1 // SPDX-License-Identifier: GPL-2.0
3 * Slab allocator functions that are independent of the allocator strategy
7 #include <linux/slab.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>
24 #include <linux/memcontrol.h>
26 #define CREATE_TRACE_POINTS
27 #include <trace/events/kmem.h>
33 enum slab_state slab_state;
34 LIST_HEAD(slab_caches);
35 DEFINE_MUTEX(slab_mutex);
36 struct kmem_cache *kmem_cache;
38 #ifdef CONFIG_HARDENED_USERCOPY
39 bool usercopy_fallback __ro_after_init =
40 IS_ENABLED(CONFIG_HARDENED_USERCOPY_FALLBACK);
41 module_param(usercopy_fallback, bool, 0400);
42 MODULE_PARM_DESC(usercopy_fallback,
43 "WARN instead of reject usercopy whitelist violations");
46 static LIST_HEAD(slab_caches_to_rcu_destroy);
47 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work);
48 static DECLARE_WORK(slab_caches_to_rcu_destroy_work,
49 slab_caches_to_rcu_destroy_workfn);
52 * Set of flags that will prevent slab merging
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)
58 #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \
59 SLAB_CACHE_DMA32 | SLAB_ACCOUNT)
62 * Merge control. If this is set then no merging of slab caches will occur.
64 static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT);
66 static int __init setup_slab_nomerge(char *str)
73 __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0);
76 __setup("slab_nomerge", setup_slab_nomerge);
79 * Determine the size of a slab object
81 unsigned int kmem_cache_size(struct kmem_cache *s)
83 return s->object_size;
85 EXPORT_SYMBOL(kmem_cache_size);
87 #ifdef CONFIG_DEBUG_VM
88 static int kmem_cache_sanity_check(const char *name, unsigned int size)
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);
96 WARN_ON(strchr(name, ' ')); /* It confuses parsers */
100 static inline int kmem_cache_sanity_check(const char *name, unsigned int size)
106 void __kmem_cache_free_bulk(struct kmem_cache *s, size_t nr, void **p)
110 for (i = 0; i < nr; i++) {
112 kmem_cache_free(s, p[i]);
118 int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t nr,
123 for (i = 0; i < nr; i++) {
124 void *x = p[i] = kmem_cache_alloc(s, flags);
126 __kmem_cache_free_bulk(s, i, p);
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.
137 static unsigned int calculate_alignment(slab_flags_t flags,
138 unsigned int align, unsigned int size)
141 * If the user wants hardware cache aligned objects then follow that
142 * suggestion if the object is sufficiently large.
144 * The hardware cache alignment cannot override the specified
145 * alignment though. If that is greater then use it.
147 if (flags & SLAB_HWCACHE_ALIGN) {
150 ralign = cache_line_size();
151 while (size <= ralign / 2)
153 align = max(align, ralign);
156 if (align < ARCH_SLAB_MINALIGN)
157 align = ARCH_SLAB_MINALIGN;
159 return ALIGN(align, sizeof(void *));
163 * Find a mergeable slab cache
165 int slab_unmergeable(struct kmem_cache *s)
167 if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
177 * We may have set a slab to be unmergeable during bootstrap.
185 struct kmem_cache *find_mergeable(unsigned int size, unsigned int align,
186 slab_flags_t flags, const char *name, void (*ctor)(void *))
188 struct kmem_cache *s;
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);
201 if (flags & SLAB_NEVER_MERGE)
204 list_for_each_entry_reverse(s, &slab_caches, list) {
205 if (slab_unmergeable(s))
211 if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
214 * Check if alignment is compatible.
