1 // SPDX-License-Identifier: GPL-2.0
3 * Slab allocator functions that are independent of the allocator strategy
5 * (C) 2012 Christoph Lameter <cl@linux.com>
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/dma-mapping.h>
21 #include <linux/swiotlb.h>
22 #include <linux/proc_fs.h>
23 #include <linux/debugfs.h>
24 #include <linux/kasan.h>
25 #include <asm/cacheflush.h>
26 #include <asm/tlbflush.h>
28 #include <linux/memcontrol.h>
29 #include <linux/stackdepot.h>
34 #define CREATE_TRACE_POINTS
35 #include <trace/events/kmem.h>
37 enum slab_state slab_state;
38 LIST_HEAD(slab_caches);
39 DEFINE_MUTEX(slab_mutex);
40 struct kmem_cache *kmem_cache;
42 static LIST_HEAD(slab_caches_to_rcu_destroy);
43 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work);
44 static DECLARE_WORK(slab_caches_to_rcu_destroy_work,
45 slab_caches_to_rcu_destroy_workfn);
48 * Set of flags that will prevent slab merging
50 #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
51 SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \
52 SLAB_FAILSLAB | kasan_never_merge())
54 #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \
55 SLAB_CACHE_DMA32 | SLAB_ACCOUNT)
58 * Merge control. If this is set then no merging of slab caches will occur.
60 static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT);
62 static int __init setup_slab_nomerge(char *str)
68 static int __init setup_slab_merge(char *str)
75 __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0);
76 __setup_param("slub_merge", slub_merge, setup_slab_merge, 0);
79 __setup("slab_nomerge", setup_slab_nomerge);
80 __setup("slab_merge", setup_slab_merge);
83 * Determine the size of a slab object
85 unsigned int kmem_cache_size(struct kmem_cache *s)
87 return s->object_size;
89 EXPORT_SYMBOL(kmem_cache_size);
91 #ifdef CONFIG_DEBUG_VM
92 static int kmem_cache_sanity_check(const char *name, unsigned int size)
94 if (!name || in_interrupt() || size > KMALLOC_MAX_SIZE) {
95 pr_err("kmem_cache_create(%s) integrity check failed\n", name);
99 WARN_ON(strchr(name, ' ')); /* It confuses parsers */
103 static inline int kmem_cache_sanity_check(const char *name, unsigned int size)
110 * Figure out what the alignment of the objects will be given a set of
111 * flags, a user specified alignment and the size of the objects.
113 static unsigned int calculate_alignment(slab_flags_t flags,
114 unsigned int align, unsigned int size)
117 * If the user wants hardware cache aligned objects then follow that
118 * suggestion if the object is sufficiently large.
120 * The hardware cache alignment cannot override the specified
121 * alignment though. If that is greater then use it.
123 if (flags & SLAB_HWCACHE_ALIGN) {
126 ralign = cache_line_size();
127 while (size <= ralign / 2)
129 align = max(align, ralign);
132 align = max(align, arch_slab_minalign());
134 return ALIGN(align, sizeof(void *));
138 * Find a mergeable slab cache
140 int slab_unmergeable(struct kmem_cache *s)
142 if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
148 #ifdef CONFIG_HARDENED_USERCOPY
154 * We may have set a slab to be unmergeable during bootstrap.
162 struct kmem_cache *find_mergeable(unsigned int size, unsigned int align,
163 slab_flags_t flags, const char *name, void (*ctor)(void *))
165 struct kmem_cache *s;
173 size = ALIGN(size, sizeof(void *));
174 align = calculate_alignment(flags, align, size);
175 size = ALIGN(size, align);
176 flags = kmem_cache_flags(size, flags, name);
178 if (flags & SLAB_NEVER_MERGE)
181 list_for_each_entry_reverse(s, &slab_caches, list) {
182 if (slab_unmergeable(s))
188 if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
191 * Check if alignment is compatible.
192 * Courtesy of Adrian Drzewiecki
194 if ((s->size & ~(align - 1)) != s->size)
197 if (s->size - size >= sizeof(void *))
200 if (IS_ENABLED(CONFIG_SLAB) && align &&
201 (align > s->align || s->align % align))
209 static struct kmem_cache *create_cache(const char *name,
210 unsigned int object_size, unsigned int align,
211 slab_flags_t flags, unsigned int useroffset,
212 unsigned int usersize, void (*ctor)(void *),
213 struct kmem_cache *root_cache)
215 struct kmem_cache *s;
218 if (WARN_ON(useroffset + usersize > object_size))
219 useroffset = usersize = 0;
222 s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
227 s->size = s->object_size = object_size;
230 #ifdef CONFIG_HARDENED_USERCOPY
231 s->useroffset = useroffset;
232 s->usersize = usersize;
235 err = __kmem_cache_create(s, flags);
240 list_add(&s->list, &slab_caches);
247 kmem_cache_free(kmem_cache, s);
252 * kmem_cache_create_usercopy - Create a cache with a region suitable
253 * for copying to userspace
254 * @name: A string which is used in /proc/slabinfo to identify this cache.
