1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3 * Written by Mark Hemment, 1996 (markhe@nextd.demon.co.uk).
4 *
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
10 */
11
12 #ifndef _LINUX_SLAB_H
13 #define _LINUX_SLAB_H
14
15 #include <linux/cache.h>
16 #include <linux/gfp.h>
17 #include <linux/overflow.h>
18 #include <linux/types.h>
19 #include <linux/workqueue.h>
20 #include <linux/percpu-refcount.h>
21 #include <linux/cleanup.h>
22 #include <linux/hash.h>
23
24
25 /*
26 * Flags to pass to kmem_cache_create().
27 * The ones marked DEBUG are only valid if CONFIG_DEBUG_SLAB is set.
28 */
29 /* DEBUG: Perform (expensive) checks on alloc/free */
30 #define SLAB_CONSISTENCY_CHECKS ((slab_flags_t __force)0x00000100U)
31 /* DEBUG: Red zone objs in a cache */
32 #define SLAB_RED_ZONE ((slab_flags_t __force)0x00000400U)
33 /* DEBUG: Poison objects */
34 #define SLAB_POISON ((slab_flags_t __force)0x00000800U)
35 /* Indicate a kmalloc slab */
36 #define SLAB_KMALLOC ((slab_flags_t __force)0x00001000U)
37 /* Align objs on cache lines */
38 #define SLAB_HWCACHE_ALIGN ((slab_flags_t __force)0x00002000U)
39 /* Use GFP_DMA memory */
40 #define SLAB_CACHE_DMA ((slab_flags_t __force)0x00004000U)
41 /* Use GFP_DMA32 memory */
42 #define SLAB_CACHE_DMA32 ((slab_flags_t __force)0x00008000U)
43 /* DEBUG: Store the last owner for bug hunting */
44 #define SLAB_STORE_USER ((slab_flags_t __force)0x00010000U)
45 /* Panic if kmem_cache_create() fails */
46 #define SLAB_PANIC ((slab_flags_t __force)0x00040000U)
47 /*
48 * SLAB_TYPESAFE_BY_RCU - **WARNING** READ THIS!
49 *
50 * This delays freeing the SLAB page by a grace period, it does _NOT_
51 * delay object freeing. This means that if you do kmem_cache_free()
52 * that memory location is free to be reused at any time. Thus it may
53 * be possible to see another object there in the same RCU grace period.
54 *
55 * This feature only ensures the memory location backing the object
56 * stays valid, the trick to using this is relying on an independent
57 * object validation pass. Something like:
58 *
59 * begin:
60 * rcu_read_lock();
61 * obj = lockless_lookup(key);
62 * if (obj) {
63 * if (!try_get_ref(obj)) // might fail for free objects
64 * rcu_read_unlock();
65 * goto begin;
66 *
67 * if (obj->key != key) { // not the object we expected
68 * put_ref(obj);
69 * rcu_read_unlock();
70 * goto begin;
71 * }
72 * }
73 * rcu_read_unlock();
74 *
75 * This is useful if we need to approach a kernel structure obliquely,
76 * from its address obtained without the usual locking. We can lock
77 * the structure to stabilize it and check it's still at the given address,
78 * only if we can be sure that the memory has not been meanwhile reused
79 * for some other kind of object (which our subsystem's lock might corrupt).
80 *
81 * rcu_read_lock before reading the address, then rcu_read_unlock after
82 * taking the spinlock within the structure expected at that address.
83 *
84 * Note that it is not possible to acquire a lock within a structure
85 * allocated with SLAB_TYPESAFE_BY_RCU without first acquiring a reference
86 * as described above. The reason is that SLAB_TYPESAFE_BY_RCU pages
87 * are not zeroed before being given to the slab, which means that any
88 * locks must be initialized after each and every kmem_struct_alloc().
89 * Alternatively, make the ctor passed to kmem_cache_create() initialize
90 * the locks at page-allocation time, as is done in __i915_request_ctor(),
91 * sighand_ctor(), and anon_vma_ctor(). Such a ctor permits readers
92 * to safely acquire those ctor-initialized locks under rcu_read_lock()
93 * protection.
94 *
95 * Note that SLAB_TYPESAFE_BY_RCU was originally named SLAB_DESTROY_BY_RCU.