215 * Courtesy of Adrian Drzewiecki
217 if ((s->size & ~(align - 1)) != s->size)
220 if (s->size - size >= sizeof(void *))
223 if (IS_ENABLED(CONFIG_SLAB) && align &&
224 (align > s->align || s->align % align))
232 static 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)
238 struct kmem_cache *s;
241 if (WARN_ON(useroffset + usersize > object_size))
242 useroffset = usersize = 0;
245 s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
250 s->size = s->object_size = object_size;
253 s->useroffset = useroffset;
254 s->usersize = usersize;
256 err = __kmem_cache_create(s, flags);
261 list_add(&s->list, &slab_caches);
268 kmem_cache_free(kmem_cache, s);
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.
279 * @useroffset: Usercopy region offset
280 * @usersize: Usercopy region size
281 * @ctor: A constructor for the objects.
283 * Cannot be called within a interrupt, but can be interrupted.
284 * The @ctor is run when new pages are allocated by the cache.
288 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
289 * to catch references to uninitialised memory.
291 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
292 * for buffer overruns.
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
298 * Return: a pointer to the cache on success, NULL on failure.
301 kmem_cache_create_usercopy(const char *name,
302 unsigned int size, unsigned int align,
304 unsigned int useroffset, unsigned int usersize,
305 void (*ctor)(void *))
307 struct kmem_cache *s = NULL;
308 const char *cache_name;
314 mutex_lock(&slab_mutex);
316 err = kmem_cache_sanity_check(name, size);
321 /* Refuse requests with allocator specific flags */
322 if (flags & ~SLAB_FLAGS_PERMITTED) {
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
333 flags &= CACHE_CREATE_MASK;
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;
341 s = __kmem_cache_alias(name, size, align, flags, ctor);
345 cache_name = kstrdup_const(name, GFP_KERNEL);
351 s = create_cache(cache_name, size,
352 calculate_alignment(flags, align, size),
353 flags, useroffset, usersize, ctor, NULL);
356 kfree_const(cache_name);
360 mutex_unlock(&slab_mutex);
366 if (flags & SLAB_PANIC)
367 panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n",
370 pr_warn("kmem_cache_create(%s) failed with error %d\n",
378 EXPORT_SYMBOL(kmem_cache_create_usercopy);
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.
386 * @ctor: A constructor for the objects.
388 * Cannot be called within a interrupt, but can be interrupted.
389 * The @ctor is run when new pages are allocated by the cache.
393 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
394 * to catch references to uninitialised memory.
396 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
397 * for buffer overruns.
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
403 * Return: a pointer to the cache on success, NULL on failure.
406 kmem_cache_create(const char *name, unsigned int size, unsigned int align,
407 slab_flags_t flags, void (*ctor)(void *))
409 return kmem_cache_create_usercopy(name, size, align, flags, 0, 0,
412 EXPORT_SYMBOL(kmem_cache_create);
414 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work)
416 LIST_HEAD(to_destroy);
417 struct kmem_cache *s, *s2;
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
428 mutex_lock(&slab_mutex);
429 list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy);
430 mutex_unlock(&slab_mutex);
432 if (list_empty(&to_destroy))
437 list_for_each_entry_safe(s, s2, &to_destroy, list) {
438 #ifdef SLAB_SUPPORTS_SYSFS
439 sysfs_slab_release(s);
441 slab_kmem_cache_release(s);
446 static int shutdown_cache(struct kmem_cache *s)
448 /* free asan quarantined objects */
449 kasan_cache_shutdown(s);
451 if (__kmem_cache_shutdown(s) != 0)
456 if (s->flags & SLAB_TYPESAFE_BY_RCU) {
457 #ifdef SLAB_SUPPORTS_SYSFS
458 sysfs_slab_unlink(s);
460 list_add_tail(&s->list, &slab_caches_to_rcu_destroy);
461 schedule_work(&slab_caches_to_rcu_destroy_work);
463 #ifdef SLAB_SUPPORTS_SYSFS
464 sysfs_slab_unlink(s);
465 sysfs_slab_release(s);
467 slab_kmem_cache_release(s);
474 void slab_kmem_cache_release(struct kmem_cache *s)
476 __kmem_cache_release(s);
477 kfree_const(s->name);
478 kmem_cache_free(kmem_cache, s);
481 void kmem_cache_destroy(struct kmem_cache *s)
491 mutex_lock(&slab_mutex);
497 err = shutdown_cache(s);
499 pr_err("kmem_cache_destroy %s: Slab cache still has objects\n",
504 mutex_unlock(&slab_mutex);
509 EXPORT_SYMBOL(kmem_cache_destroy);
512 * kmem_cache_shrink - Shrink a cache.
