1 /* SPDX-License-Identifier: GPL-2.0 */
3 * Written by Mark Hemment, 1996 (markhe@nextd.demon.co.uk).
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
*);
159 * Please use this macro to create slab caches. Simply specify the
160 * name of the structure and maybe some flags that are listed above.
162 * The alignment of the struct determines object alignment. If you
163 * f.e. add ____cacheline_aligned_in_smp to the struct declaration
164 * then the objects will be properly aligned in SMP configurations.
166 #define KMEM_CACHE(__struct, __flags) \
167 kmem_cache_create(#__struct, sizeof(struct __struct), \
168 __alignof__(struct __struct), (__flags), NULL)
171 * To whitelist a single field for copying to/from usercopy, use this
172 * macro instead for KMEM_CACHE() above.
174 #define KMEM_CACHE_USERCOPY(__struct, __flags, __field) \
175 kmem_cache_create_usercopy(#__struct, \
176 sizeof(struct __struct), \
177 __alignof__(struct __struct), (__flags), \
178 offsetof(struct __struct, __field), \
179 sizeof_field(struct __struct, __field), NULL)
182 * Common kmalloc functions provided by all allocators
184 void * __must_check
krealloc(const void *, size_t, gfp_t
);
185 void kfree(const void *);
186 void kfree_sensitive(const void *);
187 size_t __ksize(const void *);
188 size_t ksize(const void *);
189 bool kmem_valid_obj(void *object
);
190 void kmem_dump_obj(void *object
);
192 #ifdef CONFIG_HAVE_HARDENED_USERCOPY_ALLOCATOR
193 void __check_heap_object(const void *ptr
, unsigned long n
, struct page
*page
,
196 static inline void __check_heap_object(const void *ptr
, unsigned long n
,
197 struct page
*page
, bool to_user
) { }
201 * Some archs want to perform DMA into kmalloc caches and need a guaranteed
202 * alignment larger than the alignment of a 64-bit integer.
203 * Setting ARCH_KMALLOC_MINALIGN in arch headers allows that.
205 #if defined(ARCH_DMA_MINALIGN) && ARCH_DMA_MINALIGN > 8
206 #define ARCH_KMALLOC_MINALIGN ARCH_DMA_MINALIGN
207 #define KMALLOC_MIN_SIZE ARCH_DMA_MINALIGN
208 #define KMALLOC_SHIFT_LOW ilog2(ARCH_DMA_MINALIGN)
210 #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
214 * Setting ARCH_SLAB_MINALIGN in arch headers allows a different alignment.
215 * Intended for arches that get misalignment faults even for 64 bit integer
218 #ifndef ARCH_SLAB_MINALIGN
219 #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
223 * kmalloc and friends return ARCH_KMALLOC_MINALIGN aligned
224 * pointers. kmem_cache_alloc and friends return ARCH_SLAB_MINALIGN
227 #define __assume_kmalloc_alignment __assume_aligned(ARCH_KMALLOC_MINALIGN)
228 #define __assume_slab_alignment __assume_aligned(ARCH_SLAB_MINALIGN)
229 #define __assume_page_alignment __assume_aligned(PAGE_SIZE)
232 * Kmalloc array related definitions
237 * The largest kmalloc size supported by the SLAB allocators is
238 * 32 megabyte (2^25) or the maximum allocatable page order if that is
241 * WARNING: Its not easy to increase this value since the allocators have
242 * to do various tricks to work around compiler limitations in order to
243 * ensure proper constant folding.
245 #define KMALLOC_SHIFT_HIGH ((MAX_ORDER + PAGE_SHIFT - 1) <= 25 ? \
246 (MAX_ORDER + PAGE_SHIFT - 1) : 25)
247 #define KMALLOC_SHIFT_MAX KMALLOC_SHIFT_HIGH
248 #ifndef KMALLOC_SHIFT_LOW
249 #define KMALLOC_SHIFT_LOW 5
255 * SLUB directly allocates requests fitting in to an order-1 page
256 * (PAGE_SIZE*2). Larger requests are passed to the page allocator.