255 * @size: The size of objects to be created in this cache.
256 * @align: The required alignment for the objects.
258 * @useroffset: Usercopy region offset
259 * @usersize: Usercopy region size
260 * @ctor: A constructor for the objects.
262 * Cannot be called within a interrupt, but can be interrupted.
263 * The @ctor is run when new pages are allocated by the cache.
267 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
268 * to catch references to uninitialised memory.
270 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
271 * for buffer overruns.
273 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
274 * cacheline. This can be beneficial if you're counting cycles as closely
277 * Return: a pointer to the cache on success, NULL on failure.
280 kmem_cache_create_usercopy(const char *name,
281 unsigned int size, unsigned int align,
283 unsigned int useroffset, unsigned int usersize,
284 void (*ctor)(void *))
286 struct kmem_cache *s = NULL;
287 const char *cache_name;
290 #ifdef CONFIG_SLUB_DEBUG
292 * If no slub_debug was enabled globally, the static key is not yet
293 * enabled by setup_slub_debug(). Enable it if the cache is being
294 * created with any of the debugging flags passed explicitly.
295 * It's also possible that this is the first cache created with
296 * SLAB_STORE_USER and we should init stack_depot for it.
298 if (flags & SLAB_DEBUG_FLAGS)
299 static_branch_enable(&slub_debug_enabled);
300 if (flags & SLAB_STORE_USER)
304 mutex_lock(&slab_mutex);
306 err = kmem_cache_sanity_check(name, size);
311 /* Refuse requests with allocator specific flags */
312 if (flags & ~SLAB_FLAGS_PERMITTED) {
318 * Some allocators will constraint the set of valid flags to a subset
319 * of all flags. We expect them to define CACHE_CREATE_MASK in this
320 * case, and we'll just provide them with a sanitized version of the
323 flags &= CACHE_CREATE_MASK;
325 /* Fail closed on bad usersize of useroffset values. */
326 if (!IS_ENABLED(CONFIG_HARDENED_USERCOPY) ||
327 WARN_ON(!usersize && useroffset) ||
328 WARN_ON(size < usersize || size - usersize < useroffset))
329 usersize = useroffset = 0;
332 s = __kmem_cache_alias(name, size, align, flags, ctor);
336 cache_name = kstrdup_const(name, GFP_KERNEL);
342 s = create_cache(cache_name, size,
343 calculate_alignment(flags, align, size),
344 flags, useroffset, usersize, ctor, NULL);
347 kfree_const(cache_name);
351 mutex_unlock(&slab_mutex);
354 if (flags & SLAB_PANIC)
355 panic("%s: Failed to create slab '%s'. Error %d\n",
356 __func__, name, err);
358 pr_warn("%s(%s) failed with error %d\n",
359 __func__, name, err);
366 EXPORT_SYMBOL(kmem_cache_create_usercopy);
369 * kmem_cache_create - Create a cache.
370 * @name: A string which is used in /proc/slabinfo to identify this cache.
371 * @size: The size of objects to be created in this cache.
372 * @align: The required alignment for the objects.
374 * @ctor: A constructor for the objects.
376 * Cannot be called within a interrupt, but can be interrupted.
377 * The @ctor is run when new pages are allocated by the cache.
381 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
382 * to catch references to uninitialised memory.
384 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
385 * for buffer overruns.
387 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
388 * cacheline. This can be beneficial if you're counting cycles as closely
391 * Return: a pointer to the cache on success, NULL on failure.
394 kmem_cache_create(const char *name, unsigned int size, unsigned int align,
395 slab_flags_t flags, void (*ctor)(void *))
397 return kmem_cache_create_usercopy(name, size, align, flags, 0, 0,
400 EXPORT_SYMBOL(kmem_cache_create);
402 #ifdef SLAB_SUPPORTS_SYSFS
404 * For a given kmem_cache, kmem_cache_destroy() should only be called
405 * once or there will be a use-after-free problem. The actual deletion
406 * and release of the kobject does not need slab_mutex or cpu_hotplug_lock
407 * protection. So they are now done without holding those locks.
409 * Note that there will be a slight delay in the deletion of sysfs files
410 * if kmem_cache_release() is called indrectly from a work function.
