1 /* SPDX-License-Identifier: GPL-2.0 */
5 * (C) SGI 2006, Christoph Lameter
6 * Cleaned up and restructured to ease the addition of alternative
7 * implementations of SLAB allocators.
8 * (C) Linux Foundation 2008-2013
9 * Unified interface for all slab allocators
15 #include <linux/gfp.h>
16 #include <linux/overflow.h>
17 #include <linux/types.h>
18 #include <linux/workqueue.h>
19 #include <linux/percpu-refcount.h>
23 * Flags to pass to kmem_cache_create().
24 * The ones marked DEBUG are only valid if CONFIG_DEBUG_SLAB is set.
26 /* DEBUG: Perform (expensive) checks on alloc/free */
27 #define SLAB_CONSISTENCY_CHECKS ((slab_flags_t __force)0x00000100U)
28 /* DEBUG: Red zone objs in a cache */
29 #define SLAB_RED_ZONE ((slab_flags_t __force)0x00000400U)
30 /* DEBUG: Poison objects */
31 #define SLAB_POISON ((slab_flags_t __force)0x00000800U)
32 /* Align objs on cache lines */
33 #define SLAB_HWCACHE_ALIGN ((slab_flags_t __force)0x00002000U)
34 /* Use GFP_DMA memory */
35 #define SLAB_CACHE_DMA ((slab_flags_t __force)0x00004000U)
36 /* Use GFP_DMA32 memory */
37 #define SLAB_CACHE_DMA32 ((slab_flags_t __force)0x00008000U)
38 /* DEBUG: Store the last owner for bug hunting */
39 #define SLAB_STORE_USER ((slab_flags_t __force)0x00010000U)
40 /* Panic if kmem_cache_create() fails */
41 #define SLAB_PANIC ((slab_flags_t __force)0x00040000U)
43 * SLAB_TYPESAFE_BY_RCU - **WARNING** READ THIS!
45 * This delays freeing the SLAB page by a grace period, it does _NOT_
46 * delay object freeing. This means that if you do kmem_cache_free()
47 * that memory location is free to be reused at any time. Thus it may
48 * be possible to see another object there in the same RCU grace period.
50 * This feature only ensures the memory location backing the object
51 * stays valid, the trick to using this is relying on an independent
52 * object validation pass. Something like:
56 * obj = lockless_lookup(key);
58 * if (!try_get_ref(obj)) // might fail for free objects
61 * if (obj->key != key) { // not the object we expected
68 * This is useful if we need to approach a kernel structure obliquely,
69 * from its address obtained without the usual locking. We can lock
70 * the structure to stabilize it and check it's still at the given address,
71 * only if we can be sure that the memory has not been meanwhile reused
72 * for some other kind of object (which our subsystem's lock might corrupt).
74 * rcu_read_lock before reading the address, then rcu_read_unlock after
75 * taking the spinlock within the structure expected at that address.
77 * Note that SLAB_TYPESAFE_BY_RCU was originally named SLAB_DESTROY_BY_RCU.
