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/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>
26 #include <linux/memcontrol.h>
28 #define CREATE_TRACE_POINTS
29 #include <trace/events/kmem.h>
35 enum slab_state slab_state;
36 LIST_HEAD(slab_caches);
37 DEFINE_MUTEX(slab_mutex);
38 struct kmem_cache *kmem_cache;
40 static LIST_HEAD(slab_caches_to_rcu_destroy);
41 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work);
42 static DECLARE_WORK(slab_caches_to_rcu_destroy_work,
43 slab_caches_to_rcu_destroy_workfn);
46 * Set of flags that will prevent slab merging
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())
52 #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \
53 SLAB_CACHE_DMA32 | SLAB_ACCOUNT)
56 * Merge control. If this is set then no merging of slab caches will occur.
58 static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT);
60 static int __init setup_slab_nomerge(char *str)
66 static int __init setup_slab_merge(char *str)
73 __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0);
74 __setup_param("slub_merge", slub_merge, setup_slab_merge, 0);
77 __setup("slab_nomerge", setup_slab_nomerge);
78 __setup("slab_merge", setup_slab_merge);
81 * Determine the size of a slab object
83 unsigned int kmem_cache_size(struct kmem_cache *s)
85 return s->object_size;
87 EXPORT_SYMBOL(kmem_cache_size);
89 #ifdef CONFIG_DEBUG_VM
90 static int kmem_cache_sanity_check(const char *name, unsigned int size)
92 if (!name || in_interrupt() || size > KMALLOC_MAX_SIZE) {
93 pr_err("kmem_cache_create(%s) integrity check failed\n", name);
97 WARN_ON(strchr(name, ' ')); /* It confuses parsers */
101 static inline int kmem_cache_sanity_check(const char *name, unsigned int size)
107 void __kmem_cache_free_bulk(struct kmem_cache *s, size_t nr, void **p)
111 for (i = 0; i < nr; i++) {
113 kmem_cache_free(s, p[i]);
119 int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t nr,
124 for (i = 0; i < nr; i++) {
125 void *x = p[i] = kmem_cache_alloc(s, flags);
127 __kmem_cache_free_bulk(s, i, p);
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.
138 static unsigned int calculate_alignment(slab_flags_t flags,
139 unsigned int align, unsigned int size)
142 * If the user wants hardware cache aligned objects then follow that
143 * suggestion if the object is sufficiently large.
145 * The hardware cache alignment cannot override the specified
146 * alignment though. If that is greater then use it.
148 if (flags & SLAB_HWCACHE_ALIGN) {
151 ralign = cache_line_size();
152 while (size <= ralign / 2)
154 align = max(align, ralign);
157 if (align < ARCH_SLAB_MINALIGN)
158 align = ARCH_SLAB_MINALIGN;
160 return ALIGN(align, sizeof(void *));
164 * Find a mergeable slab cache
166 int slab_unmergeable(struct kmem_cache *s)
168 if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
178 * We may have set a slab to be unmergeable during bootstrap.
186 struct kmem_cache *find_mergeable(unsigned int size, unsigned int align,
187 slab_flags_t flags, const char *name, void (*ctor)(void *))
189 struct kmem_cache *s;
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);
202 if (flags & SLAB_NEVER_MERGE)
205 list_for_each_entry_reverse(s, &slab_caches, list) {
206 if (slab_unmergeable(s))
212 if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
215 * Check if alignment is compatible.
216 * Courtesy of Adrian Drzewiecki
218 if ((s->size & ~(align - 1)) != s->size)
221 if (s->size - size >= sizeof(void *))
224 if (IS_ENABLED(CONFIG_SLAB) && align &&
225 (align > s->align || s->align % align))
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)
239 struct kmem_cache *s;
242 if (WARN_ON(useroffset + usersize > object_size))
243 useroffset = usersize = 0;
246 s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
251 s->size = s->object_size = object_size;
254 s->useroffset = useroffset;
255 s->usersize = usersize;
257 err = __kmem_cache_create(s, flags);
262 list_add(&s->list, &slab_caches);
269 kmem_cache_free(kmem_cache, s);
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.
