stddef.h: Introduce sizeof_field()
[linux-block.git] / include / linux / slab.h
CommitLineData
b2441318 1/* SPDX-License-Identifier: GPL-2.0 */
1da177e4 2/*
2e892f43
CL
3 * Written by Mark Hemment, 1996 (markhe@nextd.demon.co.uk).
4 *
cde53535 5 * (C) SGI 2006, Christoph Lameter
2e892f43
CL
6 * Cleaned up and restructured to ease the addition of alternative
7 * implementations of SLAB allocators.
f1b6eb6e
CL
8 * (C) Linux Foundation 2008-2013
9 * Unified interface for all slab allocators
1da177e4
LT
10 */
11
12#ifndef _LINUX_SLAB_H
13#define _LINUX_SLAB_H
14
1b1cec4b 15#include <linux/gfp.h>
1b1cec4b 16#include <linux/types.h>
1f458cbf
GC
17#include <linux/workqueue.h>
18
1da177e4 19
2e892f43
CL
20/*
21 * Flags to pass to kmem_cache_create().
124dee09 22 * The ones marked DEBUG are only valid if CONFIG_DEBUG_SLAB is set.
1da177e4 23 */
d50112ed 24/* DEBUG: Perform (expensive) checks on alloc/free */
4fd0b46e 25#define SLAB_CONSISTENCY_CHECKS ((slab_flags_t __force)0x00000100U)
d50112ed 26/* DEBUG: Red zone objs in a cache */
4fd0b46e 27#define SLAB_RED_ZONE ((slab_flags_t __force)0x00000400U)
d50112ed 28/* DEBUG: Poison objects */
4fd0b46e 29#define SLAB_POISON ((slab_flags_t __force)0x00000800U)
d50112ed 30/* Align objs on cache lines */
4fd0b46e 31#define SLAB_HWCACHE_ALIGN ((slab_flags_t __force)0x00002000U)
d50112ed 32/* Use GFP_DMA memory */
4fd0b46e 33#define SLAB_CACHE_DMA ((slab_flags_t __force)0x00004000U)
d50112ed 34/* DEBUG: Store the last owner for bug hunting */
4fd0b46e 35#define SLAB_STORE_USER ((slab_flags_t __force)0x00010000U)
d50112ed 36/* Panic if kmem_cache_create() fails */
4fd0b46e 37#define SLAB_PANIC ((slab_flags_t __force)0x00040000U)
d7de4c1d 38/*
5f0d5a3a 39 * SLAB_TYPESAFE_BY_RCU - **WARNING** READ THIS!
d7de4c1d
PZ
40 *
41 * This delays freeing the SLAB page by a grace period, it does _NOT_
42 * delay object freeing. This means that if you do kmem_cache_free()
43 * that memory location is free to be reused at any time. Thus it may
44 * be possible to see another object there in the same RCU grace period.
45 *
46 * This feature only ensures the memory location backing the object
47 * stays valid, the trick to using this is relying on an independent
48 * object validation pass. Something like:
49 *
50 * rcu_read_lock()
51 * again:
52 * obj = lockless_lookup(key);
53 * if (obj) {
54 * if (!try_get_ref(obj)) // might fail for free objects
55 * goto again;
56 *
57 * if (obj->key != key) { // not the object we expected
58 * put_ref(obj);
59 * goto again;
60 * }
61 * }
62 * rcu_read_unlock();
63 *
68126702
JK
64 * This is useful if we need to approach a kernel structure obliquely,
65 * from its address obtained without the usual locking. We can lock
66 * the structure to stabilize it and check it's still at the given address,
67 * only if we can be sure that the memory has not been meanwhile reused
68 * for some other kind of object (which our subsystem's lock might corrupt).
69 *
70 * rcu_read_lock before reading the address, then rcu_read_unlock after
71 * taking the spinlock within the structure expected at that address.
