xref: /linux-6.15/include/linux/slab.h (revision 6edf2e37)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3  * Written by Mark Hemment, 1996 ([email protected]).
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/gfp.h>
16 #include <linux/overflow.h>
17 #include <linux/types.h>
18 #include <linux/workqueue.h>
19 
20 
21 /*
22  * Flags to pass to kmem_cache_create().
23  * The ones marked DEBUG are only valid if CONFIG_DEBUG_SLAB is set.
24  */
25 /* DEBUG: Perform (expensive) checks on alloc/free */
26 #define SLAB_CONSISTENCY_CHECKS	((slab_flags_t __force)0x00000100U)
27 /* DEBUG: Red zone objs in a cache */
28 #define SLAB_RED_ZONE		((slab_flags_t __force)0x00000400U)
29 /* DEBUG: Poison objects */
30 #define SLAB_POISON		((slab_flags_t __force)0x00000800U)
31 /* Align objs on cache lines */
32 #define SLAB_HWCACHE_ALIGN	((slab_flags_t __force)0x00002000U)
33 /* Use GFP_DMA memory */
34 #define SLAB_CACHE_DMA		((slab_flags_t __force)0x00004000U)
35 /* DEBUG: Store the last owner for bug hunting */
36 #define SLAB_STORE_USER		((slab_flags_t __force)0x00010000U)
37 /* Panic if kmem_cache_create() fails */
38 #define SLAB_PANIC		((slab_flags_t __force)0x00040000U)
39 /*
40  * SLAB_TYPESAFE_BY_RCU - **WARNING** READ THIS!
41  *
42  * This delays freeing the SLAB page by a grace period, it does _NOT_
43  * delay object freeing. This means that if you do kmem_cache_free()
44  * that memory location is free to be reused at any time. Thus it may
45  * be possible to see another object there in the same RCU grace period.
46  *
47  * This feature only ensures the memory location backing the object
48  * stays valid, the trick to using this is relying on an independent
49  * object validation pass. Something like:
50  *
51  *  rcu_read_lock()
52  * again:
53  *  obj = lockless_lookup(key);
54  *  if (obj) {
55  *    if (!try_get_ref(obj)) // might fail for free objects
56  *      goto again;
57  *
58  *    if (obj->key != key) { // not the object we expected
59  *      put_ref(obj);
60  *      goto again;
61  *    }
62  *  }
63  *  rcu_read_unlock();
64  *
65  * This is useful if we need to approach a kernel structure obliquely,
66  * from its address obtained without the usual locking. We can lock
67  * the structure to stabilize it and check it's still at the given address,
68  * only if we can be sure that the memory has not been meanwhile reused
69  * for some other kind of object (which our subsystem's lock might corrupt).
70  *
71  * rcu_read_lock before reading the address, then rcu_read_unlock after
72  * taking the spinlock within the structure expected at that address.
73  *
74  * Note that SLAB_TYPESAFE_BY_RCU was originally named SLAB_DESTROY_BY_RCU.
