mergeable slab cache
206 */
slab_unmergeable(struct kmem_cache * s)207 int slab_unmergeable(struct kmem_cache *s)
208 {
209 if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
210 return 1;
211
212 if (!is_root_cache(s))
213 return 1;
214
215 if (s->ctor)
216 return 1;
217
218 /*
219 * We may have set a slab to be unmergeable during bootstrap.
220 */
221 if (s->refcount < 0)
222 return 1;
223
224 return 0;
225 }
226
find_mergeable(size_t size,size_t align,unsigned long flags,const char * name,void (* ctor)(void *))227 struct kmem_cache *find_mergeable(size_t size, size_t align,
228 unsigned long flags, const char *name, void (*ctor)(void *))
229 {
230 struct kmem_cache *s;
231
232 if (slab_nomerge || (flags & SLAB_NEVER_MERGE))
233 return NULL;
234
235 if (ctor)
236 return NULL;
237
238 size = ALIGN(size, sizeof(void *));
239 align = calculate_alignment(flags, align, size);
240 size = ALIGN(size, align);
241 flags = kmem_cache_flags(size, flags, name, NULL);
242
243 list_for_each_entry(s, &slab_caches, list) {
244 if (slab_unmergeable(s))
245 continue;
246
247 if (size > s->size)
248 continue;
249
250 if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
251 continue;
252 /*
253 * Check if alignment is compatible.
254 * Courtesy of Adrian Drzewiecki
255 */
256 if ((s->size & ~(align - 1)) != s->size)
257 continue;
258
259 if (s->size - size >= sizeof(void *))
260 continue;
261
262 if (IS_ENABLED(CONFIG_SLAB) && align &&
263 (align > s->align || s->align % align))
264 continue;
265
266 return s;
267 }
268 return NULL;
269 }
270
271 /*
272 * Figure out what the alignment of the objects will be given a set of
273 * flags, a user specified alignment and the size of the objects.
274 */
calculate_alignment(unsigned long flags,unsigned long align,unsigned long size)275 unsigned long calculate_alignment(unsigned long flags,
276 unsigned long align, unsigned long size)
277 {
278 /*
279 * If the user wants hardware cache aligned objects then follow that
280 * suggestion if the object is sufficiently large.
281 *
282 * The hardware cache alignment cannot override the specified
283 * alignment though. If that is greater then use it.
284 */
285 if (flags & SLAB_HWCACHE_ALIGN) {
286 unsigned long ralign = cache_line_size();
287 while (size <= ralign / 2)
288 ralign /= 2;
289 align = max(align, ralign);
290 }
291
292 if (align < ARCH_SLAB_MINALIGN)
293 align = ARCH_SLAB_MINALIGN;
294
295 return ALIGN(align, sizeof(void *));
296 }
297
298 static struct kmem_cache *
do_kmem_cache_create(char * name,size_t object_size,size_t size,size_t align,unsigned long flags,void (* ctor)(void *),struct mem_cgroup * memcg,struct kmem_cache * root_cache)299 do_kmem_cache_create(char *name, size_t object_size, size_t size, size_t align,
300 unsigned long flags, void (*ctor)(void *),
301 struct mem_cgroup *memcg, struct kmem_cache *root_cache)
302 {
303 struct kmem_cache *s;
304 int err;
305
306 err = -ENOMEM;
307 s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
308 if (!s)
309 goto out;
310
311 s->name = name;
312 s->object_size = object_size;
313 s->size = size;
314 s->align = align;
315 s->ctor = ctor;
316
317 err = memcg_alloc_cache_params(memcg, s, root_cache);
318 if (err)
319 goto out_free_cache;
320
321 err = __kmem_cache_create(s, flags);
322 if (err)
323 goto out_free_cache;
324
325 s->refcount = 1;
326 list_add(&s->list, &slab_caches);
327 out:
328 if (err)
329 return ERR_PTR(err);
330 return s;
331
332 out_free_cache:
333 memcg_free_cache_params(s);
334 kfree(s);
335 goto out;
336 }
337
338 /*
339 * kmem_cache_create - Create a cache.