280 * @useroffset: Usercopy region offset
281 * @usersize: Usercopy region size
282 * @ctor: A constructor for the objects.
284 * Cannot be called within a interrupt, but can be interrupted.
285 * The @ctor is run when new pages are allocated by the cache.
289 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
290 * to catch references to uninitialised memory.
292 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
293 * for buffer overruns.
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
299 * Return: a pointer to the cache on success, NULL on failure.
302 kmem_cache_create_usercopy(const char *name,
303 unsigned int size, unsigned int align,
305 unsigned int useroffset, unsigned int usersize,
306 void (*ctor)(void *))
308 struct kmem_cache *s = NULL;
309 const char *cache_name;
312 #ifdef CONFIG_SLUB_DEBUG
314 * If no slub_debug was enabled globally, the static key is not yet
315 * enabled by setup_slub_debug(). Enable it if the cache is being
316 * created with any of the debugging flags passed explicitly.
318 if (flags & SLAB_DEBUG_FLAGS)
319 static_branch_enable(&slub_debug_enabled);
322 mutex_lock(&slab_mutex);
324 err = kmem_cache_sanity_check(name, size);
329 /* Refuse requests with allocator specific flags */
330 if (flags & ~SLAB_FLAGS_PERMITTED) {
336 * Some allocators will constraint the set of valid flags to a subset
337 * of all flags. We expect them to define CACHE_CREATE_MASK in this
338 * case, and we'll just provide them with a sanitized version of the
341 flags &= CACHE_CREATE_MASK;
343 /* Fail closed on bad usersize of useroffset values. */
344 if (WARN_ON(!usersize && useroffset) ||
345 WARN_ON(size < usersize || size - usersize < useroffset))
346 usersize = useroffset = 0;
349 s = __kmem_cache_alias(name, size, align, flags, ctor);
353 cache_name = kstrdup_const(name, GFP_KERNEL);
359 s = create_cache(cache_name, size,
360 calculate_alignment(flags, align, size),
361 flags, useroffset, usersize, ctor, NULL);
364 kfree_const(cache_name);
368 mutex_unlock(&slab_mutex);
371 if (flags & SLAB_PANIC)
372 panic("%s: Failed to create slab '%s'. Error %d\n",
373 __func__, name, err);
375 pr_warn("%s(%s) failed with error %d\n",
376 __func__, name, err);
383 EXPORT_SYMBOL(kmem_cache_create_usercopy);
386 * kmem_cache_create - Create a cache.
387 * @name: A string which is used in /proc/slabinfo to identify this cache.
388 * @size: The size of objects to be created in this cache.
389 * @align: The required alignment for the objects.
391 * @ctor: A constructor for the objects.
393 * Cannot be called within a interrupt, but can be interrupted.
394 * The @ctor is run when new pages are allocated by the cache.
398 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
399 * to catch references to uninitialised memory.
401 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
402 * for buffer overruns.
404 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
405 * cacheline. This can be beneficial if you're counting cycles as closely
408 * Return: a pointer to the cache on success, NULL on failure.