258 #define KMALLOC_SHIFT_HIGH (PAGE_SHIFT + 1)
259 #define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT - 1)
260 #ifndef KMALLOC_SHIFT_LOW
261 #define KMALLOC_SHIFT_LOW 3
267 * SLOB passes all requests larger than one page to the page allocator.
268 * No kmalloc array is necessary since objects of different sizes can
269 * be allocated from the same page.
271 #define KMALLOC_SHIFT_HIGH PAGE_SHIFT
272 #define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT - 1)
273 #ifndef KMALLOC_SHIFT_LOW
274 #define KMALLOC_SHIFT_LOW 3
278 /* Maximum allocatable size */
279 #define KMALLOC_MAX_SIZE (1UL << KMALLOC_SHIFT_MAX)
280 /* Maximum size for which we actually use a slab cache */
281 #define KMALLOC_MAX_CACHE_SIZE (1UL << KMALLOC_SHIFT_HIGH)
282 /* Maximum order allocatable via the slab allocator */
283 #define KMALLOC_MAX_ORDER (KMALLOC_SHIFT_MAX - PAGE_SHIFT)
288 #ifndef KMALLOC_MIN_SIZE
289 #define KMALLOC_MIN_SIZE (1 << KMALLOC_SHIFT_LOW)
293 * This restriction comes from byte sized index implementation.
294 * Page size is normally 2^12 bytes and, in this case, if we want to use
295 * byte sized index which can represent 2^8 entries, the size of the object
296 * should be equal or greater to 2^12 / 2^8 = 2^4 = 16.
297 * If minimum size of kmalloc is less than 16, we use it as minimum object
298 * size and give up to use byte sized index.
300 #define SLAB_OBJ_MIN_SIZE (KMALLOC_MIN_SIZE < 16 ? \
301 (KMALLOC_MIN_SIZE) : 16)
304 * Whenever changing this, take care of that kmalloc_type() and
305 * create_kmalloc_caches() still work as intended.
307 enum kmalloc_cache_type
{
310 #ifdef CONFIG_ZONE_DMA
317 extern struct kmem_cache
*
318 kmalloc_caches
[NR_KMALLOC_TYPES
][KMALLOC_SHIFT_HIGH
+ 1];
320 static __always_inline
enum kmalloc_cache_type
kmalloc_type(gfp_t flags
)
322 #ifdef CONFIG_ZONE_DMA
324 * The most common case is KMALLOC_NORMAL, so test for it
325 * with a single branch for both flags.
327 if (likely((flags
& (__GFP_DMA
| __GFP_RECLAIMABLE
)) == 0))
328 return KMALLOC_NORMAL
;
331 * At least one of the flags has to be set. If both are, __GFP_DMA
334 return flags
& __GFP_DMA
? KMALLOC_DMA
: KMALLOC_RECLAIM
;
336 return flags
& __GFP_RECLAIMABLE
? KMALLOC_RECLAIM
: KMALLOC_NORMAL
;
341 * Figure out which kmalloc slab an allocation of a certain size
345 * 2 = 129 .. 192 bytes
346 * n = 2^(n-1)+1 .. 2^n
348 static __always_inline
unsigned int kmalloc_index(size_t size
)
353 if (size
<= KMALLOC_MIN_SIZE
)
354 return KMALLOC_SHIFT_LOW
;
356 if (KMALLOC_MIN_SIZE
<= 32 && size
> 64 && size
<= 96)
358 if (KMALLOC_MIN_SIZE
<= 64 && size
> 128 && size
<= 192)
360 if (size
<= 8) return 3;
361 if (size
<= 16) return 4;
362 if (size
<= 32) return 5;
363 if (size
<= 64) return 6;
364 if (size
<= 128) return 7;
365 if (size
<= 256) return 8;
366 if (size
<= 512) return 9;
367 if (size
<= 1024) return 10;
368 if (size
<= 2 * 1024) return 11;
369 if (size
<= 4 * 1024) return 12;
370 if (size
<= 8 * 1024) return 13;
371 if (size
<= 16 * 1024) return 14;
372 if (size
<= 32 * 1024) return 15;
373 if (size
<= 64 * 1024) return 16;
374 if (size
<= 128 * 1024) return 17;
375 if (size
<= 256 * 1024) return 18;
376 if (size
<= 512 * 1024) return 19;
377 if (size
<= 1024 * 1024) return 20;
378 if (size
<= 2 * 1024 * 1024) return 21;
379 if (size
<= 4 * 1024 * 1024) return 22;