79 /* Defer freeing slabs to RCU */
80 #define SLAB_TYPESAFE_BY_RCU ((slab_flags_t __force)0x00080000U)
81 /* Spread some memory over cpuset */
82 #define SLAB_MEM_SPREAD ((slab_flags_t __force)0x00100000U)
83 /* Trace allocations and frees */
84 #define SLAB_TRACE ((slab_flags_t __force)0x00200000U)
86 /* Flag to prevent checks on free */
87 #ifdef CONFIG_DEBUG_OBJECTS
88 # define SLAB_DEBUG_OBJECTS ((slab_flags_t __force)0x00400000U)
90 # define SLAB_DEBUG_OBJECTS 0
93 /* Avoid kmemleak tracing */
94 #define SLAB_NOLEAKTRACE ((slab_flags_t __force)0x00800000U)
96 /* Fault injection mark */
97 #ifdef CONFIG_FAILSLAB
98 # define SLAB_FAILSLAB ((slab_flags_t __force)0x02000000U)
100 # define SLAB_FAILSLAB 0
102 /* Account to memcg */
103 #ifdef CONFIG_MEMCG_KMEM
104 # define SLAB_ACCOUNT ((slab_flags_t __force)0x04000000U)
106 # define SLAB_ACCOUNT 0
110 #define SLAB_KASAN ((slab_flags_t __force)0x08000000U)
115 /* The following flags affect the page allocator grouping pages by mobility */
116 /* Objects are reclaimable */
117 #define SLAB_RECLAIM_ACCOUNT ((slab_flags_t __force)0x00020000U)
118 #define SLAB_TEMPORARY SLAB_RECLAIM_ACCOUNT /* Objects are short-lived */
120 /* Slab deactivation flag */
121 #define SLAB_DEACTIVATED ((slab_flags_t __force)0x10000000U)
124 * ZERO_SIZE_PTR will be returned for zero sized kmalloc requests.
126 * Dereferencing ZERO_SIZE_PTR will lead to a distinct access fault.
128 * ZERO_SIZE_PTR can be passed to kfree though in the same way that NULL can.
129 * Both make kfree a no-op.
131 #define ZERO_SIZE_PTR ((void *)16)
133 #define ZERO_OR_NULL_PTR(x) ((unsigned long)(x) <= \
134 (unsigned long)ZERO_SIZE_PTR)
136 #include <linux/kasan.h>
140 * struct kmem_cache related prototypes
142 void __init kmem_cache_init(void);
143 bool slab_is_available(void);
145 extern bool usercopy_fallback;
147 struct kmem_cache *kmem_cache_create(const char *name, unsigned int size,
148 unsigned int align, slab_flags_t flags,
149 void (*ctor)(void *));
150 struct kmem_cache *kmem_cache_create_usercopy(const char *name,
151 unsigned int size, unsigned int align,
153 unsigned int useroffset, unsigned int usersize,
154 void (*ctor)(void *));
155 void kmem_cache_destroy(struct kmem_cache *);
156 int kmem_cache_shrink(struct kmem_cache *);
158 void memcg_create_kmem_cache(struct mem_cgroup *, struct kmem_cache *);
159 void memcg_deactivate_kmem_caches(struct mem_cgroup *, struct mem_cgroup *);
162 * Please use this macro to create slab caches. Simply specify the
163 * name of the structure and maybe some flags that are listed above.
165 * The alignment of the struct determines object alignment. If you
166 * f.e. add ____cacheline_aligned_in_smp to the struct declaration
167 * then the objects will be properly aligned in SMP configurations.
169 #define KMEM_CACHE(__struct, __flags) \
170 kmem_cache_create(#__struct, sizeof(struct __struct), \
171 __alignof__(struct __struct), (__flags), NULL)
174 * To whitelist a single field for copying to/from usercopy, use this
175 * macro instead for KMEM_CACHE() above.
177 #define KMEM_CACHE_USERCOPY(__struct, __flags, __field) \
178 kmem_cache_create_usercopy(#__struct, \
179 sizeof(struct __struct), \
180 __alignof__(struct __struct), (__flags), \
181 offsetof(struct __struct, __field), \
182 sizeof_field(struct __struct, __field), NULL)
185 * Common kmalloc functions provided by all allocators
187 void * __must_check krealloc(const void *, size_t, gfp_t);
188 void kfree(const void *);
189 void kzfree(const void *);
190 size_t __ksize(const void *);
191 size_t ksize(const void *);
193 #ifdef CONFIG_HAVE_HARDENED_USERCOPY_ALLOCATOR
194 void __check_heap_object(const void *ptr, unsigned long n, struct page *page,
197 static inline void __check_heap_object(const void *ptr, unsigned long n,
198 struct page *page, bool to_user) { }
202 * Some archs want to perform DMA into kmalloc caches and need a guaranteed
203 * alignment larger than the alignment of a 64-bit integer.