411 kmem_cache_create(const char *name, unsigned int size, unsigned int align,
412 slab_flags_t flags, void (*ctor)(void *))
414 return kmem_cache_create_usercopy(name, size, align, flags, 0, 0,
417 EXPORT_SYMBOL(kmem_cache_create);
419 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work)
421 LIST_HEAD(to_destroy);
422 struct kmem_cache *s, *s2;
425 * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the
426 * @slab_caches_to_rcu_destroy list. The slab pages are freed
427 * through RCU and the associated kmem_cache are dereferenced
428 * while freeing the pages, so the kmem_caches should be freed only
429 * after the pending RCU operations are finished. As rcu_barrier()
430 * is a pretty slow operation, we batch all pending destructions
433 mutex_lock(&slab_mutex);
434 list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy);
435 mutex_unlock(&slab_mutex);
437 if (list_empty(&to_destroy))
442 list_for_each_entry_safe(s, s2, &to_destroy, list) {
443 debugfs_slab_release(s);
444 kfence_shutdown_cache(s);
445 #ifdef SLAB_SUPPORTS_SYSFS
446 sysfs_slab_release(s);
448 slab_kmem_cache_release(s);
453 static int shutdown_cache(struct kmem_cache *s)
455 /* free asan quarantined objects */
456 kasan_cache_shutdown(s);
458 if (__kmem_cache_shutdown(s) != 0)
463 if (s->flags & SLAB_TYPESAFE_BY_RCU) {
464 #ifdef SLAB_SUPPORTS_SYSFS
465 sysfs_slab_unlink(s);
467 list_add_tail(&s->list, &slab_caches_to_rcu_destroy);
468 schedule_work(&slab_caches_to_rcu_destroy_work);
470 kfence_shutdown_cache(s);
471 debugfs_slab_release(s);
472 #ifdef SLAB_SUPPORTS_SYSFS
473 sysfs_slab_unlink(s);
474 sysfs_slab_release(s);
476 slab_kmem_cache_release(s);
483 void slab_kmem_cache_release(struct kmem_cache *s)
485 __kmem_cache_release(s);
486 kfree_const(s->name);
487 kmem_cache_free(kmem_cache, s);
490 void kmem_cache_destroy(struct kmem_cache *s)
492 if (unlikely(!s) || !kasan_check_byte(s))
496 mutex_lock(&slab_mutex);
502 WARN(shutdown_cache(s),
503 "%s %s: Slab cache still has objects when called from %pS",
504 __func__, s->name, (void *)_RET_IP_);
506 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)
525 kasan_cache_shrink(cachep);
526 ret = __kmem_cache_shrink(cachep);
530 EXPORT_SYMBOL(kmem_cache_shrink);
532 bool slab_is_available(void)
534 return slab_state >= UP;
539 * kmem_valid_obj - does the pointer reference a valid slab object?
540 * @object: pointer to query.
542 * Return: %true if the pointer is to a not-yet-freed object from
543 * kmalloc() or kmem_cache_alloc(), either %true or %false if the pointer
544 * is to an already-freed object, and %false otherwise.
546 bool kmem_valid_obj(void *object)
550 /* Some arches consider ZERO_SIZE_PTR to be a valid address. */
551 if (object < (void *)PAGE_SIZE || !virt_addr_valid(object))
553 folio = virt_to_folio(object);
554 return folio_test_slab(folio);
556 EXPORT_SYMBOL_GPL(kmem_valid_obj);
558 static void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab)
560 if (__kfence_obj_info(kpp, object, slab))
562 __kmem_obj_info(kpp, object, slab);
566 * kmem_dump_obj - Print available slab provenance information
567 * @object: slab object for which to find provenance information.
569 * This function uses pr_cont(), so that the caller is expected to have
570 * printed out whatever preamble is appropriate. The provenance information
571 * depends on the type of object and on how much debugging is enabled.
572 * For a slab-cache object, the fact that it is a slab object is printed,
573 * and, if available, the slab name, return address, and stack trace from
574 * the allocation and last free path of that object.
576 * This function will splat if passed a pointer to a non-slab object.
577 * If you are not sure what type of object you have, you should instead
578 * use mem_dump_obj().