513 * @cachep: The cache to shrink.
515 * Releases as many slabs as possible for a cache.
516 * To help debugging, a zero exit status indicates all slabs were released.
518 * Return: %0 if all slabs were released, non-zero otherwise
520 int kmem_cache_shrink(struct kmem_cache *cachep)
526 kasan_cache_shrink(cachep);
527 ret = __kmem_cache_shrink(cachep);
532 EXPORT_SYMBOL(kmem_cache_shrink);
534 bool slab_is_available(void)
536 return slab_state >= UP;
540 /* Create a cache during boot when no slab services are available yet */
541 void __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)
546 unsigned int align = ARCH_KMALLOC_MINALIGN;
549 s->size = s->object_size = size;
552 * For power of two sizes, guarantee natural alignment for kmalloc
553 * caches, regardless of SL*B debugging options.
555 if (is_power_of_2(size))
556 align = max(align, size);
557 s->align = calculate_alignment(flags, align, size);
559 s->useroffset = useroffset;
560 s->usersize = usersize;
562 err = __kmem_cache_create(s, flags);
565 panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n",
568 s->refcount = -1; /* Exempt from merging for now */
571 struct kmem_cache *__init create_kmalloc_cache(const char *name,
572 unsigned int size, slab_flags_t flags,
573 unsigned int useroffset, unsigned int usersize)
575 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
578 panic("Out of memory when creating slab %s\n", name);
580 create_boot_cache(s, name, size, flags, useroffset, usersize);
581 list_add(&s->list, &slab_caches);
587 kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1] __ro_after_init =
588 { /* initialization for https://bugs.llvm.org/show_bug.cgi?id=42570 */ };
589 EXPORT_SYMBOL(kmalloc_caches);
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
597 static u8 size_index[24] __ro_after_init = {
624 static inline unsigned int size_index_elem(unsigned int bytes)
626 return (bytes - 1) / 8;
630 * Find the kmem_cache structure that serves a given size of
633 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
639 return ZERO_SIZE_PTR;
641 index = size_index[size_index_elem(size)];
643 if (WARN_ON_ONCE(size > KMALLOC_MAX_CACHE_SIZE))
645 index = fls(size - 1);
648 return kmalloc_caches[kmalloc_type(flags)][index];
651 #ifdef CONFIG_ZONE_DMA
652 #define INIT_KMALLOC_INFO(__size, __short_size) \
654 .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \
655 .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #__short_size, \
656 .name[KMALLOC_DMA] = "dma-kmalloc-" #__short_size, \
660 #define INIT_KMALLOC_INFO(__size, __short_size) \
662 .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \
663 .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #__short_size, \
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
673 const 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)
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.
708 * Largest permitted alignment is 256 bytes due to the way we
709 * handle the index determination for the smaller caches.
711 * Make sure that nothing crazy happens if someone starts tinkering
712 * around with ARCH_KMALLOC_MINALIGN
714 void __init setup_kmalloc_cache_index_table(void)
718 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
719 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
721 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
722 unsigned int elem = size_index_elem(i);
724 if (elem >= ARRAY_SIZE(size_index))
726 size_index[elem] = KMALLOC_SHIFT_LOW;
729 if (KMALLOC_MIN_SIZE >= 64) {
731 * The 96 byte size cache is not used if the alignment
734 for (i = 64 + 8; i <= 96; i += 8)
735 size_index[size_index_elem(i)] = 7;
739 if (KMALLOC_MIN_SIZE >= 128) {
741 * The 192 byte sized cache is not used if the alignment
742 * is 128 byte. Redirect kmalloc to use the 256 byte cache
745 for (i = 128 + 8; i <= 192; i += 8)
746 size_index[size_index_elem(i)] = 8;
751 new_kmalloc_cache(int idx, enum kmalloc_cache_type type, slab_flags_t flags)
753 if (type == KMALLOC_RECLAIM)
754 flags |= SLAB_RECLAIM_ACCOUNT;
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);
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.