340 * @name: A string which is used in /proc/slabinfo to identify this cache.
341 * @size: The size of objects to be created in this cache.
342 * @align: The required alignment for the objects.
343 * @flags: SLAB flags
344 * @ctor: A constructor for the objects.
345 *
346 * Returns a ptr to the cache on success, NULL on failure.
347 * Cannot be called within a interrupt, but can be interrupted.
348 * The @ctor is run when new pages are allocated by the cache.
349 *
350 * The flags are
351 *
352 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
353 * to catch references to uninitialised memory.
354 *
355 * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check
356 * for buffer overruns.
357 *
358 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
359 * cacheline. This can be beneficial if you're counting cycles as closely
360 * as davem.
361 */
362 struct kmem_cache *
kmem_cache_create(const char * name,size_t size,size_t align,unsigned long flags,void (* ctor)(void *))363 kmem_cache_create(const char *name, size_t size, size_t align,
364 unsigned long flags, void (*ctor)(void *))
365 {
366 struct kmem_cache *s;
367 char *cache_name;
368 int err;
369
370 get_online_cpus();
371 get_online_mems();
372
373 mutex_lock(&slab_mutex);
374
375 err = kmem_cache_sanity_check(name, size);
376 if (err) {
377 s = NULL; /* suppress uninit var warning */
378 goto out_unlock;
379 }
380
381 /*
382 * Some allocators will constraint the set of valid flags to a subset
383 * of all flags. We expect them to define CACHE_CREATE_MASK in this
384 * case, and we'll just provide them with a sanitized version of the
385 * passed flags.
386 */
387 flags &= CACHE_CREATE_MASK;
388
389 s = __kmem_cache_alias(name, size, align, flags, ctor);
390 if (s)
391 goto out_unlock;
392
393 cache_name = kstrdup(name, GFP_KERNEL);
394 if (!cache_name) {
395 err = -ENOMEM;
396 goto out_unlock;
397 }
398
399 s = do_kmem_cache_create(cache_name, size, size,
400 calculate_alignment(flags, align, size),
401 flags, ctor, NULL, NULL);
402 if (IS_ERR(s)) {
403 err = PTR_ERR(s);
404 kfree(cache_name);
405 }
406
407 out_unlock:
408 mutex_unlock(&slab_mutex);
409
410 put_online_mems();
411 put_online_cpus();
412
413 if (err) {
414 if (flags & SLAB_PANIC)
415 panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n",
416 name, err);
417 else {
418 printk(KERN_WARNING "kmem_cache_create(%s) failed with error %d",
419 name, err);
420 dump_stack();
421 }
422 return NULL;
423 }
424 return s;
425 }
426 EXPORT_SYMBOL(kmem_cache_create);
427
428 #ifdef CONFIG_MEMCG_KMEM
429 /*
430 * memcg_create_kmem_cache - Create a cache for a memory cgroup.
431 * @memcg: The memory cgroup the new cache is for.
432 * @root_cache: The parent of the new cache.
433 * @memcg_name: The name of the memory cgroup (used for naming the new cache).
434 *
435 * This function attempts to create a kmem cache that will serve allocation
436 * requests going from @memcg to @root_cache. The new cache inherits properties
437 * from its parent.