96 */
97 /* Defer freeing slabs to RCU */
98 #define SLAB_TYPESAFE_BY_RCU ((slab_flags_t __force)0x00080000U)
99 /* Spread some memory over cpuset */
100 #define SLAB_MEM_SPREAD ((slab_flags_t __force)0x00100000U)
101 /* Trace allocations and frees */
102 #define SLAB_TRACE ((slab_flags_t __force)0x00200000U)
103
104 /* Flag to prevent checks on free */
105 #ifdef CONFIG_DEBUG_OBJECTS
106 # define SLAB_DEBUG_OBJECTS ((slab_flags_t __force)0x00400000U)
107 #else
108 # define SLAB_DEBUG_OBJECTS 0
109 #endif
110
111 /* Avoid kmemleak tracing */
112 #define SLAB_NOLEAKTRACE ((slab_flags_t __force)0x00800000U)
113
114 /*
115 * Prevent merging with compatible kmem caches. This flag should be used
116 * cautiously. Valid use cases:
117 *
118 * - caches created for self-tests (e.g. kunit)
119 * - general caches created and used by a subsystem, only when a
120 * (subsystem-specific) debug option is enabled
121 * - performance critical caches, should be very rare and consulted with slab
122 * maintainers, and not used together with CONFIG_SLUB_TINY
123 */
124 #define SLAB_NO_MERGE ((slab_flags_t __force)0x01000000U)
125
126 /* Fault injection mark */
127 #ifdef CONFIG_FAILSLAB
128 # define SLAB_FAILSLAB ((slab_flags_t __force)0x02000000U)
129 #else
130 # define SLAB_FAILSLAB 0
131 #endif
132 /* Account to memcg */
133 #ifdef CONFIG_MEMCG_KMEM
134 # define SLAB_ACCOUNT ((slab_flags_t __force)0x04000000U)
135 #else
136 # define SLAB_ACCOUNT 0
137 #endif
138
139 #ifdef CONFIG_KASAN_GENERIC
140 #define SLAB_KASAN ((slab_flags_t __force)0x08000000U)
141 #else
142 #define SLAB_KASAN 0
143 #endif
144
145 /*
146 * Ignore user specified debugging flags.
147 * Intended for caches created for self-tests so they have only flags
148 * specified in the code and other flags are ignored.
149 */
150 #define SLAB_NO_USER_FLAGS ((slab_flags_t __force)0x10000000U)
151
152 #ifdef CONFIG_KFENCE
153 #define SLAB_SKIP_KFENCE ((slab_flags_t __force)0x20000000U)
154 #else
155 #define SLAB_SKIP_KFENCE 0
156 #endif
157
158 /* The following flags affect the page allocator grouping pages by mobility */
159 /* Objects are reclaimable */
160 #ifndef CONFIG_SLUB_TINY
161 #define SLAB_RECLAIM_ACCOUNT ((slab_flags_t __force)0x00020000U)
162 #else
163 #define SLAB_RECLAIM_ACCOUNT ((slab_flags_t __force)0)
164 #endif
165 #define SLAB_TEMPORARY SLAB_RECLAIM_ACCOUNT /* Objects are short-lived */
166
167 /*
168 * ZERO_SIZE_PTR will be returned for zero sized kmalloc requests.
169 *
170 * Dereferencing ZERO_SIZE_PTR will lead to a distinct access fault.
171 *
172 * ZERO_SIZE_PTR can be passed to kfree though in the same way that NULL can.
173 * Both make kfree a no-op.
174 */
175 #define ZERO_SIZE_PTR ((void *)16)
176
177 #define ZERO_OR_NULL_PTR(x) ((unsigned long)(x) <= \
178 (unsigned long)ZERO_SIZE_PTR)
179
180 #include <linux/kasan.h>
181
182 struct list_lru;
183 struct mem_cgroup;
184 /*
185 * struct kmem_cache related prototypes
186 */
187 bool slab_is_available(void);
188
189 struct kmem_cache *kmem_cache_create(const char *name, unsigned int size,
190 unsigned int align, slab_flags_t flags,
191 void (*ctor)(void *));
192 struct kmem_cache *kmem_cache_create_usercopy(const char *name,
193 unsigned int size, unsigned int align,
194 slab_flags_t flags,
195 unsigned int useroffset, unsigned int usersize,
196 void (*ctor)(void *));
197 void kmem_cache_destroy(struct kmem_cache *s);
198 int kmem_cache_shrink(struct kmem_cache *s);
199
200 /*
201 * Please use this macro to create slab caches. Simply specify the
202 * name of the structure and maybe some flags that are listed above.