204 * Setting ARCH_KMALLOC_MINALIGN in arch headers allows that.
206 #if defined(ARCH_DMA_MINALIGN) && ARCH_DMA_MINALIGN > 8
207 #define ARCH_KMALLOC_MINALIGN ARCH_DMA_MINALIGN
208 #define KMALLOC_MIN_SIZE ARCH_DMA_MINALIGN
209 #define KMALLOC_SHIFT_LOW ilog2(ARCH_DMA_MINALIGN)
211 #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
215 * Setting ARCH_SLAB_MINALIGN in arch headers allows a different alignment.
216 * Intended for arches that get misalignment faults even for 64 bit integer
219 #ifndef ARCH_SLAB_MINALIGN
220 #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
224 * kmalloc and friends return ARCH_KMALLOC_MINALIGN aligned
225 * pointers. kmem_cache_alloc and friends return ARCH_SLAB_MINALIGN
228 #define __assume_kmalloc_alignment __assume_aligned(ARCH_KMALLOC_MINALIGN)
229 #define __assume_slab_alignment __assume_aligned(ARCH_SLAB_MINALIGN)
230 #define __assume_page_alignment __assume_aligned(PAGE_SIZE)
233 * Kmalloc array related definitions
238 * The largest kmalloc size supported by the SLAB allocators is
239 * 32 megabyte (2^25) or the maximum allocatable page order if that is
242 * WARNING: Its not easy to increase this value since the allocators have
243 * to do various tricks to work around compiler limitations in order to
244 * ensure proper constant folding.
246 #define KMALLOC_SHIFT_HIGH ((MAX_ORDER + PAGE_SHIFT - 1) <= 25 ? \
247 (MAX_ORDER + PAGE_SHIFT - 1) : 25)
248 #define KMALLOC_SHIFT_MAX KMALLOC_SHIFT_HIGH
249 #ifndef KMALLOC_SHIFT_LOW
250 #define KMALLOC_SHIFT_LOW 5
256 * SLUB directly allocates requests fitting in to an order-1 page
257 * (PAGE_SIZE*2). Larger requests are passed to the page allocator.
259 #define KMALLOC_SHIFT_HIGH (PAGE_SHIFT + 1)
260 #define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT - 1)
261 #ifndef KMALLOC_SHIFT_LOW
262 #define KMALLOC_SHIFT_LOW 3
268 * SLOB passes all requests larger than one page to the page allocator.
269 * No kmalloc array is necessary since objects of different sizes can
270 * be allocated from the same page.
272 #define KMALLOC_SHIFT_HIGH PAGE_SHIFT
273 #define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT - 1)
274 #ifndef KMALLOC_SHIFT_LOW
275 #define KMALLOC_SHIFT_LOW 3
279 /* Maximum allocatable size */
280 #define KMALLOC_MAX_SIZE (1UL << KMALLOC_SHIFT_MAX)
281 /* Maximum size for which we actually use a slab cache */
282 #define KMALLOC_MAX_CACHE_SIZE (1UL << KMALLOC_SHIFT_HIGH)
283 /* Maximum order allocatable via the slab allocagtor */
284 #define KMALLOC_MAX_ORDER (KMALLOC_SHIFT_MAX - PAGE_SHIFT)
289 #ifndef KMALLOC_MIN_SIZE
290 #define KMALLOC_MIN_SIZE (1 << KMALLOC_SHIFT_LOW)
294 * This restriction comes from byte sized index implementation.
295 * Page size is normally 2^12 bytes and, in this case, if we want to use
296 * byte sized index which can represent 2^8 entries, the size of the object
297 * should be equal or greater to 2^12 / 2^8 = 2^4 = 16.
298 * If minimum size of kmalloc is less than 16, we use it as minimum object
299 * size and give up to use byte sized index.