438 */
memcg_create_kmem_cache(struct mem_cgroup * memcg,struct kmem_cache * root_cache,const char * memcg_name)439 struct kmem_cache *memcg_create_kmem_cache(struct mem_cgroup *memcg,
440 struct kmem_cache *root_cache,
441 const char *memcg_name)
442 {
443 struct kmem_cache *s = NULL;
444 char *cache_name;
445
446 get_online_cpus();
447 get_online_mems();
448
449 mutex_lock(&slab_mutex);
450
451 cache_name = kasprintf(GFP_KERNEL, "%s(%d:%s)", root_cache->name,
452 memcg_cache_id(memcg), memcg_name);
453 if (!cache_name)
454 goto out_unlock;
455
456 s = do_kmem_cache_create(cache_name, root_cache->object_size,
457 root_cache->size, root_cache->align,
458 root_cache->flags, root_cache->ctor,
459 memcg, root_cache);
460 if (IS_ERR(s)) {
461 kfree(cache_name);
462 s = NULL;
463 }
464
465 out_unlock:
466 mutex_unlock(&slab_mutex);
467
468 put_online_mems();
469 put_online_cpus();
470
471 return s;
472 }
473
memcg_cleanup_cache_params(struct kmem_cache * s)474 static int memcg_cleanup_cache_params(struct kmem_cache *s)
475 {
476 int rc;
477
478 if (!s->memcg_params ||
479 !s->memcg_params->is_root_cache)
480 return 0;
481
482 mutex_unlock(&slab_mutex);
483 rc = __memcg_cleanup_cache_params(s);
484 mutex_lock(&slab_mutex);
485
486 return rc;
487 }
488 #else
memcg_cleanup_cache_params(struct kmem_cache * s)489 static int memcg_cleanup_cache_params(struct kmem_cache *s)
490 {
491 return 0;
492 }
493 #endif /* CONFIG_MEMCG_KMEM */
494
slab_kmem_cache_release(struct kmem_cache * s)495 void slab_kmem_cache_release(struct kmem_cache *s)
496 {
497 kfree(s->name);
498 kmem_cache_free(kmem_cache, s);
499 }
500
kmem_cache_destroy(struct kmem_cache * s)501 void kmem_cache_destroy(struct kmem_cache *s)
502 {
503 get_online_cpus();
504 get_online_mems();
505
506 mutex_lock(&slab_mutex);
507
508 s->refcount--;
509 if (s->refcount)
510 goto out_unlock;
511
512 if (memcg_cleanup_cache_params(s) != 0)
513 goto out_unlock;
514
515 if (__kmem_cache_shutdown(s) != 0) {
516 printk(KERN_ERR "kmem_cache_destroy %s: "
517 "Slab cache still has objects\n", s->name);
518 dump_stack();
519 goto out_unlock;
520 }
521
522 list_del(&s->list);
523
524 mutex_unlock(&slab_mutex);
525 if (s->flags & SLAB_DESTROY_BY_RCU)
526 rcu_barrier();
527
528 memcg_free_cache_params(s);
529 #ifdef SLAB_SUPPORTS_SYSFS
530 sysfs_slab_remove(s);
531 #else
532 slab_kmem_cache_release(s);
533 #endif
534 goto out;
535
536 out_unlock:
537 mutex_unlock(&slab_mutex);
538 out:
539 put_online_mems();
540 put_online_cpus();
541 }
542 EXPORT_SYMBOL(kmem_cache_destroy);
543
544 /**
545 * kmem_cache_shrink - Shrink a cache.
546 * @cachep: The cache to shrink.
547 *
548 * Releases as many slabs as possible for a cache.
549 * To help debugging, a zero exit status indicates all slabs were released.