203 *
204 * The alignment of the struct determines object alignment. If you
205 * f.e. add ____cacheline_aligned_in_smp to the struct declaration
206 * then the objects will be properly aligned in SMP configurations.
207 */
208 #define KMEM_CACHE(__struct, __flags) \
209 kmem_cache_create(#__struct, sizeof(struct __struct), \
210 __alignof__(struct __struct), (__flags), NULL)
211
212 /*
213 * To whitelist a single field for copying to/from usercopy, use this
214 * macro instead for KMEM_CACHE() above.
215 */
216 #define KMEM_CACHE_USERCOPY(__struct, __flags, __field) \
217 kmem_cache_create_usercopy(#__struct, \
218 sizeof(struct __struct), \
219 __alignof__(struct __struct), (__flags), \
220 offsetof(struct __struct, __field), \
221 sizeof_field(struct __struct, __field), NULL)
222
223 /*
224 * Common kmalloc functions provided by all allocators
225 */
226 void * __must_check krealloc(const void *objp, size_t new_size, gfp_t flags) __realloc_size(2);
227 void kfree(const void *objp);
228 void kfree_sensitive(const void *objp);
229 size_t __ksize(const void *objp);
230
231 DEFINE_FREE(kfree, void *, if (!IS_ERR_OR_NULL(_T)) kfree(_T))
232
233 /**
234 * ksize - Report actual allocation size of associated object
235 *
236 * @objp: Pointer returned from a prior kmalloc()-family allocation.
237 *
238 * This should not be used for writing beyond the originally requested
239 * allocation size. Either use krealloc() or round up the allocation size
240 * with kmalloc_size_roundup() prior to allocation. If this is used to
241 * access beyond the originally requested allocation size, UBSAN_BOUNDS
242 * and/or FORTIFY_SOURCE may trip, since they only know about the
243 * originally allocated size via the __alloc_size attribute.
244 */
245 size_t ksize(const void *objp);
246
247 #ifdef CONFIG_PRINTK
248 bool kmem_dump_obj(void *object);
249 #else
kmem_dump_obj(void * object)250 static inline bool kmem_dump_obj(void *object) { return false; }
251 #endif
252
253 /*
254 * Some archs want to perform DMA into kmalloc caches and need a guaranteed
255 * alignment larger than the alignment of a 64-bit integer.
256 * Setting ARCH_DMA_MINALIGN in arch headers allows that.
257 */
258 #ifdef ARCH_HAS_DMA_MINALIGN
259 #if ARCH_DMA_MINALIGN > 8 && !defined(ARCH_KMALLOC_MINALIGN)
260 #define ARCH_KMALLOC_MINALIGN ARCH_DMA_MINALIGN
261 #endif
262 #endif
263
264 #ifndef ARCH_KMALLOC_MINALIGN
265 #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
266 #elif ARCH_KMALLOC_MINALIGN > 8
267 #define KMALLOC_MIN_SIZE ARCH_KMALLOC_MINALIGN
268 #define KMALLOC_SHIFT_LOW ilog2(KMALLOC_MIN_SIZE)
269 #endif
270
271 /*
272 * Setting ARCH_SLAB_MINALIGN in arch headers allows a different alignment.
273 * Intended for arches that get misalignment faults even for 64 bit integer
274 * aligned buffers.
275 */
276 #ifndef ARCH_SLAB_MINALIGN
277 #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
278 #endif
279
280 /*
281 * Arches can define this function if they want to decide the minimum slab
282 * alignment at runtime. The value returned by the function must be a power
283 * of two and >= ARCH_SLAB_MINALIGN.
284 */
285 #ifndef arch_slab_minalign
arch_slab_minalign(void)286 static inline unsigned int arch_slab_minalign(void)
287 {
288 return ARCH_SLAB_MINALIGN;
289 }
290 #endif
291
292 /*
293 * kmem_cache_alloc and friends return pointers aligned to ARCH_SLAB_MINALIGN.
294 * kmalloc and friends return pointers aligned to both ARCH_KMALLOC_MINALIGN
295 * and ARCH_SLAB_MINALIGN, but here we only assume the former alignment.