301 #define SLAB_OBJ_MIN_SIZE (KMALLOC_MIN_SIZE < 16 ? \
302 (KMALLOC_MIN_SIZE) : 16)
305 * Whenever changing this, take care of that kmalloc_type() and
306 * create_kmalloc_caches() still work as intended.
308 enum kmalloc_cache_type {
311 #ifdef CONFIG_ZONE_DMA
318 extern struct kmem_cache *
319 kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1];
321 static __always_inline enum kmalloc_cache_type kmalloc_type(gfp_t flags)
323 #ifdef CONFIG_ZONE_DMA
325 * The most common case is KMALLOC_NORMAL, so test for it
326 * with a single branch for both flags.
328 if (likely((flags & (__GFP_DMA | __GFP_RECLAIMABLE)) == 0))
329 return KMALLOC_NORMAL;
332 * At least one of the flags has to be set. If both are, __GFP_DMA
335 return flags & __GFP_DMA ? KMALLOC_DMA : KMALLOC_RECLAIM;
337 return flags & __GFP_RECLAIMABLE ? KMALLOC_RECLAIM : KMALLOC_NORMAL;
342 * Figure out which kmalloc slab an allocation of a certain size
346 * 2 = 129 .. 192 bytes
347 * n = 2^(n-1)+1 .. 2^n
349 static __always_inline unsigned int kmalloc_index(size_t size)
354 if (size <= KMALLOC_MIN_SIZE)
355 return KMALLOC_SHIFT_LOW;
357 if (KMALLOC_MIN_SIZE <= 32 && size > 64 && size <= 96)
359 if (KMALLOC_MIN_SIZE <= 64 && size > 128 && size <= 192)
361 if (size <= 8) return 3;
362 if (size <= 16) return 4;
363 if (size <= 32) return 5;
364 if (size <= 64) return 6;
365 if (size <= 128) return 7;
366 if (size <= 256) return 8;
367 if (size <= 512) return 9;
368 if (size <= 1024) return 10;
369 if (size <= 2 * 1024) return 11;
370 if (size <= 4 * 1024) return 12;
371 if (size <= 8 * 1024) return 13;
372 if (size <= 16 * 1024) return 14;
373 if (size <= 32 * 1024) return 15;
374 if (size <= 64 * 1024) return 16;
375 if (size <= 128 * 1024) return 17;
376 if (size <= 256 * 1024) return 18;
377 if (size <= 512 * 1024) return 19;
378 if (size <= 1024 * 1024) return 20;
379 if (size <= 2 * 1024 * 1024) return 21;
380 if (size <= 4 * 1024 * 1024) return 22;
381 if (size <= 8 * 1024 * 1024) return 23;
382 if (size <= 16 * 1024 * 1024) return 24;
383 if (size <= 32 * 1024 * 1024) return 25;
384 if (size <= 64 * 1024 * 1024) return 26;
387 /* Will never be reached. Needed because the compiler may complain */
390 #endif /* !CONFIG_SLOB */
392 void *__kmalloc(size_t size, gfp_t flags) __assume_kmalloc_alignment __malloc;
393 void *kmem_cache_alloc(struct kmem_cache *, gfp_t flags) __assume_slab_alignment __malloc;
394 void kmem_cache_free(struct kmem_cache *, void *);
397 * Bulk allocation and freeing operations. These are accelerated in an
398 * allocator specific way to avoid taking locks repeatedly or building
399 * metadata structures unnecessarily.
401 * Note that interrupts must be enabled when calling these functions.
403 void kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
404 int kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
407 * Caller must not use kfree_bulk() on memory not originally allocated
408 * by kmalloc(), because the SLOB allocator cannot handle this.