550 */
kmem_cache_shrink(struct kmem_cache * cachep)551 int kmem_cache_shrink(struct kmem_cache *cachep)
552 {
553 int ret;
554
555 get_online_cpus();
556 get_online_mems();
557 ret = __kmem_cache_shrink(cachep);
558 put_online_mems();
559 put_online_cpus();
560 return ret;
561 }
562 EXPORT_SYMBOL(kmem_cache_shrink);
563
slab_is_available(void)564 int slab_is_available(void)
565 {
566 return slab_state >= UP;
567 }
568
569 #ifndef CONFIG_SLOB
570 /* Create a cache during boot when no slab services are available yet */
create_boot_cache(struct kmem_cache * s,const char * name,size_t size,unsigned long flags)571 void __init create_boot_cache(struct kmem_cache *s, const char *name, size_t size,
572 unsigned long flags)
573 {
574 int err;
575
576 s->name = name;
577 s->size = s->object_size = size;
578 s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size);
579 err = __kmem_cache_create(s, flags);
580
581 if (err)
582 panic("Creation of kmalloc slab %s size=%zu failed. Reason %d\n",
583 name, size, err);
584
585 s->refcount = -1; /* Exempt from merging for now */
586 }
587
create_kmalloc_cache(const char * name,size_t size,unsigned long flags)588 struct kmem_cache *__init create_kmalloc_cache(const char *name, size_t size,
589 unsigned long flags)
590 {
591 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
592
593 if (!s)
594 panic("Out of memory when creating slab %s\n", name);
595
596 create_boot_cache(s, name, size, flags);
597 list_add(&s->list, &slab_caches);
598 s->refcount = 1;
599 return s;
600 }
601
602 struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1];
603 EXPORT_SYMBOL(kmalloc_caches);
604
605 #ifdef CONFIG_ZONE_DMA
606 struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1];
607 EXPORT_SYMBOL(kmalloc_dma_caches);
608 #endif
609
610 /*
611 * Conversion table for small slabs sizes / 8 to the index in the
612 * kmalloc array. This is necessary for slabs < 192 since we have non power
613 * of two cache sizes there. The size of larger slabs can be determined using
614 * fls.
615 */
616 static s8 size_index[24] = {
617 3, /* 8 */
618 4, /* 16 */
619 5, /* 24 */
620 5, /* 32 */
621 6, /* 40 */
622 6, /* 48 */
623 6, /* 56 */
624 6, /* 64 */
625 1, /* 72 */
626 1, /* 80 */
627 1, /* 88 */
628 1, /* 96 */
629 7, /* 104 */
630 7, /* 112 */
631 7, /* 120 */
632 7, /* 128 */
633 2, /* 136 */
634 2, /* 144 */
635 2, /* 152 */
636 2, /* 160 */
637 2, /* 168 */
638 2, /* 176 */
639 2, /* 184 */
640 2 /* 192 */
641 };
642
size_index_elem(size_t bytes)643 static inline int size_index_elem(size_t bytes)
644 {
645 return (bytes - 1) / 8;
646 }
647
648 /*
649 * Find the kmem_cache structure that serves a given size of
650 * allocation
651 */
kmalloc_slab(size_t size,gfp_t flags)652 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
653 {
654 int index;
655
656 if (unlikely(size > KMALLOC_MAX_SIZE)) {
657 WARN_ON_ONCE(!(flags & __GFP_NOWARN));
658 return NULL;
659 }
660
661 if (size <= 192) {
662 if (!size)
663 return ZERO_SIZE_PTR;
664
665 index = size_index[size_index_elem(size)];
666 } else
667 index = fls(size - 1);
668
669 #ifdef CONFIG_ZONE_DMA
670 if (unlikely((flags & GFP_DMA)))
671 return kmalloc_dma_caches[index];
672
673 #endif
674 return kmalloc_caches[index];
675 }
676
677 /*
678 * Create the kmalloc array. Some of the regular kmalloc arrays
679 * may already have been created because they were needed to
680 * enable allocations for slab creation.
681 */
create_kmalloc_caches(unsigned long flags)682 void __init create_kmalloc_caches(unsigned long flags)
683 {
684 int i;
685
686 /*
687 * Patch up the size_index table if we have strange large alignment
688 * requirements for the kmalloc array. This is only the case for
689 * MIPS it seems. The standard arches will not generate any code here.
690 *
691 * Largest permitted alignment is 256 bytes due to the way we
692 * handle the index determination for the smaller caches.
693 *
694 * Make sure that nothing crazy happens if someone starts tinkering
695 * around with ARCH_KMALLOC_MINALIGN
696 */
697 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
698 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
699
700 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
701 int elem = size_index_elem(i);
702
703 if (elem >= ARRAY_SIZE(size_index))
704 break;
705 size_index[elem] = KMALLOC_SHIFT_LOW;
706 }
707
708 if (KMALLOC_MIN_SIZE >= 64) {
709 /*
710 * The 96 byte size cache is not used if the alignment
711 * is 64 byte.