75  */
76 /* Defer freeing slabs to RCU */
77 #define SLAB_TYPESAFE_BY_RCU	((slab_flags_t __force)0x00080000U)
78 /* Spread some memory over cpuset */
79 #define SLAB_MEM_SPREAD		((slab_flags_t __force)0x00100000U)
80 /* Trace allocations and frees */
81 #define SLAB_TRACE		((slab_flags_t __force)0x00200000U)
82 
83 /* Flag to prevent checks on free */
84 #ifdef CONFIG_DEBUG_OBJECTS
85 # define SLAB_DEBUG_OBJECTS	((slab_flags_t __force)0x00400000U)
86 #else
87 # define SLAB_DEBUG_OBJECTS	0
88 #endif
89 
90 /* Avoid kmemleak tracing */
91 #define SLAB_NOLEAKTRACE	((slab_flags_t __force)0x00800000U)
92 
93 /* Fault injection mark */
94 #ifdef CONFIG_FAILSLAB
95 # define SLAB_FAILSLAB		((slab_flags_t __force)0x02000000U)
96 #else
97 # define SLAB_FAILSLAB		0
98 #endif
99 /* Account to memcg */
100 #ifdef CONFIG_MEMCG_KMEM
101 # define SLAB_ACCOUNT		((slab_flags_t __force)0x04000000U)
102 #else
103 # define SLAB_ACCOUNT		0
104 #endif
105 
106 #ifdef CONFIG_KASAN
107 #define SLAB_KASAN		((slab_flags_t __force)0x08000000U)
108 #else
109 #define SLAB_KASAN		0
110 #endif
111 
112 /* The following flags affect the page allocator grouping pages by mobility */
113 /* Objects are reclaimable */
114 #define SLAB_RECLAIM_ACCOUNT	((slab_flags_t __force)0x00020000U)
115 #define SLAB_TEMPORARY		SLAB_RECLAIM_ACCOUNT	/* Objects are short-lived */
116 /*
117  * ZERO_SIZE_PTR will be returned for zero sized kmalloc requests.
118  *
119  * Dereferencing ZERO_SIZE_PTR will lead to a distinct access fault.
120  *
121  * ZERO_SIZE_PTR can be passed to kfree though in the same way that NULL can.
122  * Both make kfree a no-op.
123  */
124 #define ZERO_SIZE_PTR ((void *)16)
125 
126 #define ZERO_OR_NULL_PTR(x) ((unsigned long)(x) <= \
127 				(unsigned long)ZERO_SIZE_PTR)
128 
129 #include <linux/kasan.h>
130 
131 struct mem_cgroup;
132 /*
133  * struct kmem_cache related prototypes
134  */
135 void __init kmem_cache_init(void);
136 bool slab_is_available(void);
137 
138 extern bool usercopy_fallback;
139 
140 struct kmem_cache *kmem_cache_create(const char *name, unsigned int size,
141 			unsigned int align, slab_flags_t flags,
142 			void (*ctor)(void *));
143 struct kmem_cache *kmem_cache_create_usercopy(const char *name,
144 			unsigned int size, unsigned int align,
145 			slab_flags_t flags,
146 			unsigned int useroffset, unsigned int usersize,
147 			void (*ctor)(void *));
148 void kmem_cache_destroy(struct kmem_cache *);
149 int kmem_cache_shrink(struct kmem_cache *);
150 
151 void memcg_create_kmem_cache(struct mem_cgroup *, struct kmem_cache *);
152 void memcg_deactivate_kmem_caches(struct mem_cgroup *);
153 void memcg_destroy_kmem_caches(struct mem_cgroup *);
154 
155 /*
156  * Please use this macro to create slab caches. Simply specify the
157  * name of the structure and maybe some flags that are listed above.
158  *
159  * The alignment of the struct determines object alignment. If you
160  * f.e. add ____cacheline_aligned_in_smp to the struct declaration
161  * then the objects will be properly aligned in SMP configurations.
162  */
163 #define KMEM_CACHE(__struct, __flags)					\
164 		kmem_cache_create(#__struct, sizeof(struct __struct),	\
165 			__alignof__(struct __struct), (__flags), NULL)
166 
167 /*
168  * To whitelist a single field for copying to/from usercopy, use this
169  * macro instead for KMEM_CACHE() above.