412 static void kmem_cache_release(struct kmem_cache *s)
414 sysfs_slab_unlink(s);
415 sysfs_slab_release(s);
418 static void kmem_cache_release(struct kmem_cache *s)
420 slab_kmem_cache_release(s);
424 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work)
426 LIST_HEAD(to_destroy);
427 struct kmem_cache *s, *s2;
430 * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the
431 * @slab_caches_to_rcu_destroy list. The slab pages are freed
432 * through RCU and the associated kmem_cache are dereferenced
433 * while freeing the pages, so the kmem_caches should be freed only
434 * after the pending RCU operations are finished. As rcu_barrier()
435 * is a pretty slow operation, we batch all pending destructions
438 mutex_lock(&slab_mutex);
439 list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy);
440 mutex_unlock(&slab_mutex);
442 if (list_empty(&to_destroy))
447 list_for_each_entry_safe(s, s2, &to_destroy, list) {
448 debugfs_slab_release(s);
449 kfence_shutdown_cache(s);
450 kmem_cache_release(s);
454 static int shutdown_cache(struct kmem_cache *s)
456 /* free asan quarantined objects */
457 kasan_cache_shutdown(s);
459 if (__kmem_cache_shutdown(s) != 0)
464 if (s->flags & SLAB_TYPESAFE_BY_RCU) {
465 list_add_tail(&s->list, &slab_caches_to_rcu_destroy);
466 schedule_work(&slab_caches_to_rcu_destroy_work);
468 kfence_shutdown_cache(s);
469 debugfs_slab_release(s);
475 void slab_kmem_cache_release(struct kmem_cache *s)
477 __kmem_cache_release(s);
478 kfree_const(s->name);
479 kmem_cache_free(kmem_cache, s);
482 void kmem_cache_destroy(struct kmem_cache *s)
487 if (unlikely(!s) || !kasan_check_byte(s))
491 mutex_lock(&slab_mutex);
493 rcu_set = s->flags & SLAB_TYPESAFE_BY_RCU;
495 refcnt = --s->refcount;
499 WARN(shutdown_cache(s),
500 "%s %s: Slab cache still has objects when called from %pS",
501 __func__, s->name, (void *)_RET_IP_);
503 mutex_unlock(&slab_mutex);
505 if (!refcnt && !rcu_set)
506 kmem_cache_release(s);
508 EXPORT_SYMBOL(kmem_cache_destroy);
511 * kmem_cache_shrink - Shrink a cache.
512 * @cachep: The cache to shrink.
514 * Releases as many slabs as possible for a cache.
515 * To help debugging, a zero exit status indicates all slabs were released.
517 * Return: %0 if all slabs were released, non-zero otherwise
519 int kmem_cache_shrink(struct kmem_cache *cachep)
521 kasan_cache_shrink(cachep);
523 return __kmem_cache_shrink(cachep);
525 EXPORT_SYMBOL(kmem_cache_shrink);
527 bool slab_is_available(void)
529 return slab_state >= UP;
534 * kmem_valid_obj - does the pointer reference a valid slab object?
535 * @object: pointer to query.
537 * Return: %true if the pointer is to a not-yet-freed object from
538 * kmalloc() or kmem_cache_alloc(), either %true or %false if the pointer
539 * is to an already-freed object, and %false otherwise.
541 bool kmem_valid_obj(void *object)
545 /* Some arches consider ZERO_SIZE_PTR to be a valid address. */
546 if (object < (void *)PAGE_SIZE || !virt_addr_valid(object))
548 folio = virt_to_folio(object);
549 return folio_test_slab(folio);
551 EXPORT_SYMBOL_GPL(kmem_valid_obj);
553 static void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab)
555 if (__kfence_obj_info(kpp, object, slab))
557 __kmem_obj_info(kpp, object, slab);
561 * kmem_dump_obj - Print available slab provenance information
562 * @object: slab object for which to find provenance information.
564 * This function uses pr_cont(), so that the caller is expected to have
565 * printed out whatever preamble is appropriate. The provenance information
566 * depends on the type of object and on how much debugging is enabled.
567 * For a slab-cache object, the fact that it is a slab object is printed,
568 * and, if available, the slab name, return address, and stack trace from
569 * the allocation and last free path of that object.
571 * This function will splat if passed a pointer to a non-slab object.
572 * If you are not sure what type of object you have, you should instead
573 * use mem_dump_obj().
575 void kmem_dump_obj(void *object)
577 char *cp = IS_ENABLED(CONFIG_MMU) ? "" : "/vmalloc";
580 unsigned long ptroffset;
581 struct kmem_obj_info kp = { };
583 if (WARN_ON_ONCE(!virt_addr_valid(object)))
585 slab = virt_to_slab(object);
586 if (WARN_ON_ONCE(!slab)) {
587 pr_cont(" non-slab memory.\n");
590 kmem_obj_info(&kp, object, slab);
591 if (kp.kp_slab_cache)
592 pr_cont(" slab%s %s", cp, kp.kp_slab_cache->name);
594 pr_cont(" slab%s", cp);
595 if (is_kfence_address(object))
596 pr_cont(" (kfence)");
598 pr_cont(" start %px", kp.kp_objp);
599 if (kp.kp_data_offset)
600 pr_cont(" data offset %lu", kp.kp_data_offset);
602 ptroffset = ((char *)object - (char *)kp.kp_objp) - kp.kp_data_offset;
603 pr_cont(" pointer offset %lu", ptroffset);
605 if (kp.kp_slab_cache && kp.kp_slab_cache->object_size)
606 pr_cont(" size %u", kp.kp_slab_cache->object_size);
608 pr_cont(" allocated at %pS\n", kp.kp_ret);
611 for (i = 0; i < ARRAY_SIZE(kp.kp_stack); i++) {
614 pr_info(" %pS\n", kp.kp_stack[i]);
617 if (kp.kp_free_stack[0])
618 pr_cont(" Free path:\n");
620 for (i = 0; i < ARRAY_SIZE(kp.kp_free_stack); i++) {
621 if (!kp.kp_free_stack[i])
623 pr_info(" %pS\n", kp.kp_free_stack[i]);
627 EXPORT_SYMBOL_GPL(kmem_dump_obj);
630 /* Create a cache during boot when no slab services are available yet */
631 void __init create_boot_cache(struct kmem_cache *s, const char *name,
632 unsigned int size, slab_flags_t flags,
633 unsigned int useroffset, unsigned int usersize)
636 unsigned int align = ARCH_KMALLOC_MINALIGN;
639 s->size = s->object_size = size;
642 * For power of two sizes, guarantee natural alignment for kmalloc
643 * caches, regardless of SL*B debugging options.