5f0d5a3a
PM
72 *
73 * Note that SLAB_TYPESAFE_BY_RCU was originally named SLAB_DESTROY_BY_RCU.
d7de4c1d 74 */
d50112ed 75/* Defer freeing slabs to RCU */
4fd0b46e 76#define SLAB_TYPESAFE_BY_RCU ((slab_flags_t __force)0x00080000U)
d50112ed 77/* Spread some memory over cpuset */
4fd0b46e 78#define SLAB_MEM_SPREAD ((slab_flags_t __force)0x00100000U)
d50112ed 79/* Trace allocations and frees */
4fd0b46e 80#define SLAB_TRACE ((slab_flags_t __force)0x00200000U)
1da177e4 81
30327acf
TG
82/* Flag to prevent checks on free */
83#ifdef CONFIG_DEBUG_OBJECTS
4fd0b46e 84# define SLAB_DEBUG_OBJECTS ((slab_flags_t __force)0x00400000U)
30327acf 85#else
4fd0b46e 86# define SLAB_DEBUG_OBJECTS 0
30327acf
TG
87#endif
88
d50112ed 89/* Avoid kmemleak tracing */
4fd0b46e 90#define SLAB_NOLEAKTRACE ((slab_flags_t __force)0x00800000U)
d5cff635 91
d50112ed 92/* Fault injection mark */
4c13dd3b 93#ifdef CONFIG_FAILSLAB
4fd0b46e 94# define SLAB_FAILSLAB ((slab_flags_t __force)0x02000000U)
4c13dd3b 95#else
4fd0b46e 96# define SLAB_FAILSLAB 0
4c13dd3b 97#endif
d50112ed 98/* Account to memcg */
127424c8 99#if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
4fd0b46e 100# define SLAB_ACCOUNT ((slab_flags_t __force)0x04000000U)
230e9fc2 101#else
4fd0b46e 102# define SLAB_ACCOUNT 0
230e9fc2 103#endif
2dff4405 104
7ed2f9e6 105#ifdef CONFIG_KASAN
4fd0b46e 106#define SLAB_KASAN ((slab_flags_t __force)0x08000000U)
7ed2f9e6 107#else
4fd0b46e 108#define SLAB_KASAN 0
7ed2f9e6
AP
109#endif
110
e12ba74d 111/* The following flags affect the page allocator grouping pages by mobility */
d50112ed 112/* Objects are reclaimable */
4fd0b46e 113#define SLAB_RECLAIM_ACCOUNT ((slab_flags_t __force)0x00020000U)
e12ba74d 114#define SLAB_TEMPORARY SLAB_RECLAIM_ACCOUNT /* Objects are short-lived */
6cb8f913
CL
115/*
116 * ZERO_SIZE_PTR will be returned for zero sized kmalloc requests.
117 *
118 * Dereferencing ZERO_SIZE_PTR will lead to a distinct access fault.
119 *
120 * ZERO_SIZE_PTR can be passed to kfree though in the same way that NULL can.
121 * Both make kfree a no-op.
122 */
123#define ZERO_SIZE_PTR ((void *)16)
124
1d4ec7b1 125#define ZERO_OR_NULL_PTR(x) ((unsigned long)(x) <= \
6cb8f913
CL
126 (unsigned long)ZERO_SIZE_PTR)
127
f1b6eb6e 128#include <linux/kmemleak.h>
0316bec2 129#include <linux/kasan.h>
3b0efdfa 130
2633d7a0 131struct mem_cgroup;
2e892f43
CL
132/*
133 * struct kmem_cache related prototypes
134 */
135void __init kmem_cache_init(void);
fda90124 136bool slab_is_available(void);
1da177e4 137
2e892f43 138struct kmem_cache *kmem_cache_create(const char *, size_t, size_t,
d50112ed 139 slab_flags_t,
51cc5068 140 void (*)(void *));
2e892f43
CL
141void kmem_cache_destroy(struct kmem_cache *);
142int kmem_cache_shrink(struct kmem_cache *);
2a4db7eb
VD
143
144void memcg_create_kmem_cache(struct mem_cgroup *, struct kmem_cache *);
145void memcg_deactivate_kmem_caches(struct mem_cgroup *);
146void memcg_destroy_kmem_caches(struct mem_cgroup *);
2e892f43 147
0a31bd5f
CL
148/*
149 * Please use this macro to create slab caches. Simply specify the
150 * name of the structure and maybe some flags that are listed above.