296 */
297 #define __assume_kmalloc_alignment __assume_aligned(ARCH_KMALLOC_MINALIGN)
298 #define __assume_slab_alignment __assume_aligned(ARCH_SLAB_MINALIGN)
299 #define __assume_page_alignment __assume_aligned(PAGE_SIZE)
300
301 /*
302 * Kmalloc array related definitions
303 */
304
305 #ifdef CONFIG_SLAB
306 /*
307 * SLAB and SLUB directly allocates requests fitting in to an order-1 page
308 * (PAGE_SIZE*2). Larger requests are passed to the page allocator.
309 */
310 #define KMALLOC_SHIFT_HIGH (PAGE_SHIFT + 1)
311 #define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT)
312 #ifndef KMALLOC_SHIFT_LOW
313 #define KMALLOC_SHIFT_LOW 5
314 #endif
315 #endif
316
317 #ifdef CONFIG_SLUB
318 #define KMALLOC_SHIFT_HIGH (PAGE_SHIFT + 1)
319 #define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT)
320 #ifndef KMALLOC_SHIFT_LOW
321 #define KMALLOC_SHIFT_LOW 3
322 #endif
323 #endif
324
325 /* Maximum allocatable size */
326 #define KMALLOC_MAX_SIZE (1UL << KMALLOC_SHIFT_MAX)
327 /* Maximum size for which we actually use a slab cache */
328 #define KMALLOC_MAX_CACHE_SIZE (1UL << KMALLOC_SHIFT_HIGH)
329 /* Maximum order allocatable via the slab allocator */
330 #define KMALLOC_MAX_ORDER (KMALLOC_SHIFT_MAX - PAGE_SHIFT)
331
332 /*
333 * Kmalloc subsystem.
334 */
335 #ifndef KMALLOC_MIN_SIZE
336 #define KMALLOC_MIN_SIZE (1 << KMALLOC_SHIFT_LOW)
337 #endif
338
339 /*
340 * This restriction comes from byte sized index implementation.
341 * Page size is normally 2^12 bytes and, in this case, if we want to use
342 * byte sized index which can represent 2^8 entries, the size of the object
343 * should be equal or greater to 2^12 / 2^8 = 2^4 = 16.
344 * If minimum size of kmalloc is less than 16, we use it as minimum object
345 * size and give up to use byte sized index.
346 */
347 #define SLAB_OBJ_MIN_SIZE (KMALLOC_MIN_SIZE < 16 ? \
348 (KMALLOC_MIN_SIZE) : 16)
349
350 #ifdef CONFIG_RANDOM_KMALLOC_CACHES
351 #define RANDOM_KMALLOC_CACHES_NR 15 // # of cache copies
352 #else
353 #define RANDOM_KMALLOC_CACHES_NR 0
354 #endif
355
356 /*
357 * Whenever changing this, take care of that kmalloc_type() and
358 * create_kmalloc_caches() still work as intended.
359 *
360 * KMALLOC_NORMAL can contain only unaccounted objects whereas KMALLOC_CGROUP
361 * is for accounted but unreclaimable and non-dma objects. All the other
362 * kmem caches can have both accounted and unaccounted objects.
363 */
364 enum kmalloc_cache_type {
365 KMALLOC_NORMAL = 0,
366 #ifndef CONFIG_ZONE_DMA
367 KMALLOC_DMA = KMALLOC_NORMAL,
368 #endif
369 #ifndef CONFIG_MEMCG_KMEM
370 KMALLOC_CGROUP = KMALLOC_NORMAL,
371 #endif
372 KMALLOC_RANDOM_START = KMALLOC_NORMAL,
373 KMALLOC_RANDOM_END = KMALLOC_RANDOM_START + RANDOM_KMALLOC_CACHES_NR,
374 #ifdef CONFIG_SLUB_TINY
375 KMALLOC_RECLAIM = KMALLOC_NORMAL,
376 #else
377 KMALLOC_RECLAIM,
378 #endif
379 #ifdef CONFIG_ZONE_DMA
380 KMALLOC_DMA,
381 #endif
382 #ifdef CONFIG_MEMCG_KMEM
383 KMALLOC_CGROUP,
384 #endif
385 NR_KMALLOC_TYPES
386 };
387
388 extern struct kmem_cache *
389 kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1];
390
391 /*
392 * Define gfp bits that should not be set for KMALLOC_NORMAL.