170  */
171 #define KMEM_CACHE_USERCOPY(__struct, __flags, __field)			\
172 		kmem_cache_create_usercopy(#__struct,			\
173 			sizeof(struct __struct),			\
174 			__alignof__(struct __struct), (__flags),	\
175 			offsetof(struct __struct, __field),		\
176 			sizeof_field(struct __struct, __field), NULL)
177 
178 /*
179  * Common kmalloc functions provided by all allocators
180  */
181 void * __must_check __krealloc(const void *, size_t, gfp_t);
182 void * __must_check krealloc(const void *, size_t, gfp_t);
183 void kfree(const void *);
184 void kzfree(const void *);
185 size_t ksize(const void *);
186 
187 #ifdef CONFIG_HAVE_HARDENED_USERCOPY_ALLOCATOR
188 void __check_heap_object(const void *ptr, unsigned long n, struct page *page,
189 			bool to_user);
190 #else
191 static inline void __check_heap_object(const void *ptr, unsigned long n,
192 				       struct page *page, bool to_user) { }
193 #endif
194 
195 /*
196  * Some archs want to perform DMA into kmalloc caches and need a guaranteed
197  * alignment larger than the alignment of a 64-bit integer.
198  * Setting ARCH_KMALLOC_MINALIGN in arch headers allows that.
199  */
200 #if defined(ARCH_DMA_MINALIGN) && ARCH_DMA_MINALIGN > 8
201 #define ARCH_KMALLOC_MINALIGN ARCH_DMA_MINALIGN
202 #define KMALLOC_MIN_SIZE ARCH_DMA_MINALIGN
203 #define KMALLOC_SHIFT_LOW ilog2(ARCH_DMA_MINALIGN)
204 #else
205 #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
206 #endif
207 
208 /*
209  * Setting ARCH_SLAB_MINALIGN in arch headers allows a different alignment.
210  * Intended for arches that get misalignment faults even for 64 bit integer
211  * aligned buffers.
212  */
213 #ifndef ARCH_SLAB_MINALIGN
214 #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
215 #endif
216 
217 /*
218  * kmalloc and friends return ARCH_KMALLOC_MINALIGN aligned
219  * pointers. kmem_cache_alloc and friends return ARCH_SLAB_MINALIGN
220  * aligned pointers.
221  */
222 #define __assume_kmalloc_alignment __assume_aligned(ARCH_KMALLOC_MINALIGN)
223 #define __assume_slab_alignment __assume_aligned(ARCH_SLAB_MINALIGN)
224 #define __assume_page_alignment __assume_aligned(PAGE_SIZE)
225 
226 /*
227  * Kmalloc array related definitions
228  */
229 
230 #ifdef CONFIG_SLAB
231 /*
232  * The largest kmalloc size supported by the SLAB allocators is
233  * 32 megabyte (2^25) or the maximum allocatable page order if that is
234  * less than 32 MB.
235  *
236  * WARNING: Its not easy to increase this value since the allocators have
237  * to do various tricks to work around compiler limitations in order to
238  * ensure proper constant folding.
239  */
240 #define KMALLOC_SHIFT_HIGH	((MAX_ORDER + PAGE_SHIFT - 1) <= 25 ? \
241 				(MAX_ORDER + PAGE_SHIFT - 1) : 25)
242 #define KMALLOC_SHIFT_MAX	KMALLOC_SHIFT_HIGH
243 #ifndef KMALLOC_SHIFT_LOW
244 #define KMALLOC_SHIFT_LOW	5
245 #endif
246 #endif
247 
248 #ifdef CONFIG_SLUB
249 /*
250  * SLUB directly allocates requests fitting in to an order-1 page
251  * (PAGE_SIZE*2).  Larger requests are passed to the page allocator.
252  */
253 #define KMALLOC_SHIFT_HIGH	(PAGE_SHIFT + 1)
254 #define KMALLOC_SHIFT_MAX	(MAX_ORDER + PAGE_SHIFT - 1)
255 #ifndef KMALLOC_SHIFT_LOW
256 #define KMALLOC_SHIFT_LOW	3
257 #endif
258 #endif
259 
260 #ifdef CONFIG_SLOB
261 /*
262  * SLOB passes all requests larger than one page to the page allocator.
263  * No kmalloc array is necessary since objects of different sizes can
264  * be allocated from the same page.