151 *
152 * The alignment of the struct determines object alignment. If you
153 * f.e. add ____cacheline_aligned_in_smp to the struct declaration
154 * then the objects will be properly aligned in SMP configurations.
155 */
156#define KMEM_CACHE(__struct, __flags) kmem_cache_create(#__struct,\
157 sizeof(struct __struct), __alignof__(struct __struct),\
20c2df83 158 (__flags), NULL)
0a31bd5f 159
34504667
CL
160/*
161 * Common kmalloc functions provided by all allocators
162 */
163void * __must_check __krealloc(const void *, size_t, gfp_t);
164void * __must_check krealloc(const void *, size_t, gfp_t);
165void kfree(const void *);
166void kzfree(const void *);
167size_t ksize(const void *);
168
f5509cc1 169#ifdef CONFIG_HAVE_HARDENED_USERCOPY_ALLOCATOR
f4e6e289
KC
170void __check_heap_object(const void *ptr, unsigned long n, struct page *page,
171 bool to_user);
f5509cc1 172#else
f4e6e289
KC
173static inline void __check_heap_object(const void *ptr, unsigned long n,
174 struct page *page, bool to_user) { }
f5509cc1
KC
175#endif
176
c601fd69
CL
177/*
178 * Some archs want to perform DMA into kmalloc caches and need a guaranteed
179 * alignment larger than the alignment of a 64-bit integer.
180 * Setting ARCH_KMALLOC_MINALIGN in arch headers allows that.
181 */
182#if defined(ARCH_DMA_MINALIGN) && ARCH_DMA_MINALIGN > 8
183#define ARCH_KMALLOC_MINALIGN ARCH_DMA_MINALIGN
184#define KMALLOC_MIN_SIZE ARCH_DMA_MINALIGN
185#define KMALLOC_SHIFT_LOW ilog2(ARCH_DMA_MINALIGN)
186#else
187#define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
188#endif
189
94a58c36
RV
190/*
191 * Setting ARCH_SLAB_MINALIGN in arch headers allows a different alignment.
192 * Intended for arches that get misalignment faults even for 64 bit integer
193 * aligned buffers.
194 */
195#ifndef ARCH_SLAB_MINALIGN
196#define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
197#endif
198
199/*
200 * kmalloc and friends return ARCH_KMALLOC_MINALIGN aligned
201 * pointers. kmem_cache_alloc and friends return ARCH_SLAB_MINALIGN
202 * aligned pointers.
203 */
204#define __assume_kmalloc_alignment __assume_aligned(ARCH_KMALLOC_MINALIGN)
205#define __assume_slab_alignment __assume_aligned(ARCH_SLAB_MINALIGN)
206#define __assume_page_alignment __assume_aligned(PAGE_SIZE)
207
0aa817f0 208/*
95a05b42
CL
209 * Kmalloc array related definitions
210 */
211
212#ifdef CONFIG_SLAB
213/*
214 * The largest kmalloc size supported by the SLAB allocators is
0aa817f0
CL
215 * 32 megabyte (2^25) or the maximum allocatable page order if that is
216 * less than 32 MB.
217 *
218 * WARNING: Its not easy to increase this value since the allocators have
219 * to do various tricks to work around compiler limitations in order to
220 * ensure proper constant folding.
221 */
debee076
CL
222#define KMALLOC_SHIFT_HIGH ((MAX_ORDER + PAGE_SHIFT - 1) <= 25 ? \
223 (MAX_ORDER + PAGE_SHIFT - 1) : 25)
95a05b42 224#define KMALLOC_SHIFT_MAX KMALLOC_SHIFT_HIGH
c601fd69 225#ifndef KMALLOC_SHIFT_LOW
95a05b42 226#define KMALLOC_SHIFT_LOW 5
c601fd69 227#endif
069e2b35
CL
228#endif
229
230#ifdef CONFIG_SLUB
95a05b42 231/*
433a91ff
DH
232 * SLUB directly allocates requests fitting in to an order-1 page
233 * (PAGE_SIZE*2). Larger requests are passed to the page allocator.