645 if (is_power_of_2(size))
646 align = max(align, size);
647 s->align = calculate_alignment(flags, align, size);
649 #ifdef CONFIG_HARDENED_USERCOPY
650 s->useroffset = useroffset;
651 s->usersize = usersize;
654 err = __kmem_cache_create(s, flags);
657 panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n",
660 s->refcount = -1; /* Exempt from merging for now */
663 static struct kmem_cache *__init create_kmalloc_cache(const char *name,
667 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
670 panic("Out of memory when creating slab %s\n", name);
672 create_boot_cache(s, name, size, flags | SLAB_KMALLOC, 0, size);
673 list_add(&s->list, &slab_caches);
679 kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1] __ro_after_init =
680 { /* initialization for https://bugs.llvm.org/show_bug.cgi?id=42570 */ };
681 EXPORT_SYMBOL(kmalloc_caches);
684 * Conversion table for small slabs sizes / 8 to the index in the
685 * kmalloc array. This is necessary for slabs < 192 since we have non power
686 * of two cache sizes there. The size of larger slabs can be determined using
689 static u8 size_index[24] __ro_after_init = {
716 static inline unsigned int size_index_elem(unsigned int bytes)
718 return (bytes - 1) / 8;
722 * Find the kmem_cache structure that serves a given size of
725 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
731 return ZERO_SIZE_PTR;
733 index = size_index[size_index_elem(size)];
735 if (WARN_ON_ONCE(size > KMALLOC_MAX_CACHE_SIZE))
737 index = fls(size - 1);
740 return kmalloc_caches[kmalloc_type(flags)][index];
743 size_t kmalloc_size_roundup(size_t size)
745 struct kmem_cache *c;
747 /* Short-circuit the 0 size case. */
748 if (unlikely(size == 0))
750 /* Short-circuit saturated "too-large" case. */
751 if (unlikely(size == SIZE_MAX))
753 /* Above the smaller buckets, size is a multiple of page size. */
754 if (size > KMALLOC_MAX_CACHE_SIZE)
755 return PAGE_SIZE << get_order(size);
757 /* The flags don't matter since size_index is common to all. */
758 c = kmalloc_slab(size, GFP_KERNEL);
759 return c ? c->object_size : 0;
761 EXPORT_SYMBOL(kmalloc_size_roundup);
763 #ifdef CONFIG_ZONE_DMA
764 #define KMALLOC_DMA_NAME(sz) .name[KMALLOC_DMA] = "dma-kmalloc-" #sz,
766 #define KMALLOC_DMA_NAME(sz)
769 #ifdef CONFIG_MEMCG_KMEM
770 #define KMALLOC_CGROUP_NAME(sz) .name[KMALLOC_CGROUP] = "kmalloc-cg-" #sz,
772 #define KMALLOC_CGROUP_NAME(sz)
775 #ifndef CONFIG_SLUB_TINY
776 #define KMALLOC_RCL_NAME(sz) .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #sz,
778 #define KMALLOC_RCL_NAME(sz)
781 #define INIT_KMALLOC_INFO(__size, __short_size) \
783 .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \
784 KMALLOC_RCL_NAME(__short_size) \
785 KMALLOC_CGROUP_NAME(__short_size) \
786 KMALLOC_DMA_NAME(__short_size) \
791 * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
792 * kmalloc_index() supports up to 2^21=2MB, so the final entry of the table is
795 const struct kmalloc_info_struct kmalloc_info[] __initconst = {
796 INIT_KMALLOC_INFO(0, 0),
797 INIT_KMALLOC_INFO(96, 96),
798 INIT_KMALLOC_INFO(192, 192),
799 INIT_KMALLOC_INFO(8, 8),
800 INIT_KMALLOC_INFO(16, 16),
801 INIT_KMALLOC_INFO(32, 32),
802 INIT_KMALLOC_INFO(64, 64),
803 INIT_KMALLOC_INFO(128, 128),
804 INIT_KMALLOC_INFO(256, 256),
805 INIT_KMALLOC_INFO(512, 512),
806 INIT_KMALLOC_INFO(1024, 1k),
807 INIT_KMALLOC_INFO(2048, 2k),
808 INIT_KMALLOC_INFO(4096, 4k),
809 INIT_KMALLOC_INFO(8192, 8k),
810 INIT_KMALLOC_INFO(16384, 16k),
811 INIT_KMALLOC_INFO(32768, 32k),
812 INIT_KMALLOC_INFO(65536, 64k),
813 INIT_KMALLOC_INFO(131072, 128k),
814 INIT_KMALLOC_INFO(262144, 256k),
815 INIT_KMALLOC_INFO(524288, 512k),
816 INIT_KMALLOC_INFO(1048576, 1M),
817 INIT_KMALLOC_INFO(2097152, 2M)
821 * Patch up the size_index table if we have strange large alignment
822 * requirements for the kmalloc array. This is only the case for
823 * MIPS it seems. The standard arches will not generate any code here.