265  */
266 #define KMALLOC_SHIFT_HIGH	PAGE_SHIFT
267 #define KMALLOC_SHIFT_MAX	(MAX_ORDER + PAGE_SHIFT - 1)
268 #ifndef KMALLOC_SHIFT_LOW
269 #define KMALLOC_SHIFT_LOW	3
270 #endif
271 #endif
272 
273 /* Maximum allocatable size */
274 #define KMALLOC_MAX_SIZE	(1UL << KMALLOC_SHIFT_MAX)
275 /* Maximum size for which we actually use a slab cache */
276 #define KMALLOC_MAX_CACHE_SIZE	(1UL << KMALLOC_SHIFT_HIGH)
277 /* Maximum order allocatable via the slab allocagtor */
278 #define KMALLOC_MAX_ORDER	(KMALLOC_SHIFT_MAX - PAGE_SHIFT)
279 
280 /*
281  * Kmalloc subsystem.
282  */
283 #ifndef KMALLOC_MIN_SIZE
284 #define KMALLOC_MIN_SIZE (1 << KMALLOC_SHIFT_LOW)
285 #endif
286 
287 /*
288  * This restriction comes from byte sized index implementation.
289  * Page size is normally 2^12 bytes and, in this case, if we want to use
290  * byte sized index which can represent 2^8 entries, the size of the object
291  * should be equal or greater to 2^12 / 2^8 = 2^4 = 16.
292  * If minimum size of kmalloc is less than 16, we use it as minimum object
293  * size and give up to use byte sized index.
294  */
295 #define SLAB_OBJ_MIN_SIZE      (KMALLOC_MIN_SIZE < 16 ? \
296                                (KMALLOC_MIN_SIZE) : 16)
297 
298 /*
299  * Whenever changing this, take care of that kmalloc_type() and
300  * create_kmalloc_caches() still work as intended.
301  */
302 enum kmalloc_cache_type {
303 	KMALLOC_NORMAL = 0,
304 	KMALLOC_RECLAIM,
305 #ifdef CONFIG_ZONE_DMA
306 	KMALLOC_DMA,
307 #endif
308 	NR_KMALLOC_TYPES
309 };
310 
311 #ifndef CONFIG_SLOB
312 extern struct kmem_cache *
313 kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1];
314 
315 static __always_inline enum kmalloc_cache_type kmalloc_type(gfp_t flags)
316 {
317 	int is_dma = 0;
318 	int type_dma = 0;
319 	int is_reclaimable;
320 
321 #ifdef CONFIG_ZONE_DMA
322 	is_dma = !!(flags & __GFP_DMA);
323 	type_dma = is_dma * KMALLOC_DMA;
324 #endif
325 
326 	is_reclaimable = !!(flags & __GFP_RECLAIMABLE);
327 
328 	/*
329 	 * If an allocation is both __GFP_DMA and __GFP_RECLAIMABLE, return
330 	 * KMALLOC_DMA and effectively ignore __GFP_RECLAIMABLE
331 	 */
332 	return type_dma + (is_reclaimable & !is_dma) * KMALLOC_RECLAIM;
333 }
334 
335 /*
336  * Figure out which kmalloc slab an allocation of a certain size
337  * belongs to.