95a05b42
CL
234 */
235#define KMALLOC_SHIFT_HIGH (PAGE_SHIFT + 1)
bb1107f7 236#define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT - 1)
c601fd69 237#ifndef KMALLOC_SHIFT_LOW
95a05b42
CL
238#define KMALLOC_SHIFT_LOW 3
239#endif
c601fd69 240#endif
0aa817f0 241
069e2b35
CL
242#ifdef CONFIG_SLOB
243/*
433a91ff 244 * SLOB passes all requests larger than one page to the page allocator.
069e2b35
CL
245 * No kmalloc array is necessary since objects of different sizes can
246 * be allocated from the same page.
247 */
069e2b35 248#define KMALLOC_SHIFT_HIGH PAGE_SHIFT
bb1107f7 249#define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT - 1)
069e2b35
CL
250#ifndef KMALLOC_SHIFT_LOW
251#define KMALLOC_SHIFT_LOW 3
252#endif
253#endif
254
95a05b42
CL
255/* Maximum allocatable size */
256#define KMALLOC_MAX_SIZE (1UL << KMALLOC_SHIFT_MAX)
257/* Maximum size for which we actually use a slab cache */
258#define KMALLOC_MAX_CACHE_SIZE (1UL << KMALLOC_SHIFT_HIGH)
259/* Maximum order allocatable via the slab allocagtor */
260#define KMALLOC_MAX_ORDER (KMALLOC_SHIFT_MAX - PAGE_SHIFT)
0aa817f0 261
ce6a5026
CL
262/*
263 * Kmalloc subsystem.
264 */
c601fd69 265#ifndef KMALLOC_MIN_SIZE
95a05b42 266#define KMALLOC_MIN_SIZE (1 << KMALLOC_SHIFT_LOW)
ce6a5026
CL
267#endif
268
24f870d8
JK
269/*
270 * This restriction comes from byte sized index implementation.
271 * Page size is normally 2^12 bytes and, in this case, if we want to use
272 * byte sized index which can represent 2^8 entries, the size of the object
273 * should be equal or greater to 2^12 / 2^8 = 2^4 = 16.
274 * If minimum size of kmalloc is less than 16, we use it as minimum object
275 * size and give up to use byte sized index.
276 */
277#define SLAB_OBJ_MIN_SIZE (KMALLOC_MIN_SIZE < 16 ? \
278 (KMALLOC_MIN_SIZE) : 16)
279
069e2b35 280#ifndef CONFIG_SLOB
9425c58e
CL
281extern struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1];
282#ifdef CONFIG_ZONE_DMA
283extern struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1];
284#endif
285
ce6a5026
CL
286/*
287 * Figure out which kmalloc slab an allocation of a certain size
288 * belongs to.