825 * Largest permitted alignment is 256 bytes due to the way we
826 * handle the index determination for the smaller caches.
828 * Make sure that nothing crazy happens if someone starts tinkering
829 * around with ARCH_KMALLOC_MINALIGN
831 void __init setup_kmalloc_cache_index_table(void)
835 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
836 !is_power_of_2(KMALLOC_MIN_SIZE));
838 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
839 unsigned int elem = size_index_elem(i);
841 if (elem >= ARRAY_SIZE(size_index))
843 size_index[elem] = KMALLOC_SHIFT_LOW;
846 if (KMALLOC_MIN_SIZE >= 64) {
848 * The 96 byte sized cache is not used if the alignment
851 for (i = 64 + 8; i <= 96; i += 8)
852 size_index[size_index_elem(i)] = 7;
856 if (KMALLOC_MIN_SIZE >= 128) {
858 * The 192 byte sized cache is not used if the alignment
859 * is 128 byte. Redirect kmalloc to use the 256 byte cache
862 for (i = 128 + 8; i <= 192; i += 8)
863 size_index[size_index_elem(i)] = 8;
867 static unsigned int __kmalloc_minalign(void)
869 #ifdef CONFIG_DMA_BOUNCE_UNALIGNED_KMALLOC
870 if (io_tlb_default_mem.nslabs)
871 return ARCH_KMALLOC_MINALIGN;
873 return dma_get_cache_alignment();
877 new_kmalloc_cache(int idx, enum kmalloc_cache_type type, slab_flags_t flags)
879 unsigned int minalign = __kmalloc_minalign();
880 unsigned int aligned_size = kmalloc_info[idx].size;
881 int aligned_idx = idx;
883 if ((KMALLOC_RECLAIM != KMALLOC_NORMAL) && (type == KMALLOC_RECLAIM)) {
884 flags |= SLAB_RECLAIM_ACCOUNT;
885 } else if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_CGROUP)) {
886 if (mem_cgroup_kmem_disabled()) {
887 kmalloc_caches[type][idx] = kmalloc_caches[KMALLOC_NORMAL][idx];
890 flags |= SLAB_ACCOUNT;
891 } else if (IS_ENABLED(CONFIG_ZONE_DMA) && (type == KMALLOC_DMA)) {
892 flags |= SLAB_CACHE_DMA;
895 if (minalign > ARCH_KMALLOC_MINALIGN) {
896 aligned_size = ALIGN(aligned_size, minalign);
897 aligned_idx = __kmalloc_index(aligned_size, false);
900 if (!kmalloc_caches[type][aligned_idx])
901 kmalloc_caches[type][aligned_idx] = create_kmalloc_cache(
902 kmalloc_info[aligned_idx].name[type],
903 aligned_size, flags);
904 if (idx != aligned_idx)
905 kmalloc_caches[type][idx] = kmalloc_caches[type][aligned_idx];
908 * If CONFIG_MEMCG_KMEM is enabled, disable cache merging for
909 * KMALLOC_NORMAL caches.
911 if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_NORMAL))
912 kmalloc_caches[type][idx]->refcount = -1;
916 * Create the kmalloc array. Some of the regular kmalloc arrays
917 * may already have been created because they were needed to
918 * enable allocations for slab creation.
920 void __init create_kmalloc_caches(slab_flags_t flags)
923 enum kmalloc_cache_type type;
926 * Including KMALLOC_CGROUP if CONFIG_MEMCG_KMEM defined
928 for (type = KMALLOC_NORMAL; type < NR_KMALLOC_TYPES; type++) {
929 for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
930 if (!kmalloc_caches[type][i])
931 new_kmalloc_cache(i, type, flags);
934 * Caches that are not of the two-to-the-power-of size.