580 void kmem_dump_obj(void *object)
582 char *cp = IS_ENABLED(CONFIG_MMU) ? "" : "/vmalloc";
585 unsigned long ptroffset;
586 struct kmem_obj_info kp = { };
588 if (WARN_ON_ONCE(!virt_addr_valid(object)))
590 slab = virt_to_slab(object);
591 if (WARN_ON_ONCE(!slab)) {
592 pr_cont(" non-slab memory.\n");
595 kmem_obj_info(&kp, object, slab);
596 if (kp.kp_slab_cache)
597 pr_cont(" slab%s %s", cp, kp.kp_slab_cache->name);
599 pr_cont(" slab%s", cp);
600 if (is_kfence_address(object))
601 pr_cont(" (kfence)");
603 pr_cont(" start %px", kp.kp_objp);
604 if (kp.kp_data_offset)
605 pr_cont(" data offset %lu", kp.kp_data_offset);
607 ptroffset = ((char *)object - (char *)kp.kp_objp) - kp.kp_data_offset;
608 pr_cont(" pointer offset %lu", ptroffset);
610 if (kp.kp_slab_cache && kp.kp_slab_cache->usersize)
611 pr_cont(" size %u", kp.kp_slab_cache->usersize);
613 pr_cont(" allocated at %pS\n", kp.kp_ret);
616 for (i = 0; i < ARRAY_SIZE(kp.kp_stack); i++) {
619 pr_info(" %pS\n", kp.kp_stack[i]);
622 if (kp.kp_free_stack[0])
623 pr_cont(" Free path:\n");
625 for (i = 0; i < ARRAY_SIZE(kp.kp_free_stack); i++) {
626 if (!kp.kp_free_stack[i])
628 pr_info(" %pS\n", kp.kp_free_stack[i]);
632 EXPORT_SYMBOL_GPL(kmem_dump_obj);
636 /* Create a cache during boot when no slab services are available yet */
637 void __init create_boot_cache(struct kmem_cache *s, const char *name,
638 unsigned int size, slab_flags_t flags,
639 unsigned int useroffset, unsigned int usersize)
642 unsigned int align = ARCH_KMALLOC_MINALIGN;
645 s->size = s->object_size = size;
648 * For power of two sizes, guarantee natural alignment for kmalloc
649 * caches, regardless of SL*B debugging options.
651 if (is_power_of_2(size))
652 align = max(align, size);
653 s->align = calculate_alignment(flags, align, size);
655 s->useroffset = useroffset;
656 s->usersize = usersize;
658 err = __kmem_cache_create(s, flags);
661 panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n",
664 s->refcount = -1; /* Exempt from merging for now */
667 struct kmem_cache *__init create_kmalloc_cache(const char *name,
668 unsigned int size, slab_flags_t flags,
669 unsigned int useroffset, unsigned int usersize)
671 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
674 panic("Out of memory when creating slab %s\n", name);
676 create_boot_cache(s, name, size, flags, useroffset, usersize);
677 kasan_cache_create_kmalloc(s);
678 list_add(&s->list, &slab_caches);
684 kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1] __ro_after_init =
685 { /* initialization for https://bugs.llvm.org/show_bug.cgi?id=42570 */ };
686 EXPORT_SYMBOL(kmalloc_caches);
689 * Conversion table for small slabs sizes / 8 to the index in the
690 * kmalloc array. This is necessary for slabs < 192 since we have non power
691 * of two cache sizes there. The size of larger slabs can be determined using
694 static u8 size_index[24] __ro_after_init = {
721 static inline unsigned int size_index_elem(unsigned int bytes)
723 return (bytes - 1) / 8;
727 * Find the kmem_cache structure that serves a given size of
730 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
736 return ZERO_SIZE_PTR;
738 index = size_index[size_index_elem(size)];
740 if (WARN_ON_ONCE(size > KMALLOC_MAX_CACHE_SIZE))
742 index = fls(size - 1);
745 return kmalloc_caches[kmalloc_type(flags)][index];
748 #ifdef CONFIG_ZONE_DMA
749 #define KMALLOC_DMA_NAME(sz) .name[KMALLOC_DMA] = "dma-kmalloc-" #sz,
751 #define KMALLOC_DMA_NAME(sz)
754 #ifdef CONFIG_MEMCG_KMEM