289 * 0 = zero alloc
290 * 1 = 65 .. 96 bytes
1ed58b60
RV
291 * 2 = 129 .. 192 bytes
292 * n = 2^(n-1)+1 .. 2^n
ce6a5026
CL
293 */
294static __always_inline int kmalloc_index(size_t size)
295{
296 if (!size)
297 return 0;
298
299 if (size <= KMALLOC_MIN_SIZE)
300 return KMALLOC_SHIFT_LOW;
301
302 if (KMALLOC_MIN_SIZE <= 32 && size > 64 && size <= 96)
303 return 1;
304 if (KMALLOC_MIN_SIZE <= 64 && size > 128 && size <= 192)
305 return 2;
306 if (size <= 8) return 3;
307 if (size <= 16) return 4;
308 if (size <= 32) return 5;
309 if (size <= 64) return 6;
310 if (size <= 128) return 7;
311 if (size <= 256) return 8;
312 if (size <= 512) return 9;
313 if (size <= 1024) return 10;
314 if (size <= 2 * 1024) return 11;
315 if (size <= 4 * 1024) return 12;
316 if (size <= 8 * 1024) return 13;
317 if (size <= 16 * 1024) return 14;
318 if (size <= 32 * 1024) return 15;
319 if (size <= 64 * 1024) return 16;
320 if (size <= 128 * 1024) return 17;
321 if (size <= 256 * 1024) return 18;
322 if (size <= 512 * 1024) return 19;
323 if (size <= 1024 * 1024) return 20;
324 if (size <= 2 * 1024 * 1024) return 21;
325 if (size <= 4 * 1024 * 1024) return 22;
326 if (size <= 8 * 1024 * 1024) return 23;
327 if (size <= 16 * 1024 * 1024) return 24;
328 if (size <= 32 * 1024 * 1024) return 25;
329 if (size <= 64 * 1024 * 1024) return 26;
330 BUG();
331
332 /* Will never be reached. Needed because the compiler may complain */
333 return -1;
334}
069e2b35 335#endif /* !CONFIG_SLOB */
ce6a5026 336
48a27055
RV
337void *__kmalloc(size_t size, gfp_t flags) __assume_kmalloc_alignment __malloc;
338void *kmem_cache_alloc(struct kmem_cache *, gfp_t flags) __assume_slab_alignment __malloc;
2a4db7eb 339void kmem_cache_free(struct kmem_cache *, void *);
f1b6eb6e 340
484748f0 341/*
9f706d68 342 * Bulk allocation and freeing operations. These are accelerated in an
484748f0
CL
343 * allocator specific way to avoid taking locks repeatedly or building
344 * metadata structures unnecessarily.
345 *
346 * Note that interrupts must be enabled when calling these functions.
347 */
348void kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
865762a8 349int kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
484748f0 350
ca257195
JDB
351/*
352 * Caller must not use kfree_bulk() on memory not originally allocated
353 * by kmalloc(), because the SLOB allocator cannot handle this.
354 */
355static __always_inline void kfree_bulk(size_t size, void **p)
356{
357 kmem_cache_free_bulk(NULL, size, p);
358}
359
f1b6eb6e 360#ifdef CONFIG_NUMA
48a27055
RV
361void *__kmalloc_node(size_t size, gfp_t flags, int node) __assume_kmalloc_alignment __malloc;
362void *kmem_cache_alloc_node(struct kmem_cache *, gfp_t flags, int node) __assume_slab_alignment __malloc;
f1b6eb6e
CL
363#else
364static __always_inline void *__kmalloc_node(size_t size, gfp_t flags, int node)
365{
366 return __kmalloc(size, flags);
367}
368
369static __always_inline void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t flags, int node)
370{
371 return kmem_cache_alloc(s, flags);
372}
373#endif
374
375#ifdef CONFIG_TRACING
48a27055 376extern void *kmem_cache_alloc_trace(struct kmem_cache *, gfp_t, size_t) __assume_slab_alignment __malloc;
f1b6eb6e
CL
377
378#ifdef CONFIG_NUMA
379extern void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
380 gfp_t gfpflags,
48a27055 381 int node, size_t size) __assume_slab_alignment __malloc;
f1b6eb6e
CL
382#else
383static __always_inline void *
384kmem_cache_alloc_node_trace(struct kmem_cache *s,
385 gfp_t gfpflags,
386 int node, size_t size)
387{
388 return kmem_cache_alloc_trace(s, gfpflags, size);
389}
390#endif /* CONFIG_NUMA */
391
392#else /* CONFIG_TRACING */
393static __always_inline void *kmem_cache_alloc_trace(struct kmem_cache *s,
394 gfp_t flags, size_t size)
395{
0316bec2
AR
396 void *ret = kmem_cache_alloc(s, flags);
397
505f5dcb 398 kasan_kmalloc(s, ret, size, flags);
0316bec2 399 return ret;
f1b6eb6e
CL
400}
401
402static __always_inline void *
403kmem_cache_alloc_node_trace(struct kmem_cache *s,
404 gfp_t gfpflags,
405 int node, size_t size)
406{
0316bec2
AR
407 void *ret = kmem_cache_alloc_node(s, gfpflags, node);
408
505f5dcb 409 kasan_kmalloc(s, ret, size, gfpflags);
0316bec2 410 return ret;
f1b6eb6e
CL
411}
412#endif /* CONFIG_TRACING */
413
48a27055 414extern void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) __assume_page_alignment __malloc;
f1b6eb6e
CL
415
416#ifdef CONFIG_TRACING
48a27055 417extern void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) __assume_page_alignment __malloc;
f1b6eb6e
CL
418#else
419static __always_inline void *
420kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
421{
422 return kmalloc_order(size, flags, order);
423}
ce6a5026
CL
424#endif
425
f1b6eb6e
CL
426static __always_inline void *kmalloc_large(size_t size, gfp_t flags)
427{
428 unsigned int order = get_order(size);
429 return kmalloc_order_trace(size, flags, order);
430}
431
432/**
433 * kmalloc - allocate memory
434 * @size: how many bytes of memory are required.