935 * These have to be created immediately after the
936 * earlier power of two caches
938 if (KMALLOC_MIN_SIZE <= 32 && i == 6 &&
939 !kmalloc_caches[type][1])
940 new_kmalloc_cache(1, type, flags);
941 if (KMALLOC_MIN_SIZE <= 64 && i == 7 &&
942 !kmalloc_caches[type][2])
943 new_kmalloc_cache(2, type, flags);
947 /* Kmalloc array is now usable */
951 void free_large_kmalloc(struct folio *folio, void *object)
953 unsigned int order = folio_order(folio);
955 if (WARN_ON_ONCE(order == 0))
956 pr_warn_once("object pointer: 0x%p\n", object);
958 kmemleak_free(object);
959 kasan_kfree_large(object);
960 kmsan_kfree_large(object);
962 mod_lruvec_page_state(folio_page(folio, 0), NR_SLAB_UNRECLAIMABLE_B,
963 -(PAGE_SIZE << order));
964 __free_pages(folio_page(folio, 0), order);
967 static void *__kmalloc_large_node(size_t size, gfp_t flags, int node);
968 static __always_inline
969 void *__do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller)
971 struct kmem_cache *s;
974 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
975 ret = __kmalloc_large_node(size, flags, node);
976 trace_kmalloc(caller, ret, size,
977 PAGE_SIZE << get_order(size), flags, node);
981 s = kmalloc_slab(size, flags);
983 if (unlikely(ZERO_OR_NULL_PTR(s)))
986 ret = __kmem_cache_alloc_node(s, flags, node, size, caller);
987 ret = kasan_kmalloc(s, ret, size, flags);
988 trace_kmalloc(caller, ret, size, s->size, flags, node);
992 void *__kmalloc_node(size_t size, gfp_t flags, int node)
994 return __do_kmalloc_node(size, flags, node, _RET_IP_);
996 EXPORT_SYMBOL(__kmalloc_node);
998 void *__kmalloc(size_t size, gfp_t flags)
1000 return __do_kmalloc_node(size, flags, NUMA_NO_NODE, _RET_IP_);
1002 EXPORT_SYMBOL(__kmalloc);
1004 void *__kmalloc_node_track_caller(size_t size, gfp_t flags,
1005 int node, unsigned long caller)
1007 return __do_kmalloc_node(size, flags, node, caller);
1009 EXPORT_SYMBOL(__kmalloc_node_track_caller);
1012 * kfree - free previously allocated memory
1013 * @object: pointer returned by kmalloc() or kmem_cache_alloc()
1015 * If @object is NULL, no operation is performed.
1017 void kfree(const void *object)
1019 struct folio *folio;
1021 struct kmem_cache *s;
1023 trace_kfree(_RET_IP_, object);
1025 if (unlikely(ZERO_OR_NULL_PTR(object)))
1028 folio = virt_to_folio(object);
1029 if (unlikely(!folio_test_slab(folio))) {
1030 free_large_kmalloc(folio, (void *)object);
1034 slab = folio_slab(folio);
1035 s = slab->slab_cache;
1036 __kmem_cache_free(s, (void *)object, _RET_IP_);
1038 EXPORT_SYMBOL(kfree);
1041 * __ksize -- Report full size of underlying allocation
1042 * @object: pointer to the object
1044 * This should only be used internally to query the true size of allocations.
1045 * It is not meant to be a way to discover the usable size of an allocation
1046 * after the fact. Instead, use kmalloc_size_roundup(). Using memory beyond
1047 * the originally requested allocation size may trigger KASAN, UBSAN_BOUNDS,
1048 * and/or FORTIFY_SOURCE.
1050 * Return: size of the actual memory used by @object in bytes
1052 size_t __ksize(const void *object)
1054 struct folio *folio;
1056 if (unlikely(object == ZERO_SIZE_PTR))
1059 folio = virt_to_folio(object);
1061 if (unlikely(!folio_test_slab(folio))) {
1062 if (WARN_ON(folio_size(folio) <= KMALLOC_MAX_CACHE_SIZE))
1064 if (WARN_ON(object != folio_address(folio)))
1066 return folio_size(folio);
1069 #ifdef CONFIG_SLUB_DEBUG
1070 skip_orig_size_check(folio_slab(folio)->slab_cache, object);
1073 return slab_ksize(folio_slab(folio)->slab_cache);
1076 void *kmalloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
1078 void *ret = __kmem_cache_alloc_node(s, gfpflags, NUMA_NO_NODE,
1081 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags, NUMA_NO_NODE);
1083 ret = kasan_kmalloc(s, ret, size, gfpflags);
1086 EXPORT_SYMBOL(kmalloc_trace);
1088 void *kmalloc_node_trace(struct kmem_cache *s, gfp_t gfpflags,
1089 int node, size_t size)
1091 void *ret = __kmem_cache_alloc_node(s, gfpflags, node, size, _RET_IP_);
1093 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags, node);
1095 ret = kasan_kmalloc(s, ret, size, gfpflags);
1098 EXPORT_SYMBOL(kmalloc_node_trace);
1100 gfp_t kmalloc_fix_flags(gfp_t flags)
1102 gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
1104 flags &= ~GFP_SLAB_BUG_MASK;
1105 pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
1106 invalid_mask, &invalid_mask, flags, &flags);
1113 * To avoid unnecessary overhead, we pass through large allocation requests
1114 * directly to the page allocator. We use __GFP_COMP, because we will need to
1115 * know the allocation order to free the pages properly in kfree.