410 static __always_inline void kfree_bulk(size_t size, void **p)
412 kmem_cache_free_bulk(NULL, size, p);
416 void *__kmalloc_node(size_t size, gfp_t flags, int node) __assume_kmalloc_alignment __malloc;
417 void *kmem_cache_alloc_node(struct kmem_cache *, gfp_t flags, int node) __assume_slab_alignment __malloc;
419 static __always_inline void *__kmalloc_node(size_t size, gfp_t flags, int node)
421 return __kmalloc(size, flags);
424 static __always_inline void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t flags, int node)
426 return kmem_cache_alloc(s, flags);
430 #ifdef CONFIG_TRACING
431 extern void *kmem_cache_alloc_trace(struct kmem_cache *, gfp_t, size_t) __assume_slab_alignment __malloc;
434 extern void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
436 int node, size_t size) __assume_slab_alignment __malloc;
438 static __always_inline void *
439 kmem_cache_alloc_node_trace(struct kmem_cache *s,
441 int node, size_t size)
443 return kmem_cache_alloc_trace(s, gfpflags, size);
445 #endif /* CONFIG_NUMA */
447 #else /* CONFIG_TRACING */
448 static __always_inline void *kmem_cache_alloc_trace(struct kmem_cache *s,
449 gfp_t flags, size_t size)
451 void *ret = kmem_cache_alloc(s, flags);
453 ret = kasan_kmalloc(s, ret, size, flags);
457 static __always_inline void *
458 kmem_cache_alloc_node_trace(struct kmem_cache *s,
460 int node, size_t size)
462 void *ret = kmem_cache_alloc_node(s, gfpflags, node);
464 ret = kasan_kmalloc(s, ret, size, gfpflags);
467 #endif /* CONFIG_TRACING */
469 extern void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) __assume_page_alignment __malloc;
471 #ifdef CONFIG_TRACING
472 extern void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) __assume_page_alignment __malloc;
474 static __always_inline void *
475 kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
477 return kmalloc_order(size, flags, order);
481 static __always_inline void *kmalloc_large(size_t size, gfp_t flags)
483 unsigned int order = get_order(size);
484 return kmalloc_order_trace(size, flags, order);
488 * kmalloc - allocate memory
489 * @size: how many bytes of memory are required.
490 * @flags: the type of memory to allocate.
492 * kmalloc is the normal method of allocating memory
493 * for objects smaller than page size in the kernel.
495 * The allocated object address is aligned to at least ARCH_KMALLOC_MINALIGN
496 * bytes. For @size of power of two bytes, the alignment is also guaranteed
497 * to be at least to the size.
499 * The @flags argument may be one of the GFP flags defined at
500 * include/linux/gfp.h and described at
501 * :ref:`Documentation/core-api/mm-api.rst <mm-api-gfp-flags>`
503 * The recommended usage of the @flags is described at
504 * :ref:`Documentation/core-api/memory-allocation.rst <memory_allocation>`
506 * Below is a brief outline of the most useful GFP flags
509 * Allocate normal kernel ram. May sleep.
512 * Allocation will not sleep.
515 * Allocation will not sleep. May use emergency pools.
518 * Allocate memory from high memory on behalf of user.
520 * Also it is possible to set different flags by OR'ing
521 * in one or more of the following additional @flags:
524 * This allocation has high priority and may use emergency pools.
527 * Indicate that this allocation is in no way allowed to fail
528 * (think twice before using).
531 * If memory is not immediately available,
532 * then give up at once.
535 * If allocation fails, don't issue any warnings.
537 * %__GFP_RETRY_MAYFAIL
538 * Try really hard to succeed the allocation but fail
541 static __always_inline void *kmalloc(size_t size, gfp_t flags)
543 if (__builtin_constant_p(size)) {
547 if (size > KMALLOC_MAX_CACHE_SIZE)
548 return kmalloc_large(size, flags);
550 index = kmalloc_index(size);
553 return ZERO_SIZE_PTR;
555 return kmem_cache_alloc_trace(
556 kmalloc_caches[kmalloc_type(flags)][index],
560 return __kmalloc(size, flags);
563 static __always_inline void *kmalloc_node(size_t size, gfp_t flags, int node)
566 if (__builtin_constant_p(size) &&
567 size <= KMALLOC_MAX_CACHE_SIZE) {
568 unsigned int i = kmalloc_index(size);
571 return ZERO_SIZE_PTR;
573 return kmem_cache_alloc_node_trace(
574 kmalloc_caches[kmalloc_type(flags)][i],
578 return __kmalloc_node(size, flags, node);
581 int memcg_update_all_caches(int num_memcgs);
584 * kmalloc_array - allocate memory for an array.