338  * 0 = zero alloc
339  * 1 =  65 .. 96 bytes
340  * 2 = 129 .. 192 bytes
341  * n = 2^(n-1)+1 .. 2^n
342  */
343 static __always_inline unsigned int kmalloc_index(size_t size)
344 {
345 	if (!size)
346 		return 0;
347 
348 	if (size <= KMALLOC_MIN_SIZE)
349 		return KMALLOC_SHIFT_LOW;
350 
351 	if (KMALLOC_MIN_SIZE <= 32 && size > 64 && size <= 96)
352 		return 1;
353 	if (KMALLOC_MIN_SIZE <= 64 && size > 128 && size <= 192)
354 		return 2;
355 	if (size <=          8) return 3;
356 	if (size <=         16) return 4;
357 	if (size <=         32) return 5;
358 	if (size <=         64) return 6;
359 	if (size <=        128) return 7;
360 	if (size <=        256) return 8;
361 	if (size <=        512) return 9;
362 	if (size <=       1024) return 10;
363 	if (size <=   2 * 1024) return 11;
364 	if (size <=   4 * 1024) return 12;
365 	if (size <=   8 * 1024) return 13;
366 	if (size <=  16 * 1024) return 14;
367 	if (size <=  32 * 1024) return 15;
368 	if (size <=  64 * 1024) return 16;
369 	if (size <= 128 * 1024) return 17;
370 	if (size <= 256 * 1024) return 18;
371 	if (size <= 512 * 1024) return 19;
372 	if (size <= 1024 * 1024) return 20;
373 	if (size <=  2 * 1024 * 1024) return 21;
374 	if (size <=  4 * 1024 * 1024) return 22;
375 	if (size <=  8 * 1024 * 1024) return 23;
376 	if (size <=  16 * 1024 * 1024) return 24;
377 	if (size <=  32 * 1024 * 1024) return 25;
378 	if (size <=  64 * 1024 * 1024) return 26;
379 	BUG();
380 
381 	/* Will never be reached. Needed because the compiler may complain */
382 	return -1;
383 }
384 #endif /* !CONFIG_SLOB */
385 
386 void *__kmalloc(size_t size, gfp_t flags) __assume_kmalloc_alignment __malloc;
387 void *kmem_cache_alloc(struct kmem_cache *, gfp_t flags) __assume_slab_alignment __malloc;
388 void kmem_cache_free(struct kmem_cache *, void *);
389 
390 /*
391  * Bulk allocation and freeing operations. These are accelerated in an
392  * allocator specific way to avoid taking locks repeatedly or building
393  * metadata structures unnecessarily.
394  *
395  * Note that interrupts must be enabled when calling these functions.
396  */
397 void kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
398 int kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
399 
400 /*
401  * Caller must not use kfree_bulk() on memory not originally allocated
402  * by kmalloc(), because the SLOB allocator cannot handle this.
403  */
404 static __always_inline void kfree_bulk(size_t size, void **p)
405 {
406 	kmem_cache_free_bulk(NULL, size, p);
407 }
408 
409 #ifdef CONFIG_NUMA
410 void *__kmalloc_node(size_t size, gfp_t flags, int node) __assume_kmalloc_alignment __malloc;
411 void *kmem_cache_alloc_node(struct kmem_cache *, gfp_t flags, int node) __assume_slab_alignment __malloc;
412 #else
413 static __always_inline void *__kmalloc_node(size_t size, gfp_t flags, int node)
414 {
415 	return __kmalloc(size, flags);
416 }
417 
418 static __always_inline void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t flags, int node)
419 {
420 	return kmem_cache_alloc(s, flags);
421 }
422 #endif
423 
424 #ifdef CONFIG_TRACING
425 extern void *kmem_cache_alloc_trace(struct kmem_cache *, gfp_t, size_t) __assume_slab_alignment __malloc;
426 
427 #ifdef CONFIG_NUMA
428 extern void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
429 					   gfp_t gfpflags,
430 					   int node, size_t size) __assume_slab_alignment __malloc;
431 #else
432 static __always_inline void *
433 kmem_cache_alloc_node_trace(struct kmem_cache *s,
434 			      gfp_t gfpflags,
435 			      int node, size_t size)
436 {
437 	return kmem_cache_alloc_trace(s, gfpflags, size);
438 }
439 #endif /* CONFIG_NUMA */
440 
441 #else /* CONFIG_TRACING */