393 */
394 #define KMALLOC_NOT_NORMAL_BITS \
395 (__GFP_RECLAIMABLE | \
396 (IS_ENABLED(CONFIG_ZONE_DMA) ? __GFP_DMA : 0) | \
397 (IS_ENABLED(CONFIG_MEMCG_KMEM) ? __GFP_ACCOUNT : 0))
398
399 extern unsigned long random_kmalloc_seed;
400
kmalloc_type(gfp_t flags,unsigned long caller)401 static __always_inline enum kmalloc_cache_type kmalloc_type(gfp_t flags, unsigned long caller)
402 {
403 /*
404 * The most common case is KMALLOC_NORMAL, so test for it
405 * with a single branch for all the relevant flags.
406 */
407 if (likely((flags & KMALLOC_NOT_NORMAL_BITS) == 0))
408 #ifdef CONFIG_RANDOM_KMALLOC_CACHES
409 /* RANDOM_KMALLOC_CACHES_NR (=15) copies + the KMALLOC_NORMAL */
410 return KMALLOC_RANDOM_START + hash_64(caller ^ random_kmalloc_seed,
411 ilog2(RANDOM_KMALLOC_CACHES_NR + 1));
412 #else
413 return KMALLOC_NORMAL;
414 #endif
415
416 /*
417 * At least one of the flags has to be set. Their priorities in
418 * decreasing order are:
419 * 1) __GFP_DMA
420 * 2) __GFP_RECLAIMABLE
421 * 3) __GFP_ACCOUNT
422 */
423 if (IS_ENABLED(CONFIG_ZONE_DMA) && (flags & __GFP_DMA))
424 return KMALLOC_DMA;
425 if (!IS_ENABLED(CONFIG_MEMCG_KMEM) || (flags & __GFP_RECLAIMABLE))
426 return KMALLOC_RECLAIM;
427 else
428 return KMALLOC_CGROUP;
429 }
430
431 /*
432 * Figure out which kmalloc slab an allocation of a certain size
433 * belongs to.
434 * 0 = zero alloc
435 * 1 = 65 .. 96 bytes
436 * 2 = 129 .. 192 bytes
437 * n = 2^(n-1)+1 .. 2^n
438 *
439 * Note: __kmalloc_index() is compile-time optimized, and not runtime optimized;
440 * typical usage is via kmalloc_index() and therefore evaluated at compile-time.
441 * Callers where !size_is_constant should only be test modules, where runtime
442 * overheads of __kmalloc_index() can be tolerated. Also see kmalloc_slab().
443 */
__kmalloc_index(size_t size,bool size_is_constant)444 static __always_inline unsigned int __kmalloc_index(size_t size,
445 bool size_is_constant)
446 {
447 if (!size)
448 return 0;
449
450 if (size <= KMALLOC_MIN_SIZE)
451 return KMALLOC_SHIFT_LOW;
452
453 if (KMALLOC_MIN_SIZE <= 32 && size > 64 && size <= 96)
454 return 1;
455 if (KMALLOC_MIN_SIZE <= 64 && size > 128 && size <= 192)
456 return 2;
457 if (size <= 8) return 3;
458 if (size <= 16) return 4;
459 if (size <= 32) return 5;
460 if (size <= 64) return 6;
461 if (size <= 128) return 7;
462 if (size <= 256) return 8;
463 if (size <= 512) return 9;
464 if (size <= 1024) return 10;
465 if (size <= 2 * 1024) return 11;
466 if (size <= 4 * 1024) return 12;
467 if (size <= 8 * 1024) return 13;
468 if (size <= 16 * 1024) return 14;
469 if (size <= 32 * 1024) return 15;
470 if (size <= 64 * 1024) return 16;
471 if (size <= 128 * 1024) return 17;
472 if (size <= 256 * 1024) return 18;
473 if (size <= 512 * 1024) return 19;
474 if (size <= 1024 * 1024) return 20;
475 if (size <= 2 * 1024 * 1024) return 21;
476
477 if (!IS_ENABLED(CONFIG_PROFILE_ALL_BRANCHES) && size_is_constant)
478 BUILD_BUG_ON_MSG(1, "unexpected size in kmalloc_index()");
479 else
480 BUG();
481
482 /* Will never be reached. Needed because the compiler may complain */
483 return -1;
484 }
485 static_assert(PAGE_SHIFT <= 20);
486 #define kmalloc_index(s) __kmalloc_index(s, true)
487
488 void *__kmalloc(size_t size, gfp_t flags) __assume_kmalloc_alignment __alloc_size(1);
489
490 /**
491 * kmem_cache_alloc - Allocate an object
492 * @cachep: The cache to allocate from.