755 #define KMALLOC_CGROUP_NAME(sz) .name[KMALLOC_CGROUP] = "kmalloc-cg-" #sz,
757 #define KMALLOC_CGROUP_NAME(sz)
760 #define INIT_KMALLOC_INFO(__size, __short_size) \
762 .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \
763 .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #__short_size, \
764 KMALLOC_CGROUP_NAME(__short_size) \
765 KMALLOC_DMA_NAME(__short_size) \
770 * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
771 * kmalloc_index() supports up to 2^25=32MB, so the final entry of the table is
774 const struct kmalloc_info_struct kmalloc_info[] __initconst = {
775 INIT_KMALLOC_INFO(0, 0),
776 INIT_KMALLOC_INFO(96, 96),
777 INIT_KMALLOC_INFO(192, 192),
778 INIT_KMALLOC_INFO(8, 8),
779 INIT_KMALLOC_INFO(16, 16),
780 INIT_KMALLOC_INFO(32, 32),
781 INIT_KMALLOC_INFO(64, 64),
782 INIT_KMALLOC_INFO(128, 128),
783 INIT_KMALLOC_INFO(256, 256),
784 INIT_KMALLOC_INFO(512, 512),
785 INIT_KMALLOC_INFO(1024, 1k),
786 INIT_KMALLOC_INFO(2048, 2k),
787 INIT_KMALLOC_INFO(4096, 4k),
788 INIT_KMALLOC_INFO(8192, 8k),
789 INIT_KMALLOC_INFO(16384, 16k),
790 INIT_KMALLOC_INFO(32768, 32k),
791 INIT_KMALLOC_INFO(65536, 64k),
792 INIT_KMALLOC_INFO(131072, 128k),
793 INIT_KMALLOC_INFO(262144, 256k),
794 INIT_KMALLOC_INFO(524288, 512k),
795 INIT_KMALLOC_INFO(1048576, 1M),
796 INIT_KMALLOC_INFO(2097152, 2M),
797 INIT_KMALLOC_INFO(4194304, 4M),
798 INIT_KMALLOC_INFO(8388608, 8M),
799 INIT_KMALLOC_INFO(16777216, 16M),
800 INIT_KMALLOC_INFO(33554432, 32M)
804 * Patch up the size_index table if we have strange large alignment
805 * requirements for the kmalloc array. This is only the case for
806 * MIPS it seems. The standard arches will not generate any code here.
808 * Largest permitted alignment is 256 bytes due to the way we
809 * handle the index determination for the smaller caches.
811 * Make sure that nothing crazy happens if someone starts tinkering
812 * around with ARCH_KMALLOC_MINALIGN
814 void __init setup_kmalloc_cache_index_table(void)
818 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
819 !is_power_of_2(KMALLOC_MIN_SIZE));
821 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
822 unsigned int elem = size_index_elem(i);
824 if (elem >= ARRAY_SIZE(size_index))
826 size_index[elem] = KMALLOC_SHIFT_LOW;
829 if (KMALLOC_MIN_SIZE >= 64) {
831 * The 96 byte sized cache is not used if the alignment
834 for (i = 64 + 8; i <= 96; i += 8)
835 size_index[size_index_elem(i)] = 7;
839 if (KMALLOC_MIN_SIZE >= 128) {
841 * The 192 byte sized cache is not used if the alignment
842 * is 128 byte. Redirect kmalloc to use the 256 byte cache
845 for (i = 128 + 8; i <= 192; i += 8)
846 size_index[size_index_elem(i)] = 8;
851 new_kmalloc_cache(int idx, enum kmalloc_cache_type type, slab_flags_t flags)
853 if (type == KMALLOC_RECLAIM) {
854 flags |= SLAB_RECLAIM_ACCOUNT;
855 } else if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_CGROUP)) {
856 if (mem_cgroup_kmem_disabled()) {
857 kmalloc_caches[type][idx] = kmalloc_caches[KMALLOC_NORMAL][idx];
860 flags |= SLAB_ACCOUNT;
863 kmalloc_caches[type][idx] = create_kmalloc_cache(
864 kmalloc_info[idx].name[type],
865 kmalloc_info[idx].size, flags, 0,
866 kmalloc_info[idx].size);
869 * If CONFIG_MEMCG_KMEM is enabled, disable cache merging for
870 * KMALLOC_NORMAL caches.
872 if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_NORMAL))
873 kmalloc_caches[type][idx]->refcount = -1;
877 * Create the kmalloc array. Some of the regular kmalloc arrays
878 * may already have been created because they were needed to
879 * enable allocations for slab creation.