7e3528c3 435 * @flags: the type of memory to allocate.
f1b6eb6e
CL
436 *
437 * kmalloc is the normal method of allocating memory
438 * for objects smaller than page size in the kernel.
7e3528c3
RD
439 *
440 * The @flags argument may be one of:
441 *
442 * %GFP_USER - Allocate memory on behalf of user. May sleep.
443 *
444 * %GFP_KERNEL - Allocate normal kernel ram. May sleep.
445 *
446 * %GFP_ATOMIC - Allocation will not sleep. May use emergency pools.
447 * For example, use this inside interrupt handlers.
448 *
449 * %GFP_HIGHUSER - Allocate pages from high memory.
450 *
451 * %GFP_NOIO - Do not do any I/O at all while trying to get memory.
452 *
453 * %GFP_NOFS - Do not make any fs calls while trying to get memory.
454 *
455 * %GFP_NOWAIT - Allocation will not sleep.
456 *
e97ca8e5 457 * %__GFP_THISNODE - Allocate node-local memory only.
7e3528c3
RD
458 *
459 * %GFP_DMA - Allocation suitable for DMA.
460 * Should only be used for kmalloc() caches. Otherwise, use a
461 * slab created with SLAB_DMA.
462 *
463 * Also it is possible to set different flags by OR'ing
464 * in one or more of the following additional @flags:
465 *
7e3528c3
RD
466 * %__GFP_HIGH - This allocation has high priority and may use emergency pools.
467 *
468 * %__GFP_NOFAIL - Indicate that this allocation is in no way allowed to fail
469 * (think twice before using).
470 *
471 * %__GFP_NORETRY - If memory is not immediately available,
472 * then give up at once.
473 *
474 * %__GFP_NOWARN - If allocation fails, don't issue any warnings.
475 *
dcda9b04
MH
476 * %__GFP_RETRY_MAYFAIL - Try really hard to succeed the allocation but fail
477 * eventually.
7e3528c3
RD
478 *
479 * There are other flags available as well, but these are not intended
480 * for general use, and so are not documented here. For a full list of
481 * potential flags, always refer to linux/gfp.h.
f1b6eb6e
CL
482 */
483static __always_inline void *kmalloc(size_t size, gfp_t flags)
484{
485 if (__builtin_constant_p(size)) {
486 if (size > KMALLOC_MAX_CACHE_SIZE)
487 return kmalloc_large(size, flags);
488#ifndef CONFIG_SLOB
489 if (!(flags & GFP_DMA)) {
490 int index = kmalloc_index(size);
491
492 if (!index)
493 return ZERO_SIZE_PTR;
494
495 return kmem_cache_alloc_trace(kmalloc_caches[index],
496 flags, size);
497 }
498#endif
499 }
500 return __kmalloc(size, flags);
501}
502
ce6a5026
CL
503/*
504 * Determine size used for the nth kmalloc cache.