1118 static void *__kmalloc_large_node(size_t size, gfp_t flags, int node)
1122 unsigned int order = get_order(size);
1124 if (unlikely(flags & GFP_SLAB_BUG_MASK))
1125 flags = kmalloc_fix_flags(flags);
1127 flags |= __GFP_COMP;
1128 page = alloc_pages_node(node, flags, order);
1130 ptr = page_address(page);
1131 mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B,
1132 PAGE_SIZE << order);
1135 ptr = kasan_kmalloc_large(ptr, size, flags);
1136 /* As ptr might get tagged, call kmemleak hook after KASAN. */
1137 kmemleak_alloc(ptr, size, 1, flags);
1138 kmsan_kmalloc_large(ptr, size, flags);
1143 void *kmalloc_large(size_t size, gfp_t flags)
1145 void *ret = __kmalloc_large_node(size, flags, NUMA_NO_NODE);
1147 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << get_order(size),
1148 flags, NUMA_NO_NODE);
1151 EXPORT_SYMBOL(kmalloc_large);
1153 void *kmalloc_large_node(size_t size, gfp_t flags, int node)
1155 void *ret = __kmalloc_large_node(size, flags, node);
1157 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << get_order(size),
1161 EXPORT_SYMBOL(kmalloc_large_node);
1163 #ifdef CONFIG_SLAB_FREELIST_RANDOM
1164 /* Randomize a generic freelist */
1165 static void freelist_randomize(struct rnd_state *state, unsigned int *list,
1171 for (i = 0; i < count; i++)
1174 /* Fisher-Yates shuffle */
1175 for (i = count - 1; i > 0; i--) {
1176 rand = prandom_u32_state(state);
1178 swap(list[i], list[rand]);
1182 /* Create a random sequence per cache */
1183 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
1186 struct rnd_state state;
1188 if (count < 2 || cachep->random_seq)
1191 cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp);
1192 if (!cachep->random_seq)
1195 /* Get best entropy at this stage of boot */
1196 prandom_seed_state(&state, get_random_long());
1198 freelist_randomize(&state, cachep->random_seq, count);
1202 /* Destroy the per-cache random freelist sequence */
1203 void cache_random_seq_destroy(struct kmem_cache *cachep)
1205 kfree(cachep->random_seq);
1206 cachep->random_seq = NULL;
1208 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
1210 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
1212 #define SLABINFO_RIGHTS (0600)
1214 #define SLABINFO_RIGHTS (0400)
1217 static void print_slabinfo_header(struct seq_file *m)
1220 * Output format version, so at least we can change it
1221 * without _too_ many complaints.
1223 #ifdef CONFIG_DEBUG_SLAB
1224 seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
1226 seq_puts(m, "slabinfo - version: 2.1\n");
1228 seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
1229 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
1230 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
1231 #ifdef CONFIG_DEBUG_SLAB
1232 seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
1233 seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
1238 static void *slab_start(struct seq_file *m, loff_t *pos)
1240 mutex_lock(&slab_mutex);
1241 return seq_list_start(&slab_caches, *pos);
1244 static void *slab_next(struct seq_file *m, void *p, loff_t *pos)
1246 return seq_list_next(p, &slab_caches, pos);
1249 static void slab_stop(struct seq_file *m, void *p)
1251 mutex_unlock(&slab_mutex);
1254 static void cache_show(struct kmem_cache *s, struct seq_file *m)
1256 struct slabinfo sinfo;
1258 memset(&sinfo, 0, sizeof(sinfo));
1259 get_slabinfo(s, &sinfo);
1261 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
1262 s->name, sinfo.active_objs, sinfo.num_objs, s->size,
1263 sinfo.objects_per_slab, (1 << sinfo.cache_order));
1265 seq_printf(m, " : tunables %4u %4u %4u",
1266 sinfo.limit, sinfo.batchcount, sinfo.shared);
1267 seq_printf(m, " : slabdata %6lu %6lu %6lu",
1268 sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
1269 slabinfo_show_stats(m, s);
1273 static int slab_show(struct seq_file *m, void *p)
1275 struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
1277 if (p == slab_caches.next)
1278 print_slabinfo_header(m);
1283 void dump_unreclaimable_slab(void)
1285 struct kmem_cache *s;
1286 struct slabinfo sinfo;
1289 * Here acquiring slab_mutex is risky since we don't prefer to get
1290 * sleep in oom path. But, without mutex hold, it may introduce a
1292 * Use mutex_trylock to protect the list traverse, dump nothing
1293 * without acquiring the mutex.