380 if (size
<= 8 * 1024 * 1024) return 23;
381 if (size
<= 16 * 1024 * 1024) return 24;
382 if (size
<= 32 * 1024 * 1024) return 25;
383 if (size
<= 64 * 1024 * 1024) return 26;
386 /* Will never be reached. Needed because the compiler may complain */
389 #endif /* !CONFIG_SLOB */
391 void *__kmalloc(size_t size
, gfp_t flags
) __assume_kmalloc_alignment __malloc
;
392 void *kmem_cache_alloc(struct kmem_cache
*, gfp_t flags
) __assume_slab_alignment __malloc
;
393 void kmem_cache_free(struct kmem_cache
*, void *);
396 * Bulk allocation and freeing operations. These are accelerated in an
397 * allocator specific way to avoid taking locks repeatedly or building
398 * metadata structures unnecessarily.
400 * Note that interrupts must be enabled when calling these functions.
402 void kmem_cache_free_bulk(struct kmem_cache
*, size_t, void **);
403 int kmem_cache_alloc_bulk(struct kmem_cache
*, gfp_t
, size_t, void **);
406 * Caller must not use kfree_bulk() on memory not originally allocated
407 * by kmalloc(), because the SLOB allocator cannot handle this.
409 static __always_inline
void kfree_bulk(size_t size
, void **p
)
411 kmem_cache_free_bulk(NULL
, size
, p
);
415 void *__kmalloc_node(size_t size
, gfp_t flags
, int node
) __assume_kmalloc_alignment __malloc
;
416 void *kmem_cache_alloc_node(struct kmem_cache
*, gfp_t flags
, int node
) __assume_slab_alignment __malloc
;
418 static __always_inline
void *__kmalloc_node(size_t size
, gfp_t flags
, int node
)
420 return __kmalloc(size
, flags
);
423 static __always_inline
void *kmem_cache_alloc_node(struct kmem_cache
*s
, gfp_t flags
, int node
)
425 return kmem_cache_alloc(s
, flags
);
429 #ifdef CONFIG_TRACING
430 extern void *kmem_cache_alloc_trace(struct kmem_cache
*, gfp_t
, size_t) __assume_slab_alignment __malloc
;
433 extern void *kmem_cache_alloc_node_trace(struct kmem_cache
*s
,
435 int node
, size_t size
) __assume_slab_alignment __malloc
;
437 static __always_inline
void *
438 kmem_cache_alloc_node_trace(struct kmem_cache
*s
,
440 int node
, size_t size
)
442 return kmem_cache_alloc_trace(s
, gfpflags
, size
);
444 #endif /* CONFIG_NUMA */
446 #else /* CONFIG_TRACING */
447 static __always_inline
void *kmem_cache_alloc_trace(struct kmem_cache
*s
,
448 gfp_t flags
, size_t size
)
450 void *ret
= kmem_cache_alloc(s
, flags
);
452 ret
= kasan_kmalloc(s
, ret
, size
, flags
);
456 static __always_inline
void *
457 kmem_cache_alloc_node_trace(struct kmem_cache
*s
,
459 int node
, size_t size
)
461 void *ret
= kmem_cache_alloc_node(s
, gfpflags
, node
);
463 ret
= kasan_kmalloc(s
, ret
, size
, gfpflags
);
466 #endif /* CONFIG_TRACING */
468 extern void *kmalloc_order(size_t size
, gfp_t flags
, unsigned int order
) __assume_page_alignment __malloc
;
470 #ifdef CONFIG_TRACING
471 extern void *kmalloc_order_trace(size_t size
, gfp_t flags
, unsigned int order
) __assume_page_alignment __malloc
;
473 static __always_inline
void *
474 kmalloc_order_trace(size_t size
, gfp_t flags
, unsigned int order
)
476 return kmalloc_order(size
, flags
, order
);
480 static __always_inline
void *kmalloc_large(size_t size
, gfp_t flags
)
482 unsigned int order
= get_order(size
);
483 return kmalloc_order_trace(size
, flags
, order
);
487 * kmalloc - allocate memory
488 * @size: how many bytes of memory are required.