505 * return size or 0 if a kmalloc cache for that
506 * size does not exist
507 */
508static __always_inline int kmalloc_size(int n)
509{
069e2b35 510#ifndef CONFIG_SLOB
ce6a5026
CL
511 if (n > 2)
512 return 1 << n;
513
514 if (n == 1 && KMALLOC_MIN_SIZE <= 32)
515 return 96;
516
517 if (n == 2 && KMALLOC_MIN_SIZE <= 64)
518 return 192;
069e2b35 519#endif
ce6a5026
CL
520 return 0;
521}
ce6a5026 522
f1b6eb6e
CL
523static __always_inline void *kmalloc_node(size_t size, gfp_t flags, int node)
524{
525#ifndef CONFIG_SLOB
526 if (__builtin_constant_p(size) &&
23774a2f 527 size <= KMALLOC_MAX_CACHE_SIZE && !(flags & GFP_DMA)) {
f1b6eb6e
CL
528 int i = kmalloc_index(size);
529
530 if (!i)
531 return ZERO_SIZE_PTR;
532
533 return kmem_cache_alloc_node_trace(kmalloc_caches[i],
534 flags, node, size);
535 }
536#endif
537 return __kmalloc_node(size, flags, node);
538}
539
f7ce3190
VD
540struct memcg_cache_array {
541 struct rcu_head rcu;
542 struct kmem_cache *entries[0];
543};
544
ba6c496e
GC
545/*
546 * This is the main placeholder for memcg-related information in kmem caches.
ba6c496e
GC
547 * Both the root cache and the child caches will have it. For the root cache,
548 * this will hold a dynamically allocated array large enough to hold
f8570263
VD
549 * information about the currently limited memcgs in the system. To allow the
550 * array to be accessed without taking any locks, on relocation we free the old
551 * version only after a grace period.
ba6c496e 552 *
9eeadc8b 553 * Root and child caches hold different metadata.
ba6c496e 554 *
9eeadc8b
TH
555 * @root_cache: Common to root and child caches. NULL for root, pointer to
556 * the root cache for children.
426589f5 557 *
9eeadc8b
TH
558 * The following fields are specific to root caches.
559 *
560 * @memcg_caches: kmemcg ID indexed table of child caches. This table is
561 * used to index child cachces during allocation and cleared
562 * early during shutdown.
563 *
510ded33
TH
564 * @root_caches_node: List node for slab_root_caches list.
565 *
9eeadc8b
TH
566 * @children: List of all child caches. While the child caches are also
567 * reachable through @memcg_caches, a child cache remains on
568 * this list until it is actually destroyed.
569 *
570 * The following fields are specific to child caches.
571 *
572 * @memcg: Pointer to the memcg this cache belongs to.
573 *
574 * @children_node: List node for @root_cache->children list.
bc2791f8
TH
575 *
576 * @kmem_caches_node: List node for @memcg->kmem_caches list.
ba6c496e
GC
577 */
578struct memcg_cache_params {
9eeadc8b 579 struct kmem_cache *root_cache;
ba6c496e 580 union {
9eeadc8b
TH
581 struct {
582 struct memcg_cache_array __rcu *memcg_caches;
510ded33 583 struct list_head __root_caches_node;
9eeadc8b
TH
584 struct list_head children;
585 };
2633d7a0
GC
586 struct {
587 struct mem_cgroup *memcg;
9eeadc8b 588 struct list_head children_node;
bc2791f8 589 struct list_head kmem_caches_node;
01fb58bc
TH
590
591 void (*deact_fn)(struct kmem_cache *);
592 union {
593 struct rcu_head deact_rcu_head;
594 struct work_struct deact_work;
595 };
2633d7a0 596 };
ba6c496e
GC
597 };
598};
599
2633d7a0
GC
600int memcg_update_all_caches(int num_memcgs);
601
e7efa615
MO
602/**
603 * kmalloc_array - allocate memory for an array.
604 * @n: number of elements.
605 * @size: element size.
606 * @flags: the type of memory to allocate (see kmalloc).