1295 if (!mutex_trylock(&slab_mutex)) {
1296 pr_warn("excessive unreclaimable slab but cannot dump stats\n");
1300 pr_info("Unreclaimable slab info:\n");
1301 pr_info("Name Used Total\n");
1303 list_for_each_entry(s, &slab_caches, list) {
1304 if (s->flags & SLAB_RECLAIM_ACCOUNT)
1307 get_slabinfo(s, &sinfo);
1309 if (sinfo.num_objs > 0)
1310 pr_info("%-17s %10luKB %10luKB\n", s->name,
1311 (sinfo.active_objs * s->size) / 1024,
1312 (sinfo.num_objs * s->size) / 1024);
1314 mutex_unlock(&slab_mutex);
1318 * slabinfo_op - iterator that generates /proc/slabinfo
1327 * num-pages-per-slab
1328 * + further values on SMP and with statistics enabled
1330 static const struct seq_operations slabinfo_op = {
1331 .start = slab_start,
1337 static int slabinfo_open(struct inode *inode, struct file *file)
1339 return seq_open(file, &slabinfo_op);
1342 static const struct proc_ops slabinfo_proc_ops = {
1343 .proc_flags = PROC_ENTRY_PERMANENT,
1344 .proc_open = slabinfo_open,
1345 .proc_read = seq_read,
1346 .proc_write = slabinfo_write,
1347 .proc_lseek = seq_lseek,
1348 .proc_release = seq_release,
1351 static int __init slab_proc_init(void)
1353 proc_create("slabinfo", SLABINFO_RIGHTS, NULL, &slabinfo_proc_ops);
1356 module_init(slab_proc_init);
1358 #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */
1360 static __always_inline __realloc_size(2) void *
1361 __do_krealloc(const void *p, size_t new_size, gfp_t flags)
1366 /* Check for double-free before calling ksize. */
1367 if (likely(!ZERO_OR_NULL_PTR(p))) {
1368 if (!kasan_check_byte(p))
1374 /* If the object still fits, repoison it precisely. */
1375 if (ks >= new_size) {
1376 p = kasan_krealloc((void *)p, new_size, flags);
1380 ret = kmalloc_track_caller(new_size, flags);
1382 /* Disable KASAN checks as the object's redzone is accessed. */
1383 kasan_disable_current();
1384 memcpy(ret, kasan_reset_tag(p), ks);
1385 kasan_enable_current();
1392 * krealloc - reallocate memory. The contents will remain unchanged.
1393 * @p: object to reallocate memory for.
1394 * @new_size: how many bytes of memory are required.
1395 * @flags: the type of memory to allocate.
1397 * The contents of the object pointed to are preserved up to the
1398 * lesser of the new and old sizes (__GFP_ZERO flag is effectively ignored).
1399 * If @p is %NULL, krealloc() behaves exactly like kmalloc(). If @new_size
1400 * is 0 and @p is not a %NULL pointer, the object pointed to is freed.
1402 * Return: pointer to the allocated memory or %NULL in case of error
1404 void *krealloc(const void *p, size_t new_size, gfp_t flags)
1408 if (unlikely(!new_size)) {
1410 return ZERO_SIZE_PTR;
1413 ret = __do_krealloc(p, new_size, flags);
1414 if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret))
1419 EXPORT_SYMBOL(krealloc);
1422 * kfree_sensitive - Clear sensitive information in memory before freeing
1423 * @p: object to free memory of
1425 * The memory of the object @p points to is zeroed before freed.
1426 * If @p is %NULL, kfree_sensitive() does nothing.
1428 * Note: this function zeroes the whole allocated buffer which can be a good
1429 * deal bigger than the requested buffer size passed to kmalloc(). So be
1430 * careful when using this function in performance sensitive code.
1432 void kfree_sensitive(const void *p)
1435 void *mem = (void *)p;
1439 kasan_unpoison_range(mem, ks);
1440 memzero_explicit(mem, ks);
1444 EXPORT_SYMBOL(kfree_sensitive);
1446 size_t ksize(const void *objp)
1449 * We need to first check that the pointer to the object is valid.
1450 * The KASAN report printed from ksize() is more useful, then when
1451 * it's printed later when the behaviour could be undefined due to
1452 * a potential use-after-free or double-free.
1454 * We use kasan_check_byte(), which is supported for the hardware
1455 * tag-based KASAN mode, unlike kasan_check_read/write().
1457 * If the pointed to memory is invalid, we return 0 to avoid users of
1458 * ksize() writing to and potentially corrupting the memory region.
1460 * We want to perform the check before __ksize(), to avoid potentially
1461 * crashing in __ksize() due to accessing invalid metadata.
1463 if (unlikely(ZERO_OR_NULL_PTR(objp)) || !kasan_check_byte(objp))
1466 return kfence_ksize(objp) ?: __ksize(objp);
1468 EXPORT_SYMBOL(ksize);
1470 /* Tracepoints definitions. */
1471 EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1472 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1473 EXPORT_TRACEPOINT_SYMBOL(kfree);
1474 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);
1476 int should_failslab(struct kmem_cache *s, gfp_t gfpflags)
1478 if (__should_failslab(s, gfpflags))
1482 ALLOW_ERROR_INJECTION(should_failslab, ERRNO);