767 void __init create_kmalloc_caches(slab_flags_t flags)
770 enum kmalloc_cache_type type;
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);
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
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);
791 /* Kmalloc array is now usable */
794 #ifdef CONFIG_ZONE_DMA
795 for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
796 struct kmem_cache *s = kmalloc_caches[KMALLOC_NORMAL][i];
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);
808 #endif /* !CONFIG_SLOB */
810 gfp_t kmalloc_fix_flags(gfp_t flags)
812 gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
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);
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.
827 void *kmalloc_order(size_t size, gfp_t flags, unsigned int order)
832 if (unlikely(flags & GFP_SLAB_BUG_MASK))
833 flags = kmalloc_fix_flags(flags);
836 page = alloc_pages(flags, order);
838 ret = page_address(page);
839 mod_node_page_state(page_pgdat(page), NR_SLAB_UNRECLAIMABLE_B,
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);
847 EXPORT_SYMBOL(kmalloc_order);
849 #ifdef CONFIG_TRACING
850 void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
852 void *ret = kmalloc_order(size, flags, order);
853 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
856 EXPORT_SYMBOL(kmalloc_order_trace);
859 #ifdef CONFIG_SLAB_FREELIST_RANDOM
860 /* Randomize a generic freelist */
861 static void freelist_randomize(struct rnd_state *state, unsigned int *list,
867 for (i = 0; i < count; i++)
870 /* Fisher-Yates shuffle */
871 for (i = count - 1; i > 0; i--) {
872 rand = prandom_u32_state(state);
874 swap(list[i], list[rand]);
878 /* Create a random sequence per cache */
879 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
882 struct rnd_state state;
884 if (count < 2 || cachep->random_seq)
887 cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp);
888 if (!cachep->random_seq)
891 /* Get best entropy at this stage of boot */
892 prandom_seed_state(&state, get_random_long());
894 freelist_randomize(&state, cachep->random_seq, count);
898 /* Destroy the per-cache random freelist sequence */
899 void cache_random_seq_destroy(struct kmem_cache *cachep)
901 kfree(cachep->random_seq);
902 cachep->random_seq = NULL;
904 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
906 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
908 #define SLABINFO_RIGHTS (0600)
910 #define SLABINFO_RIGHTS (0400)
913 static void print_slabinfo_header(struct seq_file *m)
916 * Output format version, so at least we can change it
917 * without _too_ many complaints.
919 #ifdef CONFIG_DEBUG_SLAB
920 seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
922 seq_puts(m, "slabinfo - version: 2.1\n");
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>");
934 void *slab_start(struct seq_file *m, loff_t *pos)
936 mutex_lock(&slab_mutex);
937 return seq_list_start(&slab_caches, *pos);
940 void *slab_next(struct seq_file *m, void *p, loff_t *pos)
942 return seq_list_next(p, &slab_caches, pos);
945 void slab_stop(struct seq_file *m, void *p)
947 mutex_unlock(&slab_mutex);
950 static void cache_show(struct kmem_cache *s, struct seq_file *m)
952 struct slabinfo sinfo;
954 memset(&sinfo, 0, sizeof(sinfo));
955 get_slabinfo(s, &sinfo);
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));
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);
969 static int slab_show(struct seq_file *m, void *p)
971 struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
973 if (p == slab_caches.next)
974 print_slabinfo_header(m);
979 void dump_unreclaimable_slab(void)
981 struct kmem_cache *s;
982 struct slabinfo sinfo;
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
988 * Use mutex_trylock to protect the list traverse, dump nothing
989 * without acquiring the mutex.