800590f5 607 */
a8203725 608static inline void *kmalloc_array(size_t n, size_t size, gfp_t flags)
1da177e4 609{
a3860c1c 610 if (size != 0 && n > SIZE_MAX / size)
6193a2ff 611 return NULL;
91c6a05f
AD
612 if (__builtin_constant_p(n) && __builtin_constant_p(size))
613 return kmalloc(n * size, flags);
a8203725
XW
614 return __kmalloc(n * size, flags);
615}
616
617/**
618 * kcalloc - allocate memory for an array. The memory is set to zero.
619 * @n: number of elements.
620 * @size: element size.
621 * @flags: the type of memory to allocate (see kmalloc).
622 */
623static inline void *kcalloc(size_t n, size_t size, gfp_t flags)
624{
625 return kmalloc_array(n, size, flags | __GFP_ZERO);
1da177e4
LT
626}
627
1d2c8eea
CH
628/*
629 * kmalloc_track_caller is a special version of kmalloc that records the
630 * calling function of the routine calling it for slab leak tracking instead
631 * of just the calling function (confusing, eh?).
632 * It's useful when the call to kmalloc comes from a widely-used standard
633 * allocator where we care about the real place the memory allocation
634 * request comes from.
635 */
ce71e27c 636extern void *__kmalloc_track_caller(size_t, gfp_t, unsigned long);
1d2c8eea 637#define kmalloc_track_caller(size, flags) \
ce71e27c 638 __kmalloc_track_caller(size, flags, _RET_IP_)
1da177e4 639
5799b255
JT
640static inline void *kmalloc_array_node(size_t n, size_t size, gfp_t flags,
641 int node)
642{
643 if (size != 0 && n > SIZE_MAX / size)
644 return NULL;
645 if (__builtin_constant_p(n) && __builtin_constant_p(size))
646 return kmalloc_node(n * size, flags, node);
647 return __kmalloc_node(n * size, flags, node);
648}
649
650static inline void *kcalloc_node(size_t n, size_t size, gfp_t flags, int node)
651{
652 return kmalloc_array_node(n, size, flags | __GFP_ZERO, node);
653}
654
655
97e2bde4 656#ifdef CONFIG_NUMA
ce71e27c 657extern void *__kmalloc_node_track_caller(size_t, gfp_t, int, unsigned long);
8b98c169
CH
658#define kmalloc_node_track_caller(size, flags, node) \
659 __kmalloc_node_track_caller(size, flags, node, \
ce71e27c 660 _RET_IP_)
2e892f43 661
8b98c169 662#else /* CONFIG_NUMA */
8b98c169
CH
663
664#define kmalloc_node_track_caller(size, flags, node) \
665 kmalloc_track_caller(size, flags)
97e2bde4 666
dfcd3610 667#endif /* CONFIG_NUMA */
10cef602 668
81cda662
CL
669/*
670 * Shortcuts
671 */
672static inline void *kmem_cache_zalloc(struct kmem_cache *k, gfp_t flags)
673{
674 return kmem_cache_alloc(k, flags | __GFP_ZERO);
675}
676
677/**
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).
681 */
682static inline void *kzalloc(size_t size, gfp_t flags)
683{
684 return kmalloc(size, flags | __GFP_ZERO);
685}
686
979b0fea
JL
687/**
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
692 */
693static inline void *kzalloc_node(size_t size, gfp_t flags, int node)
694{
695 return kmalloc_node(size, flags | __GFP_ZERO, node);
696}
697
07f361b2 698unsigned int kmem_cache_size(struct kmem_cache *s);
7e85ee0c
PE
699void __init kmem_cache_init_late(void);
700
6731d4f1
SAS
701#if defined(CONFIG_SMP) && defined(CONFIG_SLAB)
702int slab_prepare_cpu(unsigned int cpu);
703int slab_dead_cpu(unsigned int cpu);
704#else
705#define slab_prepare_cpu NULL
706#define slab_dead_cpu NULL
707#endif
708
1da177e4 709#endif /* _LINUX_SLAB_H */