442 static __always_inline void *kmem_cache_alloc_trace(struct kmem_cache *s,
443 		gfp_t flags, size_t size)
444 {
445 	void *ret = kmem_cache_alloc(s, flags);
446 
447 	kasan_kmalloc(s, ret, size, flags);
448 	return ret;
449 }
450 
451 static __always_inline void *
452 kmem_cache_alloc_node_trace(struct kmem_cache *s,
453 			      gfp_t gfpflags,
454 			      int node, size_t size)
455 {
456 	void *ret = kmem_cache_alloc_node(s, gfpflags, node);
457 
458 	kasan_kmalloc(s, ret, size, gfpflags);
459 	return ret;
460 }
461 #endif /* CONFIG_TRACING */
462 
463 extern void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) __assume_page_alignment __malloc;
464 
465 #ifdef CONFIG_TRACING
466 extern void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) __assume_page_alignment __malloc;
467 #else
468 static __always_inline void *
469 kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
470 {
471 	return kmalloc_order(size, flags, order);
472 }
473 #endif
474 
475 static __always_inline void *kmalloc_large(size_t size, gfp_t flags)
476 {
477 	unsigned int order = get_order(size);
478 	return kmalloc_order_trace(size, flags, order);
479 }
480 
481 /**
482  * kmalloc - allocate memory
483  * @size: how many bytes of memory are required.
484  * @flags: the type of memory to allocate.
485  *
486  * kmalloc is the normal method of allocating memory
487  * for objects smaller than page size in the kernel.
488  *
489  * The @flags argument may be one of:
490  *
491  * %GFP_USER - Allocate memory on behalf of user.  May sleep.
492  *
493  * %GFP_KERNEL - Allocate normal kernel ram.  May sleep.
494  *
495  * %GFP_ATOMIC - Allocation will not sleep.  May use emergency pools.
496  *   For example, use this inside interrupt handlers.
497  *
498  * %GFP_HIGHUSER - Allocate pages from high memory.
499  *
500  * %GFP_NOIO - Do not do any I/O at all while trying to get memory.
501  *
502  * %GFP_NOFS - Do not make any fs calls while trying to get memory.
503  *
504  * %GFP_NOWAIT - Allocation will not sleep.
505  *
506  * %__GFP_THISNODE - Allocate node-local memory only.
507  *
508  * %GFP_DMA - Allocation suitable for DMA.
509  *   Should only be used for kmalloc() caches. Otherwise, use a
510  *   slab created with SLAB_DMA.
511  *
512  * Also it is possible to set different flags by OR'ing
513  * in one or more of the following additional @flags:
514  *
515  * %__GFP_HIGH - This allocation has high priority and may use emergency pools.
516  *
517  * %__GFP_NOFAIL - Indicate that this allocation is in no way allowed to fail
518  *   (think twice before using).
519  *
520  * %__GFP_NORETRY - If memory is not immediately available,
521  *   then give up at once.
522  *
523  * %__GFP_NOWARN - If allocation fails, don't issue any warnings.
524  *
525  * %__GFP_RETRY_MAYFAIL - Try really hard to succeed the allocation but fail
526  *   eventually.
527  *
528  * There are other flags available as well, but these are not intended
529  * for general use, and so are not documented here. For a full list of
530  * potential flags, always refer to linux/gfp.h.
531  */
532 static __always_inline void *kmalloc(size_t size, gfp_t flags)
533 {
534 	if (__builtin_constant_p(size)) {
535 #ifndef CONFIG_SLOB
536 		unsigned int index;
537 #endif
538 		if (size > KMALLOC_MAX_CACHE_SIZE)
539 			return kmalloc_large(size, flags);
540 #ifndef CONFIG_SLOB
541 		index = kmalloc_index(size);
542 
543 		if (!index)
544 			return ZERO_SIZE_PTR;
545 
546 		return kmem_cache_alloc_trace(
547 				kmalloc_caches[kmalloc_type(flags)][index],
548 				flags, size);
549 #endif
550 	}
551 	return __kmalloc(size, flags);
552 }
553 
554 /*
555  * Determine size used for the nth kmalloc cache.