712 */
713 for (i = 64 + 8; i <= 96; i += 8)
714 size_index[size_index_elem(i)] = 7;
715
716 }
717
718 if (KMALLOC_MIN_SIZE >= 128) {
719 /*
720 * The 192 byte sized cache is not used if the alignment
721 * is 128 byte. Redirect kmalloc to use the 256 byte cache
722 * instead.
723 */
724 for (i = 128 + 8; i <= 192; i += 8)
725 size_index[size_index_elem(i)] = 8;
726 }
727 for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
728 if (!kmalloc_caches[i]) {
729 kmalloc_caches[i] = create_kmalloc_cache(NULL,
730 1 << i, flags);
731 }
732
733 /*
734 * Caches that are not of the two-to-the-power-of size.
735 * These have to be created immediately after the
736 * earlier power of two caches
737 */
738 if (KMALLOC_MIN_SIZE <= 32 && !kmalloc_caches[1] && i == 6)
739 kmalloc_caches[1] = create_kmalloc_cache(NULL, 96, flags);
740
741 if (KMALLOC_MIN_SIZE <= 64 && !kmalloc_caches[2] && i == 7)
742 kmalloc_caches[2] = create_kmalloc_cache(NULL, 192, flags);
743 }
744
745 /* Kmalloc array is now usable */
746 slab_state = UP;
747
748 for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
749 struct kmem_cache *s = kmalloc_caches[i];
750 char *n;
751
752 if (s) {
753 n = kasprintf(GFP_NOWAIT, "kmalloc-%d", kmalloc_size(i));
754
755 BUG_ON(!n);
756 s->name = n;
757 }
758 }
759
760 #ifdef CONFIG_ZONE_DMA
761 for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
762 struct kmem_cache *s = kmalloc_caches[i];
763
764 if (s) {
765 int size = kmalloc_size(i);
766 char *n = kasprintf(GFP_NOWAIT,
767 "dma-kmalloc-%d", size);
768
769 BUG_ON(!n);
770 kmalloc_dma_caches[i] = create_kmalloc_cache(n,
771 size, SLAB_CACHE_DMA | flags);
772 }
773 }
774 #endif
775 }
776 #endif /* !CONFIG_SLOB */
777
778 /*
779 * To avoid unnecessary overhead, we pass through large allocation requests
780 * directly to the page allocator. We use __GFP_COMP, because we will need to
781 * know the allocation order to free the pages properly in kfree.
782 */
kmalloc_order(size_t size,gfp_t flags,unsigned int order)783 void *kmalloc_order(size_t size, gfp_t flags, unsigned int order)
784 {
785 void *ret;
786 struct page *page;
787
788 flags |= __GFP_COMP;
789 page = alloc_kmem_pages(flags, order);
790 ret = page ? page_address(page) : NULL;
791 kmemleak_alloc(ret, size, 1, flags);
792 return ret;
793 }
794 EXPORT_SYMBOL(kmalloc_order);
795
796 #ifdef CONFIG_TRACING
kmalloc_order_trace(size_t size,gfp_t flags,unsigned int order)797 void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
798 {
799 void *ret = kmalloc_order(size, flags, order);
800 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
801 return ret;
802 }
803 EXPORT_SYMBOL(kmalloc_order_trace);
804 #endif
805
806 #ifdef CONFIG_SLABINFO
807
808 #ifdef CONFIG_SLAB
809 #define SLABINFO_RIGHTS (S_IWUSR | S_IRUSR)
810 #else
811 #define SLABINFO_RIGHTS S_IRUSR
812 #endif
813
print_slabinfo_header(struct seq_file * m)814 void print_slabinfo_header(struct seq_file *m)
815 {
816 /*
817 * Output format version, so at least we can change it
818 * without _too_ many complaints.