493 * @flags: See kmalloc().
494 *
495 * Allocate an object from this cache.
496 * See kmem_cache_zalloc() for a shortcut of adding __GFP_ZERO to flags.
497 *
498 * Return: pointer to the new object or %NULL in case of error
499 */
500 void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags) __assume_slab_alignment __malloc;
501 void *kmem_cache_alloc_lru(struct kmem_cache *s, struct list_lru *lru,
502 gfp_t gfpflags) __assume_slab_alignment __malloc;
503 void kmem_cache_free(struct kmem_cache *s, void *objp);
504
505 /*
506 * Bulk allocation and freeing operations. These are accelerated in an
507 * allocator specific way to avoid taking locks repeatedly or building
508 * metadata structures unnecessarily.
509 *
510 * Note that interrupts must be enabled when calling these functions.
511 */
512 void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p);
513 int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, void **p);
514
kfree_bulk(size_t size,void ** p)515 static __always_inline void kfree_bulk(size_t size, void **p)
516 {
517 kmem_cache_free_bulk(NULL, size, p);
518 }
519
520 void *__kmalloc_node(size_t size, gfp_t flags, int node) __assume_kmalloc_alignment
521 __alloc_size(1);
522 void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t flags, int node) __assume_slab_alignment
523 __malloc;
524
525 void *kmalloc_trace(struct kmem_cache *s, gfp_t flags, size_t size)
526 __assume_kmalloc_alignment __alloc_size(3);
527
528 void *kmalloc_node_trace(struct kmem_cache *s, gfp_t gfpflags,
529 int node, size_t size) __assume_kmalloc_alignment
530 __alloc_size(4);
531 void *kmalloc_large(size_t size, gfp_t flags) __assume_page_alignment
532 __alloc_size(1);
533
534 void *kmalloc_large_node(size_t size, gfp_t flags, int node) __assume_page_alignment
535 __alloc_size(1);
536
537 /**
538 * kmalloc - allocate kernel memory
539 * @size: how many bytes of memory are required.
540 * @flags: describe the allocation context
541 *
542 * kmalloc is the normal method of allocating memory
543 * for objects smaller than page size in the kernel.
544 *
545 * The allocated object address is aligned to at least ARCH_KMALLOC_MINALIGN
546 * bytes. For @size of power of two bytes, the alignment is also guaranteed
547 * to be at least to the size.
548 *
549 * The @flags argument may be one of the GFP flags defined at
550 * include/linux/gfp_types.h and described at
551 * :ref:`Documentation/core-api/mm-api.rst <mm-api-gfp-flags>`
552 *
553 * The recommended usage of the @flags is described at
554 * :ref:`Documentation/core-api/memory-allocation.rst <memory_allocation>`
555 *
556 * Below is a brief outline of the most useful GFP flags
557 *
558 * %GFP_KERNEL
559 * Allocate normal kernel ram. May sleep.
560 *
561 * %GFP_NOWAIT
562 * Allocation will not sleep.
563 *
564 * %GFP_ATOMIC
565 * Allocation will not sleep. May use emergency pools.
566 *
567 * Also it is possible to set different flags by OR'ing
568 * in one or more of the following additional @flags:
569 *
570 * %__GFP_ZERO
571 * Zero the allocated memory before returning. Also see kzalloc().
572 *
573 * %__GFP_HIGH
574 * This allocation has high priority and may use emergency pools.
575 *
576 * %__GFP_NOFAIL
577 * Indicate that this allocation is in no way allowed to fail
578 * (think twice before using).
579 *
580 * %__GFP_NORETRY
581 * If memory is not immediately available,
582 * then give up at once.
583 *
584 * %__GFP_NOWARN
585 * If allocation fails, don't issue any warnings.
586 *
587 * %__GFP_RETRY_MAYFAIL
588 * Try really hard to succeed the allocation but fail
589 * eventually.