881 void __init create_kmalloc_caches(slab_flags_t flags)
884 enum kmalloc_cache_type type;
887 * Including KMALLOC_CGROUP if CONFIG_MEMCG_KMEM defined
889 for (type = KMALLOC_NORMAL; type <= KMALLOC_RECLAIM; type++) {
890 for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
891 if (!kmalloc_caches[type][i])
892 new_kmalloc_cache(i, type, flags);
895 * Caches that are not of the two-to-the-power-of size.
896 * These have to be created immediately after the
897 * earlier power of two caches
899 if (KMALLOC_MIN_SIZE <= 32 && i == 6 &&
900 !kmalloc_caches[type][1])
901 new_kmalloc_cache(1, type, flags);
902 if (KMALLOC_MIN_SIZE <= 64 && i == 7 &&
903 !kmalloc_caches[type][2])
904 new_kmalloc_cache(2, type, flags);
908 /* Kmalloc array is now usable */
911 #ifdef CONFIG_ZONE_DMA
912 for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
913 struct kmem_cache *s = kmalloc_caches[KMALLOC_NORMAL][i];
916 kmalloc_caches[KMALLOC_DMA][i] = create_kmalloc_cache(
917 kmalloc_info[i].name[KMALLOC_DMA],
918 kmalloc_info[i].size,
919 SLAB_CACHE_DMA | flags, 0,
920 kmalloc_info[i].size);
925 #endif /* !CONFIG_SLOB */
927 gfp_t kmalloc_fix_flags(gfp_t flags)
929 gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
931 flags &= ~GFP_SLAB_BUG_MASK;
932 pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
933 invalid_mask, &invalid_mask, flags, &flags);
940 * To avoid unnecessary overhead, we pass through large allocation requests
941 * directly to the page allocator. We use __GFP_COMP, because we will need to
942 * know the allocation order to free the pages properly in kfree.
944 void *kmalloc_order(size_t size, gfp_t flags, unsigned int order)
949 if (unlikely(flags & GFP_SLAB_BUG_MASK))
950 flags = kmalloc_fix_flags(flags);
953 page = alloc_pages(flags, order);
955 ret = page_address(page);
956 mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B,
959 ret = kasan_kmalloc_large(ret, size, flags);
960 /* As ret might get tagged, call kmemleak hook after KASAN. */
961 kmemleak_alloc(ret, size, 1, flags);
964 EXPORT_SYMBOL(kmalloc_order);
966 #ifdef CONFIG_TRACING
967 void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
969 void *ret = kmalloc_order(size, flags, order);
970 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
973 EXPORT_SYMBOL(kmalloc_order_trace);
976 #ifdef CONFIG_SLAB_FREELIST_RANDOM
977 /* Randomize a generic freelist */
978 static void freelist_randomize(struct rnd_state *state, unsigned int *list,
984 for (i = 0; i < count; i++)
987 /* Fisher-Yates shuffle */
988 for (i = count - 1; i > 0; i--) {
989 rand = prandom_u32_state(state);
991 swap(list[i], list[rand]);
995 /* Create a random sequence per cache */
996 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
999 struct rnd_state state;
1001 if (count < 2 || cachep->random_seq)
1004 cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp);
1005 if (!cachep->random_seq)
1008 /* Get best entropy at this stage of boot */
1009 prandom_seed_state(&state, get_random_long());
1011 freelist_randomize(&state, cachep->random_seq, count);
1015 /* Destroy the per-cache random freelist sequence */
1016 void cache_random_seq_destroy(struct kmem_cache *cachep)
1018 kfree(cachep->random_seq);
1019 cachep->random_seq = NULL;
1021 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
1023 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
1025 #define SLABINFO_RIGHTS (0600)
1027 #define SLABINFO_RIGHTS (0400)
1030 static void print_slabinfo_header(struct seq_file *m)
1033 * Output format version, so at least we can change it
1034 * without _too_ many complaints.