556  * return size or 0 if a kmalloc cache for that
557  * size does not exist
558  */
559 static __always_inline unsigned int kmalloc_size(unsigned int n)
560 {
561 #ifndef CONFIG_SLOB
562 	if (n > 2)
563 		return 1U << n;
564 
565 	if (n == 1 && KMALLOC_MIN_SIZE <= 32)
566 		return 96;
567 
568 	if (n == 2 && KMALLOC_MIN_SIZE <= 64)
569 		return 192;
570 #endif
571 	return 0;
572 }
573 
574 static __always_inline void *kmalloc_node(size_t size, gfp_t flags, int node)
575 {
576 #ifndef CONFIG_SLOB
577 	if (__builtin_constant_p(size) &&
578 		size <= KMALLOC_MAX_CACHE_SIZE) {
579 		unsigned int i = kmalloc_index(size);
580 
581 		if (!i)
582 			return ZERO_SIZE_PTR;
583 
584 		return kmem_cache_alloc_node_trace(
585 				kmalloc_caches[kmalloc_type(flags)][i],
586 						flags, node, size);
587 	}
588 #endif
589 	return __kmalloc_node(size, flags, node);
590 }
591 
592 struct memcg_cache_array {
593 	struct rcu_head rcu;
594 	struct kmem_cache *entries[0];
595 };
596 
597 /*
598  * This is the main placeholder for memcg-related information in kmem caches.
599  * Both the root cache and the child caches will have it. For the root cache,
600  * this will hold a dynamically allocated array large enough to hold
601  * information about the currently limited memcgs in the system. To allow the
602  * array to be accessed without taking any locks, on relocation we free the old
603  * version only after a grace period.
604  *
605  * Root and child caches hold different metadata.
606  *
607  * @root_cache:	Common to root and child caches.  NULL for root, pointer to
608  *		the root cache for children.
609  *
610  * The following fields are specific to root caches.
611  *
612  * @memcg_caches: kmemcg ID indexed table of child caches.  This table is
613  *		used to index child cachces during allocation and cleared
614  *		early during shutdown.
615  *
616  * @root_caches_node: List node for slab_root_caches list.
617  *
618  * @children:	List of all child caches.  While the child caches are also
619  *		reachable through @memcg_caches, a child cache remains on
620  *		this list until it is actually destroyed.
621  *
622  * The following fields are specific to child caches.
623  *
624  * @memcg:	Pointer to the memcg this cache belongs to.
625  *
626  * @children_node: List node for @root_cache->children list.
627  *
628  * @kmem_caches_node: List node for @memcg->kmem_caches list.
629  */
630 struct memcg_cache_params {
631 	struct kmem_cache *root_cache;
632 	union {
633 		struct {
634 			struct memcg_cache_array __rcu *memcg_caches;
635 			struct list_head __root_caches_node;
636 			struct list_head children;
637 			bool dying;
638 		};
639 		struct {
640 			struct mem_cgroup *memcg;
641 			struct list_head children_node;
642 			struct list_head kmem_caches_node;
643 
644 			void (*deact_fn)(struct kmem_cache *);
645 			union {
646 				struct rcu_head deact_rcu_head;
647 				struct work_struct deact_work;
648 			};
649 		};
650 	};
651 };
652 
653 int memcg_update_all_caches(int num_memcgs);
654 
655 /**
656  * kmalloc_array - allocate memory for an array.
657  * @n: number of elements.
658  * @size: element size.
659  * @flags: the type of memory to allocate (see kmalloc).
660  */
661 static inline void *kmalloc_array(size_t n, size_t size, gfp_t flags)
662 {
663 	size_t bytes;
664 
665 	if (unlikely(check_mul_overflow(n, size, &bytes)))
666 		return NULL;
667 	if (__builtin_constant_p(n) && __builtin_constant_p(size))
668 		return kmalloc(bytes, flags);
669 	return __kmalloc(bytes, flags);
670 }
671 
672 /**
673  * kcalloc - allocate memory for an array. The memory is set to zero.