991 if (!mutex_trylock(&slab_mutex)) {
992 pr_warn("excessive unreclaimable slab but cannot dump stats\n");
996 pr_info("Unreclaimable slab info:\n");
997 pr_info("Name Used Total\n");
999 list_for_each_entry(s, &slab_caches, list) {
1000 if (s->flags & SLAB_RECLAIM_ACCOUNT)
1003 get_slabinfo(s, &sinfo);
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);
1010 mutex_unlock(&slab_mutex);
1013 #if defined(CONFIG_MEMCG_KMEM)
1014 int memcg_slab_show(struct seq_file *m, void *p)
1018 * Please, take a look at tools/cgroup/slabinfo.py .
1025 * slabinfo_op - iterator that generates /proc/slabinfo
1034 * num-pages-per-slab
1035 * + further values on SMP and with statistics enabled
1037 static const struct seq_operations slabinfo_op = {
1038 .start = slab_start,
1044 static int slabinfo_open(struct inode *inode, struct file *file)
1046 return seq_open(file, &slabinfo_op);
1049 static 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,
1058 static int __init slab_proc_init(void)
1060 proc_create("slabinfo", SLABINFO_RIGHTS, NULL, &slabinfo_proc_ops);
1063 module_init(slab_proc_init);
1065 #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */
1067 static __always_inline void *__do_krealloc(const void *p, size_t new_size,
1075 if (ks >= new_size) {
1076 p = kasan_krealloc((void *)p, new_size, flags);
1080 ret = kmalloc_track_caller(new_size, flags);
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.
1093 * The contents of the object pointed to are preserved up to the
1094 * lesser of the new and old sizes (__GFP_ZERO flag is effectively ignored).
1095 * If @p is %NULL, krealloc() behaves exactly like kmalloc(). If @new_size
1096 * is 0 and @p is not a %NULL pointer, the object pointed to is freed.
1098 * Return: pointer to the allocated memory or %NULL in case of error
1100 void *krealloc(const void *p, size_t new_size, gfp_t flags)
1104 if (unlikely(!new_size)) {
1106 return ZERO_SIZE_PTR;
1109 ret = __do_krealloc(p, new_size, flags);
1110 if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret))
1115 EXPORT_SYMBOL(krealloc);
1118 * kfree_sensitive - Clear sensitive information in memory before freeing
1119 * @p: object to free memory of
1121 * The memory of the object @p points to is zeroed before freed.
1122 * If @p is %NULL, kfree_sensitive() does nothing.
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.
1128 void kfree_sensitive(const void *p)
1131 void *mem = (void *)p;
1135 memzero_explicit(mem, ks);
1138 EXPORT_SYMBOL(kfree_sensitive);
1141 * ksize - get the actual amount of memory allocated for a given object
1142 * @objp: Pointer to the object
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.
1152 * Return: size of the actual memory used by @objp in bytes
1154 size_t ksize(const void *objp)
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).
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.
1168 * We want to perform the check before __ksize(), to avoid potentially
1169 * crashing in __ksize() due to accessing invalid metadata.
1171 if (unlikely(ZERO_OR_NULL_PTR(objp)) || !__kasan_check_read(objp, 1))
1174 size = __ksize(objp);
1176 * We assume that ksize callers could use whole allocated area,
1177 * so we need to unpoison this area.
1179 kasan_unpoison_range(objp, size);
1182 EXPORT_SYMBOL(ksize);
1184 /* Tracepoints definitions. */
1185 EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1186 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1187 EXPORT_TRACEPOINT_SYMBOL(kmalloc_node);
1188 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node);
1189 EXPORT_TRACEPOINT_SYMBOL(kfree);
1190 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);
1192 int should_failslab(struct kmem_cache *s, gfp_t gfpflags)
1194 if (__should_failslab(s, gfpflags))
1198 ALLOW_ERROR_INJECTION(should_failslab, ERRNO);