489 * @flags: the type of memory to allocate.
491 * kmalloc is the normal method of allocating memory
492 * for objects smaller than page size in the kernel.
494 * The allocated object address is aligned to at least ARCH_KMALLOC_MINALIGN
495 * bytes. For @size of power of two bytes, the alignment is also guaranteed
496 * to be at least to the size.
498 * The @flags argument may be one of the GFP flags defined at
499 * include/linux/gfp.h and described at
500 * :ref:`Documentation/core-api/mm-api.rst <mm-api-gfp-flags>`
502 * The recommended usage of the @flags is described at
503 * :ref:`Documentation/core-api/memory-allocation.rst <memory_allocation>`
505 * Below is a brief outline of the most useful GFP flags
508 * Allocate normal kernel ram. May sleep.
511 * Allocation will not sleep.
514 * Allocation will not sleep. May use emergency pools.
517 * Allocate memory from high memory on behalf of user.
519 * Also it is possible to set different flags by OR'ing
520 * in one or more of the following additional @flags:
523 * This allocation has high priority and may use emergency pools.
526 * Indicate that this allocation is in no way allowed to fail
527 * (think twice before using).
530 * If memory is not immediately available,
531 * then give up at once.
534 * If allocation fails, don't issue any warnings.
536 * %__GFP_RETRY_MAYFAIL
537 * Try really hard to succeed the allocation but fail
540 static __always_inline
void *kmalloc(size_t size
, gfp_t flags
)
542 if (__builtin_constant_p(size
)) {
546 if (size
> KMALLOC_MAX_CACHE_SIZE
)
547 return kmalloc_large(size
, flags
);
549 index
= kmalloc_index(size
);
552 return ZERO_SIZE_PTR
;
554 return kmem_cache_alloc_trace(
555 kmalloc_caches
[kmalloc_type(flags
)][index
],
559 return __kmalloc(size
, flags
);
562 static __always_inline
void *kmalloc_node(size_t size
, gfp_t flags
, int node
)
565 if (__builtin_constant_p(size
) &&
566 size
<= KMALLOC_MAX_CACHE_SIZE
) {
567 unsigned int i
= kmalloc_index(size
);
570 return ZERO_SIZE_PTR
;
572 return kmem_cache_alloc_node_trace(
573 kmalloc_caches
[kmalloc_type(flags
)][i
],
577 return __kmalloc_node(size
, flags
, node
);
581 * kmalloc_array - allocate memory for an array.
582 * @n: number of elements.
583 * @size: element size.
584 * @flags: the type of memory to allocate (see kmalloc).
586 static inline void *kmalloc_array(size_t n
, size_t size
, gfp_t flags
)
590 if (unlikely(check_mul_overflow(n
, size
, &bytes
)))
592 if (__builtin_constant_p(n
) && __builtin_constant_p(size
))
593 return kmalloc(bytes
, flags
);
594 return __kmalloc(bytes
, flags
);
598 * krealloc_array - reallocate memory for an array.