1036 #ifdef CONFIG_DEBUG_SLAB
1037 seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
1039 seq_puts(m, "slabinfo - version: 2.1\n");
1041 seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
1042 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
1043 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
1044 #ifdef CONFIG_DEBUG_SLAB
1045 seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
1046 seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
1051 static void *slab_start(struct seq_file *m, loff_t *pos)
1053 mutex_lock(&slab_mutex);
1054 return seq_list_start(&slab_caches, *pos);
1057 static void *slab_next(struct seq_file *m, void *p, loff_t *pos)
1059 return seq_list_next(p, &slab_caches, pos);
1062 static void slab_stop(struct seq_file *m, void *p)
1064 mutex_unlock(&slab_mutex);
1067 static void cache_show(struct kmem_cache *s, struct seq_file *m)
1069 struct slabinfo sinfo;
1071 memset(&sinfo, 0, sizeof(sinfo));
1072 get_slabinfo(s, &sinfo);
1074 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
1075 s->name, sinfo.active_objs, sinfo.num_objs, s->size,
1076 sinfo.objects_per_slab, (1 << sinfo.cache_order));
1078 seq_printf(m, " : tunables %4u %4u %4u",
1079 sinfo.limit, sinfo.batchcount, sinfo.shared);
1080 seq_printf(m, " : slabdata %6lu %6lu %6lu",
1081 sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
1082 slabinfo_show_stats(m, s);
1086 static int slab_show(struct seq_file *m, void *p)
1088 struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
1090 if (p == slab_caches.next)
1091 print_slabinfo_header(m);
1096 void dump_unreclaimable_slab(void)
1098 struct kmem_cache *s;
1099 struct slabinfo sinfo;
1102 * Here acquiring slab_mutex is risky since we don't prefer to get
1103 * sleep in oom path. But, without mutex hold, it may introduce a
1105 * Use mutex_trylock to protect the list traverse, dump nothing
1106 * without acquiring the mutex.
1108 if (!mutex_trylock(&slab_mutex)) {
1109 pr_warn("excessive unreclaimable slab but cannot dump stats\n");
1113 pr_info("Unreclaimable slab info:\n");
1114 pr_info("Name Used Total\n");
1116 list_for_each_entry(s, &slab_caches, list) {
1117 if (s->flags & SLAB_RECLAIM_ACCOUNT)
1120 get_slabinfo(s, &sinfo);
1122 if (sinfo.num_objs > 0)
1123 pr_info("%-17s %10luKB %10luKB\n", s->name,
1124 (sinfo.active_objs * s->size) / 1024,
1125 (sinfo.num_objs * s->size) / 1024);
1127 mutex_unlock(&slab_mutex);
1131 * slabinfo_op - iterator that generates /proc/slabinfo
1140 * num-pages-per-slab
1141 * + further values on SMP and with statistics enabled
1143 static const struct seq_operations slabinfo_op = {
1144 .start = slab_start,
1150 static int slabinfo_open(struct inode *inode, struct file *file)
1152 return seq_open(file, &slabinfo_op);
1155 static const struct proc_ops slabinfo_proc_ops = {
1156 .proc_flags = PROC_ENTRY_PERMANENT,
1157 .proc_open = slabinfo_open,
1158 .proc_read = seq_read,
1159 .proc_write = slabinfo_write,
1160 .proc_lseek = seq_lseek,
1161 .proc_release = seq_release,
1164 static int __init slab_proc_init(void)
1166 proc_create("slabinfo", SLABINFO_RIGHTS, NULL, &slabinfo_proc_ops);
1169 module_init(slab_proc_init);
1171 #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */
1173 static __always_inline void *__do_krealloc(const void *p, size_t new_size,
1179 /* Don't use instrumented ksize to allow precise KASAN poisoning. */
1180 if (likely(!ZERO_OR_NULL_PTR(p))) {
1181 if (!kasan_check_byte(p))
1183 ks = kfence_ksize(p) ?: __ksize(p);
1187 /* If the object still fits, repoison it precisely. */
1188 if (ks >= new_size) {
1189 p = kasan_krealloc((void *)p, new_size, flags);
1193 ret = kmalloc_track_caller(new_size, flags);
1195 /* Disable KASAN checks as the object's redzone is accessed. */
1196 kasan_disable_current();
1197 memcpy(ret, kasan_reset_tag(p), ks);
1198 kasan_enable_current();
1205 * krealloc - reallocate memory. The contents will remain unchanged.