819 */
820 #ifdef CONFIG_DEBUG_SLAB
821 seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
822 #else
823 seq_puts(m, "slabinfo - version: 2.1\n");
824 #endif
825 seq_puts(m, "# name <active_objs> <num_objs> <objsize> "
826 "<objperslab> <pagesperslab>");
827 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
828 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
829 #ifdef CONFIG_DEBUG_SLAB
830 seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> "
831 "<error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
832 seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
833 #endif
834 seq_putc(m, '\n');
835 }
836
s_start(struct seq_file * m,loff_t * pos)837 static void *s_start(struct seq_file *m, loff_t *pos)
838 {
839 loff_t n = *pos;
840
841 mutex_lock(&slab_mutex);
842 if (!n)
843 print_slabinfo_header(m);
844
845 return seq_list_start(&slab_caches, *pos);
846 }
847
slab_next(struct seq_file * m,void * p,loff_t * pos)848 void *slab_next(struct seq_file *m, void *p, loff_t *pos)
849 {
850 return seq_list_next(p, &slab_caches, pos);
851 }
852
slab_stop(struct seq_file * m,void * p)853 void slab_stop(struct seq_file *m, void *p)
854 {
855 mutex_unlock(&slab_mutex);
856 }
857
858 static void
memcg_accumulate_slabinfo(struct kmem_cache * s,struct slabinfo * info)859 memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info)
860 {
861 struct kmem_cache *c;
862 struct slabinfo sinfo;
863 int i;
864
865 if (!is_root_cache(s))
866 return;
867
868 for_each_memcg_cache_index(i) {
869 c = cache_from_memcg_idx(s, i);
870 if (!c)
871 continue;
872
873 memset(&sinfo, 0, sizeof(sinfo));
874 get_slabinfo(c, &sinfo);
875
876 info->active_slabs += sinfo.active_slabs;
877 info->num_slabs += sinfo.num_slabs;
878 info->shared_avail += sinfo.shared_avail;
879 info->active_objs += sinfo.active_objs;
880 info->num_objs += sinfo.num_objs;
881 }
882 }
883
cache_show(struct kmem_cache * s,struct seq_file * m)884 int cache_show(struct kmem_cache *s, struct seq_file *m)
885 {
886 struct slabinfo sinfo;
887
888 memset(&sinfo, 0, sizeof(sinfo));
889 get_slabinfo(s, &sinfo);
890
891 memcg_accumulate_slabinfo(s, &sinfo);
892
893 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
894 cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size,
895 sinfo.objects_per_slab, (1 << sinfo.cache_order));
896
897 seq_printf(m, " : tunables %4u %4u %4u",
898 sinfo.limit, sinfo.batchcount, sinfo.shared);
899 seq_printf(m, " : slabdata %6lu %6lu %6lu",
900 sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
901 slabinfo_show_stats(m, s);
902 seq_putc(m, '\n');
903 return 0;
904 }
905
s_show(struct seq_file * m,void * p)906 static int s_show(struct seq_file *m, void *p)
907 {
908 struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
909
910 if (!is_root_cache(s))
911 return 0;
912 return cache_show(s, m);
913 }
914
915 /*
916 * slabinfo_op - iterator that generates /proc/slabinfo
917 *
918 * Output layout:
919 * cache-name
920 * num-active-objs
921 * total-objs
922 * object size
923 * num-active-slabs
924 * total-slabs
925 * num-pages-per-slab
926 * + further values on SMP and with statistics enabled
927 */
928 static const struct seq_operations slabinfo_op = {
929 .start = s_start,
930 .next = slab_next,
931 .stop = slab_stop,
932 .show = s_show,
933 };
934
slabinfo_open(struct inode * inode,struct file * file)935 static int slabinfo_open(struct inode *inode, struct file *file)
936 {
937 return seq_open(file, &slabinfo_op);
938 }
939
940 static const struct file_operations proc_slabinfo_operations = {
941 .open = slabinfo_open,
942 .read = seq_read,
943 .write = slabinfo_write,
944 .llseek = seq_lseek,
945 .release = seq_release,
946 };
947
slab_proc_init(void)948 static int __init slab_proc_init(void)
949 {
950 proc_create("slabinfo", SLABINFO_RIGHTS, NULL,
951 &proc_slabinfo_operations);
952 return 0;
953 }
954 module_init(slab_proc_init);
955 #endif /* CONFIG_SLABINFO */
956
__do_krealloc(const void * p,size_t new_size,gfp_t flags)957 static __always_inline void *__do_krealloc(const void *p, size_t new_size,
958 gfp_t flags)
959 {
960 void *ret;
961 size_t ks = 0;
962
963 if (p)
964 ks = ksize(p);
965
966 if (ks >= new_size)
967 return (void *)p;
968
969 ret = kmalloc_track_caller(new_size, flags);
970 if (ret && p)
971 memcpy(ret, p, ks);
972
973 return ret;
974 }
975
976 /**
977 * __krealloc - like krealloc() but don't free @p.