590 */
kmalloc(size_t size,gfp_t flags)591 static __always_inline __alloc_size(1) void *kmalloc(size_t size, gfp_t flags)
592 {
593 if (__builtin_constant_p(size) && size) {
594 unsigned int index;
595
596 if (size > KMALLOC_MAX_CACHE_SIZE)
597 return kmalloc_large(size, flags);
598
599 index = kmalloc_index(size);
600 return kmalloc_trace(
601 kmalloc_caches[kmalloc_type(flags, _RET_IP_)][index],
602 flags, size);
603 }
604 return __kmalloc(size, flags);
605 }
606
kmalloc_node(size_t size,gfp_t flags,int node)607 static __always_inline __alloc_size(1) void *kmalloc_node(size_t size, gfp_t flags, int node)
608 {
609 if (__builtin_constant_p(size) && size) {
610 unsigned int index;
611
612 if (size > KMALLOC_MAX_CACHE_SIZE)
613 return kmalloc_large_node(size, flags, node);
614
615 index = kmalloc_index(size);
616 return kmalloc_node_trace(
617 kmalloc_caches[kmalloc_type(flags, _RET_IP_)][index],
618 flags, node, size);
619 }
620 return __kmalloc_node(size, flags, node);
621 }
622
623 /**
624 * kmalloc_array - allocate memory for an array.
625 * @n: number of elements.
626 * @size: element size.
627 * @flags: the type of memory to allocate (see kmalloc).
628 */
kmalloc_array(size_t n,size_t size,gfp_t flags)629 static inline __alloc_size(1, 2) void *kmalloc_array(size_t n, size_t size, gfp_t flags)
630 {
631 size_t bytes;
632
633 if (unlikely(check_mul_overflow(n, size, &bytes)))
634 return NULL;
635 if (__builtin_constant_p(n) && __builtin_constant_p(size))
636 return kmalloc(bytes, flags);
637 return __kmalloc(bytes, flags);
638 }
639
640 /**
641 * krealloc_array - reallocate memory for an array.
642 * @p: pointer to the memory chunk to reallocate
643 * @new_n: new number of elements to alloc
644 * @new_size: new size of a single member of the array
645 * @flags: the type of memory to allocate (see kmalloc)
646 */
krealloc_array(void * p,size_t new_n,size_t new_size,gfp_t flags)647 static inline __realloc_size(2, 3) void * __must_check krealloc_array(void *p,
648 size_t new_n,
649 size_t new_size,
650 gfp_t flags)
651 {
652 size_t bytes;
653
654 if (unlikely(check_mul_overflow(new_n, new_size, &bytes)))
655 return NULL;
656
657 return krealloc(p, bytes, flags);
658 }
659
660 /**
661 * kcalloc - allocate memory for an array. The memory is set to zero.
662 * @n: number of elements.
663 * @size: element size.
664 * @flags: the type of memory to allocate (see kmalloc).
665 */
kcalloc(size_t n,size_t size,gfp_t flags)666 static inline __alloc_size(1, 2) void *kcalloc(size_t n, size_t size, gfp_t flags)
667 {
668 return kmalloc_array(n, size, flags | __GFP_ZERO);
669 }
670
671 void *__kmalloc_node_track_caller(size_t size, gfp_t flags, int node,
672 unsigned long caller) __alloc_size(1);
673 #define kmalloc_node_track_caller(size, flags, node) \
674 __kmalloc_node_track_caller(size, flags, node, \
675 _RET_IP_)
676
677 /*
678 * kmalloc_track_caller is a special version of kmalloc that records the
679 * calling function of the routine calling it for slab leak tracking instead
680 * of just the calling function (confusing, eh?).
681 * It's useful when the call to kmalloc comes from a widely-used standard
682 * allocator where we care about the real place the memory allocation
683 * request comes from.
684 */
685 #define kmalloc_track_caller(size, flags) \
686 __kmalloc_node_track_caller(size, flags, \
687 NUMA_NO_NODE, _RET_IP_)
688
kmalloc_array_node(size_t n,size_t size,gfp_t flags,int node)689 static inline __alloc_size(1, 2) void *kmalloc_array_node(size_t n, size_t size, gfp_t flags,
690 int node)
691 {
692 size_t bytes;
693
694 if (unlikely(check_mul_overflow(n, size, &bytes)))
695 return NULL;
696 if (__builtin_constant_p(n) && __builtin_constant_p(size))
697 return kmalloc_node(bytes, flags, node);
698 return __kmalloc_node(bytes, flags, node);
699 }
700
kcalloc_node(size_t n,size_t size,gfp_t flags,int node)701 static inline __alloc_size(1, 2) void *kcalloc_node(size_t n, size_t size, gfp_t flags, int node)
702 {
703 return kmalloc_array_node(n, size, flags | __GFP_ZERO, node);
704 }
705
706 /*
707 * Shortcuts
708 */
kmem_cache_zalloc(struct kmem_cache * k,gfp_t flags)709 static inline void *kmem_cache_zalloc(struct kmem_cache *k, gfp_t flags)
710 {
711 return kmem_cache_alloc(k, flags | __GFP_ZERO);
712 }
713
714 /**
715 * kzalloc - allocate memory. The memory is set to zero.
