blob: 04a7f7993e678d6454b6efaa608db3b88ea78eec [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Christoph Lameter2e892f42006-12-13 00:34:23 -08002 * Written by Mark Hemment, 1996 (markhe@nextd.demon.co.uk).
3 *
Christoph Lametercde53532008-07-04 09:59:22 -07004 * (C) SGI 2006, Christoph Lameter
Christoph Lameter2e892f42006-12-13 00:34:23 -08005 * Cleaned up and restructured to ease the addition of alternative
6 * implementations of SLAB allocators.
Christoph Lameterf1b6eb62013-09-04 16:35:34 +00007 * (C) Linux Foundation 2008-2013
8 * Unified interface for all slab allocators
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 */
10
11#ifndef _LINUX_SLAB_H
12#define _LINUX_SLAB_H
13
Andrew Morton1b1cec42006-12-06 20:33:22 -080014#include <linux/gfp.h>
Andrew Morton1b1cec42006-12-06 20:33:22 -080015#include <linux/types.h>
Glauber Costa1f458cb2012-12-18 14:22:50 -080016#include <linux/workqueue.h>
17
Linus Torvalds1da177e2005-04-16 15:20:36 -070018
Christoph Lameter2e892f42006-12-13 00:34:23 -080019/*
20 * Flags to pass to kmem_cache_create().
David Rientjes124dee02015-04-14 15:44:28 -070021 * The ones marked DEBUG are only valid if CONFIG_DEBUG_SLAB is set.
Linus Torvalds1da177e2005-04-16 15:20:36 -070022 */
Laura Abbottbecfda62016-03-15 14:55:06 -070023#define SLAB_CONSISTENCY_CHECKS 0x00000100UL /* DEBUG: Perform (expensive) checks on alloc/free */
Christoph Lameter55935a32006-12-13 00:34:24 -080024#define SLAB_RED_ZONE 0x00000400UL /* DEBUG: Red zone objs in a cache */
25#define SLAB_POISON 0x00000800UL /* DEBUG: Poison objects */
26#define SLAB_HWCACHE_ALIGN 0x00002000UL /* Align objs on cache lines */
Christoph Lameter2e892f42006-12-13 00:34:23 -080027#define SLAB_CACHE_DMA 0x00004000UL /* Use GFP_DMA memory */
Christoph Lameter2e892f42006-12-13 00:34:23 -080028#define SLAB_STORE_USER 0x00010000UL /* DEBUG: Store the last owner for bug hunting */
Christoph Lameter2e892f42006-12-13 00:34:23 -080029#define SLAB_PANIC 0x00040000UL /* Panic if kmem_cache_create() fails */
Peter Zijlstrad7de4c12008-11-13 20:40:12 +020030/*
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -080031 * SLAB_TYPESAFE_BY_RCU - **WARNING** READ THIS!
Peter Zijlstrad7de4c12008-11-13 20:40:12 +020032 *
33 * This delays freeing the SLAB page by a grace period, it does _NOT_
34 * delay object freeing. This means that if you do kmem_cache_free()
35 * that memory location is free to be reused at any time. Thus it may
36 * be possible to see another object there in the same RCU grace period.
37 *
38 * This feature only ensures the memory location backing the object
39 * stays valid, the trick to using this is relying on an independent
40 * object validation pass. Something like:
41 *
42 * rcu_read_lock()
43 * again:
44 * obj = lockless_lookup(key);
45 * if (obj) {
46 * if (!try_get_ref(obj)) // might fail for free objects
47 * goto again;
48 *
49 * if (obj->key != key) { // not the object we expected
50 * put_ref(obj);
51 * goto again;
52 * }
53 * }
54 * rcu_read_unlock();
55 *
Joonsoo Kim68126702013-10-24 10:07:42 +090056 * This is useful if we need to approach a kernel structure obliquely,
57 * from its address obtained without the usual locking. We can lock
58 * the structure to stabilize it and check it's still at the given address,
59 * only if we can be sure that the memory has not been meanwhile reused
60 * for some other kind of object (which our subsystem's lock might corrupt).
61 *
62 * rcu_read_lock before reading the address, then rcu_read_unlock after
63 * taking the spinlock within the structure expected at that address.
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -080064 *
65 * Note that SLAB_TYPESAFE_BY_RCU was originally named SLAB_DESTROY_BY_RCU.