599 * @p: pointer to the memory chunk to reallocate
600 * @new_n: new number of elements to alloc
601 * @new_size: new size of a single member of the array
602 * @flags: the type of memory to allocate (see kmalloc)
604 static __must_check
inline void *
605 krealloc_array(void *p
, size_t new_n
, size_t new_size
, gfp_t flags
)
609 if (unlikely(check_mul_overflow(new_n
, new_size
, &bytes
)))
612 return krealloc(p
, bytes
, flags
);
616 * kcalloc - allocate memory for an array. The memory is set to zero.
617 * @n: number of elements.
618 * @size: element size.
619 * @flags: the type of memory to allocate (see kmalloc).
621 static inline void *kcalloc(size_t n
, size_t size
, gfp_t flags
)
623 return kmalloc_array(n
, size
, flags
| __GFP_ZERO
);
627 * kmalloc_track_caller is a special version of kmalloc that records the
628 * calling function of the routine calling it for slab leak tracking instead
629 * of just the calling function (confusing, eh?).
630 * It's useful when the call to kmalloc comes from a widely-used standard
631 * allocator where we care about the real place the memory allocation
632 * request comes from.
634 extern void *__kmalloc_track_caller(size_t, gfp_t
, unsigned long);
635 #define kmalloc_track_caller(size, flags) \
636 __kmalloc_track_caller(size, flags, _RET_IP_)
638 static inline void *kmalloc_array_node(size_t n
, size_t size
, gfp_t flags
,
643 if (unlikely(check_mul_overflow(n
, size
, &bytes
)))
645 if (__builtin_constant_p(n
) && __builtin_constant_p(size
))
646 return kmalloc_node(bytes
, flags
, node
);
647 return __kmalloc_node(bytes
, flags
, node
);
650 static inline void *kcalloc_node(size_t n
, size_t size
, gfp_t flags
, int node
)
652 return kmalloc_array_node(n
, size
, flags
| __GFP_ZERO
, node
);
657 extern void *__kmalloc_node_track_caller(size_t, gfp_t
, int, unsigned long);
658 #define kmalloc_node_track_caller(size, flags, node) \
659 __kmalloc_node_track_caller(size, flags, node, \
662 #else /* CONFIG_NUMA */
664 #define kmalloc_node_track_caller(size, flags, node) \
665 kmalloc_track_caller(size, flags)
667 #endif /* CONFIG_NUMA */
672 static inline void *kmem_cache_zalloc(struct kmem_cache
*k
, gfp_t flags
)
674 return kmem_cache_alloc(k
, flags
| __GFP_ZERO
);
678 * kzalloc - allocate memory. The memory is set to zero.
679 * @size: how many bytes of memory are required.
680 * @flags: the type of memory to allocate (see kmalloc).
682 static inline void *kzalloc(size_t size
, gfp_t flags
)
684 return kmalloc(size
, flags
| __GFP_ZERO
);
688 * kzalloc_node - allocate zeroed memory from a particular memory node.
689 * @size: how many bytes of memory are required.
690 * @flags: the type of memory to allocate (see kmalloc).
691 * @node: memory node from which to allocate
693 static inline void *kzalloc_node(size_t size
, gfp_t flags
, int node
)
695 return kmalloc_node(size
, flags
| __GFP_ZERO
, node
);
698 unsigned int kmem_cache_size(struct kmem_cache
*s
);
699 void __init
kmem_cache_init_late(void);
701 #if defined(CONFIG_SMP) && defined(CONFIG_SLAB)
702 int slab_prepare_cpu(unsigned int cpu
);
703 int slab_dead_cpu(unsigned int cpu
);
705 #define slab_prepare_cpu NULL
706 #define slab_dead_cpu NULL
709 #endif /* _LINUX_SLAB_H */