716 * @size: how many bytes of memory are required.
717 * @flags: the type of memory to allocate (see kmalloc).
718 */
kzalloc(size_t size,gfp_t flags)719 static inline __alloc_size(1) void *kzalloc(size_t size, gfp_t flags)
720 {
721 return kmalloc(size, flags | __GFP_ZERO);
722 }
723
724 /**
725 * kzalloc_node - allocate zeroed memory from a particular memory node.
726 * @size: how many bytes of memory are required.
727 * @flags: the type of memory to allocate (see kmalloc).
728 * @node: memory node from which to allocate
729 */
kzalloc_node(size_t size,gfp_t flags,int node)730 static inline __alloc_size(1) void *kzalloc_node(size_t size, gfp_t flags, int node)
731 {
732 return kmalloc_node(size, flags | __GFP_ZERO, node);
733 }
734
735 extern void *kvmalloc_node(size_t size, gfp_t flags, int node) __alloc_size(1);
kvmalloc(size_t size,gfp_t flags)736 static inline __alloc_size(1) void *kvmalloc(size_t size, gfp_t flags)
737 {
738 return kvmalloc_node(size, flags, NUMA_NO_NODE);
739 }
kvzalloc_node(size_t size,gfp_t flags,int node)740 static inline __alloc_size(1) void *kvzalloc_node(size_t size, gfp_t flags, int node)
741 {
742 return kvmalloc_node(size, flags | __GFP_ZERO, node);
743 }
kvzalloc(size_t size,gfp_t flags)744 static inline __alloc_size(1) void *kvzalloc(size_t size, gfp_t flags)
745 {
746 return kvmalloc(size, flags | __GFP_ZERO);
747 }
748
kvmalloc_array(size_t n,size_t size,gfp_t flags)749 static inline __alloc_size(1, 2) void *kvmalloc_array(size_t n, size_t size, gfp_t flags)
750 {
751 size_t bytes;
752
753 if (unlikely(check_mul_overflow(n, size, &bytes)))
754 return NULL;
755
756 return kvmalloc(bytes, flags);
757 }
758
kvcalloc(size_t n,size_t size,gfp_t flags)759 static inline __alloc_size(1, 2) void *kvcalloc(size_t n, size_t size, gfp_t flags)
760 {
761 return kvmalloc_array(n, size, flags | __GFP_ZERO);
762 }
763
764 extern void *kvrealloc(const void *p, size_t oldsize, size_t newsize, gfp_t flags)
765 __realloc_size(3);
766 extern void kvfree(const void *addr);
767 extern void kvfree_sensitive(const void *addr, size_t len);
768
769 unsigned int kmem_cache_size(struct kmem_cache *s);
770
771 /**
772 * kmalloc_size_roundup - Report allocation bucket size for the given size
773 *
774 * @size: Number of bytes to round up from.
775 *
776 * This returns the number of bytes that would be available in a kmalloc()
777 * allocation of @size bytes. For example, a 126 byte request would be
778 * rounded up to the next sized kmalloc bucket, 128 bytes. (This is strictly
779 * for the general-purpose kmalloc()-based allocations, and is not for the
780 * pre-sized kmem_cache_alloc()-based allocations.)
781 *
782 * Use this to kmalloc() the full bucket size ahead of time instead of using
783 * ksize() to query the size after an allocation.
784 */
785 size_t kmalloc_size_roundup(size_t size);
786
787 void __init kmem_cache_init_late(void);
788
789 #if defined(CONFIG_SMP) && defined(CONFIG_SLAB)
790 int slab_prepare_cpu(unsigned int cpu);
791 int slab_dead_cpu(unsigned int cpu);
792 #else
793 #define slab_prepare_cpu NULL
794 #define slab_dead_cpu NULL
795 #endif
796
797 #endif /* _LINUX_SLAB_H */
798