978 * @p: object to reallocate memory for.
979 * @new_size: how many bytes of memory are required.
980 * @flags: the type of memory to allocate.
981 *
982 * This function is like krealloc() except it never frees the originally
983 * allocated buffer. Use this if you don't want to free the buffer immediately
984 * like, for example, with RCU.
985 */
__krealloc(const void * p,size_t new_size,gfp_t flags)986 void *__krealloc(const void *p, size_t new_size, gfp_t flags)
987 {
988 if (unlikely(!new_size))
989 return ZERO_SIZE_PTR;
990
991 return __do_krealloc(p, new_size, flags);
992
993 }
994 EXPORT_SYMBOL(__krealloc);
995
996 /**
997 * krealloc - reallocate memory. The contents will remain unchanged.
998 * @p: object to reallocate memory for.
999 * @new_size: how many bytes of memory are required.
1000 * @flags: the type of memory to allocate.
1001 *
1002 * The contents of the object pointed to are preserved up to the
1003 * lesser of the new and old sizes. If @p is %NULL, krealloc()
1004 * behaves exactly like kmalloc(). If @new_size is 0 and @p is not a
1005 * %NULL pointer, the object pointed to is freed.
1006 */
krealloc(const void * p,size_t new_size,gfp_t flags)1007 void *krealloc(const void *p, size_t new_size, gfp_t flags)
1008 {
1009 void *ret;
1010
1011 if (unlikely(!new_size)) {
1012 kfree(p);
1013 return ZERO_SIZE_PTR;
1014 }
1015
1016 ret = __do_krealloc(p, new_size, flags);
1017 if (ret && p != ret)
1018 kfree(p);
1019
1020 return ret;
1021 }
1022 EXPORT_SYMBOL(krealloc);
1023
1024 /**
1025 * kzfree - like kfree but zero memory
1026 * @p: object to free memory of
1027 *
1028 * The memory of the object @p points to is zeroed before freed.
1029 * If @p is %NULL, kzfree() does nothing.
1030 *
1031 * Note: this function zeroes the whole allocated buffer which can be a good
1032 * deal bigger than the requested buffer size passed to kmalloc(). So be
1033 * careful when using this function in performance sensitive code.
1034 */
kzfree(const void * p)1035 void kzfree(const void *p)
1036 {
1037 size_t ks;
1038 void *mem = (void *)p;
1039
1040 if (unlikely(ZERO_OR_NULL_PTR(mem)))
1041 return;
1042 ks = ksize(mem);
1043 memset(mem, 0, ks);
1044 kfree(mem);
1045 }
1046 EXPORT_SYMBOL(kzfree);
1047
1048 /* Tracepoints definitions. */
1049 EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1050 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1051 EXPORT_TRACEPOINT_SYMBOL(kmalloc_node);
1052 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node);
1053 EXPORT_TRACEPOINT_SYMBOL(kfree);
1054 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);
1055