585 * @n: number of elements.
586 * @size: element size.
587 * @flags: the type of memory to allocate (see kmalloc).
589 static inline void *kmalloc_array(size_t n, size_t size, gfp_t flags)
593 if (unlikely(check_mul_overflow(n, size, &bytes)))
595 if (__builtin_constant_p(n) && __builtin_constant_p(size))
596 return kmalloc(bytes, flags);
597 return __kmalloc(bytes, flags);
601 * kcalloc - allocate memory for an array. The memory is set to zero.
602 * @n: number of elements.
603 * @size: element size.
604 * @flags: the type of memory to allocate (see kmalloc).
606 static inline void *kcalloc(size_t n, size_t size, gfp_t flags)
608 return kmalloc_array(n, size, flags | __GFP_ZERO);
612 * kmalloc_track_caller is a special version of kmalloc that records the
613 * calling function of the routine calling it for slab leak tracking instead
614 * of just the calling function (confusing, eh?).
615 * It's useful when the call to kmalloc comes from a widely-used standard
616 * allocator where we care about the real place the memory allocation
617 * request comes from.
619 extern void *__kmalloc_track_caller(size_t, gfp_t, unsigned long);
620 #define kmalloc_track_caller(size, flags) \
621 __kmalloc_track_caller(size, flags, _RET_IP_)
623 static inline void *kmalloc_array_node(size_t n, size_t size, gfp_t flags,
628 if (unlikely(check_mul_overflow(n, size, &bytes)))
630 if (__builtin_constant_p(n) && __builtin_constant_p(size))
631 return kmalloc_node(bytes, flags, node);
632 return __kmalloc_node(bytes, flags, node);
635 static inline void *kcalloc_node(size_t n, size_t size, gfp_t flags, int node)
637 return kmalloc_array_node(n, size, flags | __GFP_ZERO, node);
642 extern void *__kmalloc_node_track_caller(size_t, gfp_t, int, unsigned long);
643 #define kmalloc_node_track_caller(size, flags, node) \
644 __kmalloc_node_track_caller(size, flags, node, \
647 #else /* CONFIG_NUMA */
649 #define kmalloc_node_track_caller(size, flags, node) \
650 kmalloc_track_caller(size, flags)
652 #endif /* CONFIG_NUMA */
657 static inline void *kmem_cache_zalloc(struct kmem_cache *k, gfp_t flags)
659 return kmem_cache_alloc(k, flags | __GFP_ZERO);
663 * kzalloc - allocate memory. The memory is set to zero.
664 * @size: how many bytes of memory are required.
665 * @flags: the type of memory to allocate (see kmalloc).
667 static inline void *kzalloc(size_t size, gfp_t flags)
669 return kmalloc(size, flags | __GFP_ZERO);
673 * kzalloc_node - allocate zeroed memory from a particular memory node.
674 * @size: how many bytes of memory are required.
675 * @flags: the type of memory to allocate (see kmalloc).
676 * @node: memory node from which to allocate
678 static inline void *kzalloc_node(size_t size, gfp_t flags, int node)
680 return kmalloc_node(size, flags | __GFP_ZERO, node);
683 unsigned int kmem_cache_size(struct kmem_cache *s);
684 void __init kmem_cache_init_late(void);
686 #if defined(CONFIG_SMP) && defined(CONFIG_SLAB)
687 int slab_prepare_cpu(unsigned int cpu);
688 int slab_dead_cpu(unsigned int cpu);
690 #define slab_prepare_cpu NULL
691 #define slab_dead_cpu NULL
694 #endif /* _LINUX_SLAB_H */