674  * @n: number of elements.
675  * @size: element size.
676  * @flags: the type of memory to allocate (see kmalloc).
677  */
678 static inline void *kcalloc(size_t n, size_t size, gfp_t flags)
679 {
680 	return kmalloc_array(n, size, flags | __GFP_ZERO);
681 }
682 
683 /*
684  * kmalloc_track_caller is a special version of kmalloc that records the
685  * calling function of the routine calling it for slab leak tracking instead
686  * of just the calling function (confusing, eh?).
687  * It's useful when the call to kmalloc comes from a widely-used standard
688  * allocator where we care about the real place the memory allocation
689  * request comes from.
690  */
691 extern void *__kmalloc_track_caller(size_t, gfp_t, unsigned long);
692 #define kmalloc_track_caller(size, flags) \
693 	__kmalloc_track_caller(size, flags, _RET_IP_)
694 
695 static inline void *kmalloc_array_node(size_t n, size_t size, gfp_t flags,
696 				       int node)
697 {
698 	size_t bytes;
699 
700 	if (unlikely(check_mul_overflow(n, size, &bytes)))
701 		return NULL;
702 	if (__builtin_constant_p(n) && __builtin_constant_p(size))
703 		return kmalloc_node(bytes, flags, node);
704 	return __kmalloc_node(bytes, flags, node);
705 }
706 
707 static inline void *kcalloc_node(size_t n, size_t size, gfp_t flags, int node)
708 {
709 	return kmalloc_array_node(n, size, flags | __GFP_ZERO, node);
710 }
711 
712 
713 #ifdef CONFIG_NUMA
714 extern void *__kmalloc_node_track_caller(size_t, gfp_t, int, unsigned long);
715 #define kmalloc_node_track_caller(size, flags, node) \
716 	__kmalloc_node_track_caller(size, flags, node, \
717 			_RET_IP_)
718 
719 #else /* CONFIG_NUMA */
720 
721 #define kmalloc_node_track_caller(size, flags, node) \
722 	kmalloc_track_caller(size, flags)
723 
724 #endif /* CONFIG_NUMA */
725 
726 /*
727  * Shortcuts
728  */
729 static inline void *kmem_cache_zalloc(struct kmem_cache *k, gfp_t flags)
730 {
731 	return kmem_cache_alloc(k, flags | __GFP_ZERO);
732 }
733 
734 /**
735  * kzalloc - allocate memory. The memory is set to zero.
736  * @size: how many bytes of memory are required.
737  * @flags: the type of memory to allocate (see kmalloc).
738  */
739 static inline void *kzalloc(size_t size, gfp_t flags)
740 {
741 	return kmalloc(size, flags | __GFP_ZERO);
742 }
743 
744 /**
745  * kzalloc_node - allocate zeroed memory from a particular memory node.
746  * @size: how many bytes of memory are required.
747  * @flags: the type of memory to allocate (see kmalloc).
748  * @node: memory node from which to allocate
749  */
750 static inline void *kzalloc_node(size_t size, gfp_t flags, int node)
751 {
752 	return kmalloc_node(size, flags | __GFP_ZERO, node);
753 }
754 
755 unsigned int kmem_cache_size(struct kmem_cache *s);
756 void __init kmem_cache_init_late(void);
757 
758 #if defined(CONFIG_SMP) && defined(CONFIG_SLAB)
759 int slab_prepare_cpu(unsigned int cpu);
760 int slab_dead_cpu(unsigned int cpu);
761 #else
762 #define slab_prepare_cpu	NULL
763 #define slab_dead_cpu		NULL
764 #endif
765 
766 #endif	/* _LINUX_SLAB_H */
767