1206 * @p: object to reallocate memory for.
1207 * @new_size: how many bytes of memory are required.
1208 * @flags: the type of memory to allocate.
1210 * The contents of the object pointed to are preserved up to the
1211 * lesser of the new and old sizes (__GFP_ZERO flag is effectively ignored).
1212 * If @p is %NULL, krealloc() behaves exactly like kmalloc(). If @new_size
1213 * is 0 and @p is not a %NULL pointer, the object pointed to is freed.
1215 * Return: pointer to the allocated memory or %NULL in case of error
1217 void *krealloc(const void *p, size_t new_size, gfp_t flags)
1221 if (unlikely(!new_size)) {
1223 return ZERO_SIZE_PTR;
1226 ret = __do_krealloc(p, new_size, flags);
1227 if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret))
1232 EXPORT_SYMBOL(krealloc);
1235 * kfree_sensitive - Clear sensitive information in memory before freeing
1236 * @p: object to free memory of
1238 * The memory of the object @p points to is zeroed before freed.
1239 * If @p is %NULL, kfree_sensitive() does nothing.
1241 * Note: this function zeroes the whole allocated buffer which can be a good
1242 * deal bigger than the requested buffer size passed to kmalloc(). So be
1243 * careful when using this function in performance sensitive code.
1245 void kfree_sensitive(const void *p)
1248 void *mem = (void *)p;
1252 memzero_explicit(mem, ks);
1255 EXPORT_SYMBOL(kfree_sensitive);
1258 * ksize - get the actual amount of memory allocated for a given object
1259 * @objp: Pointer to the object
1261 * kmalloc may internally round up allocations and return more memory
1262 * than requested. ksize() can be used to determine the actual amount of
1263 * memory allocated. The caller may use this additional memory, even though
1264 * a smaller amount of memory was initially specified with the kmalloc call.
1265 * The caller must guarantee that objp points to a valid object previously
1266 * allocated with either kmalloc() or kmem_cache_alloc(). The object
1267 * must not be freed during the duration of the call.
1269 * Return: size of the actual memory used by @objp in bytes
1271 size_t ksize(const void *objp)
1276 * We need to first check that the pointer to the object is valid, and
1277 * only then unpoison the memory. The report printed from ksize() is
1278 * more useful, then when it's printed later when the behaviour could
1279 * be undefined due to a potential use-after-free or double-free.
1281 * We use kasan_check_byte(), which is supported for the hardware
1282 * tag-based KASAN mode, unlike kasan_check_read/write().
1284 * If the pointed to memory is invalid, we return 0 to avoid users of
1285 * ksize() writing to and potentially corrupting the memory region.
1287 * We want to perform the check before __ksize(), to avoid potentially
1288 * crashing in __ksize() due to accessing invalid metadata.
1290 if (unlikely(ZERO_OR_NULL_PTR(objp)) || !kasan_check_byte(objp))
1293 size = kfence_ksize(objp) ?: __ksize(objp);
1295 * We assume that ksize callers could use whole allocated area,
1296 * so we need to unpoison this area.
1298 kasan_unpoison_range(objp, size);
1301 EXPORT_SYMBOL(ksize);
1303 /* Tracepoints definitions. */
1304 EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1305 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1306 EXPORT_TRACEPOINT_SYMBOL(kmalloc_node);
1307 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node);
1308 EXPORT_TRACEPOINT_SYMBOL(kfree);
1309 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);
1311 int should_failslab(struct kmem_cache *s, gfp_t gfpflags)
1313 if (__should_failslab(s, gfpflags))
1317 ALLOW_ERROR_INJECTION(should_failslab, ERRNO);