Peter Zijlstrad7de4c12008-11-13 20:40:12 +020066 */
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -080067#define SLAB_TYPESAFE_BY_RCU 0x00080000UL /* Defer freeing slabs to RCU */
Paul Jackson101a5002006-03-24 03:16:07 -080068#define SLAB_MEM_SPREAD 0x00100000UL /* Spread some memory over cpuset */
Christoph Lameter81819f02007-05-06 14:49:36 -070069#define SLAB_TRACE 0x00200000UL /* Trace allocations and frees */
Linus Torvalds1da177e2005-04-16 15:20:36 -070070
Thomas Gleixner30327ac2008-04-30 00:54:59 -070071/* Flag to prevent checks on free */
72#ifdef CONFIG_DEBUG_OBJECTS
73# define SLAB_DEBUG_OBJECTS 0x00400000UL
74#else
75# define SLAB_DEBUG_OBJECTS 0x00000000UL
76#endif
77
Catalin Marinasd5cff632009-06-11 13:22:40 +010078#define SLAB_NOLEAKTRACE 0x00800000UL /* Avoid kmemleak tracing */
79
Vegard Nossum2dff4402008-05-31 15:56:17 +020080/* Don't track use of uninitialized memory */
81#ifdef CONFIG_KMEMCHECK
82# define SLAB_NOTRACK 0x01000000UL
83#else
84# define SLAB_NOTRACK 0x00000000UL
85#endif
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +030086#ifdef CONFIG_FAILSLAB
87# define SLAB_FAILSLAB 0x02000000UL /* Fault injection mark */
88#else
89# define SLAB_FAILSLAB 0x00000000UL
90#endif
Johannes Weiner127424c2016-01-20 15:02:32 -080091#if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
Vladimir Davydov230e9fc2016-01-14 15:18:15 -080092# define SLAB_ACCOUNT 0x04000000UL /* Account to memcg */
93#else
94# define SLAB_ACCOUNT 0x00000000UL
95#endif
Vegard Nossum2dff4402008-05-31 15:56:17 +020096
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -070097#ifdef CONFIG_KASAN
98#define SLAB_KASAN 0x08000000UL
99#else
100#define SLAB_KASAN 0x00000000UL
101#endif
102
Mel Gormane12ba742007-10-16 01:25:52 -0700103/* The following flags affect the page allocator grouping pages by mobility */
104#define SLAB_RECLAIM_ACCOUNT 0x00020000UL /* Objects are reclaimable */
105#define SLAB_TEMPORARY SLAB_RECLAIM_ACCOUNT /* Objects are short-lived */
Christoph Lameter2e892f42006-12-13 00:34:23 -0800106/*
Christoph Lameter6cb8f912007-07-17 04:03:22 -0700107 * ZERO_SIZE_PTR will be returned for zero sized kmalloc requests.
108 *
109 * Dereferencing ZERO_SIZE_PTR will lead to a distinct access fault.
110 *
111 * ZERO_SIZE_PTR can be passed to kfree though in the same way that NULL can.
112 * Both make kfree a no-op.
113 */
114#define ZERO_SIZE_PTR ((void *)16)
115
Roland Dreier1d4ec7b2007-07-20 12:13:20 -0700116#define ZERO_OR_NULL_PTR(x) ((unsigned long)(x) <= \
Christoph Lameter6cb8f912007-07-17 04:03:22 -0700117 (unsigned long)ZERO_SIZE_PTR)
118
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000119#include <linux/kmemleak.h>
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800120#include <linux/kasan.h>
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500121
Glauber Costa2633d7a2012-12-18 14:22:34 -0800122struct mem_cgroup;
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500123/*
Christoph Lameter2e892f42006-12-13 00:34:23 -0800124 * struct kmem_cache related prototypes
125 */
126void __init kmem_cache_init(void);
Denis Kirjanovfda90122015-11-05 18:44:59 -0800127bool slab_is_available(void);
Matt Mackall10cef602006-01-08 01:01:45 -0800128
Christoph Lameter2e892f42006-12-13 00:34:23 -0800129struct kmem_cache *kmem_cache_create(const char *, size_t, size_t,
Christoph Lameterebe29732006-12-06 20:32:59 -0800130 unsigned long,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700131 void (*)(void *));
Christoph Lameter2e892f42006-12-13 00:34:23 -0800132void kmem_cache_destroy(struct kmem_cache *);
133int kmem_cache_shrink(struct kmem_cache *);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -0800134
135void memcg_create_kmem_cache(struct mem_cgroup *, struct kmem_cache *);
136void memcg_deactivate_kmem_caches(struct mem_cgroup *);
137void memcg_destroy_kmem_caches(struct mem_cgroup *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138
Christoph Lameter0a31bd52007-05-06 14:49:57 -0700139/*
140 * Please use this macro to create slab caches. Simply specify the
141 * name of the structure and maybe some flags that are listed above.
142 *
143 * The alignment of the struct determines object alignment. If you
144 * f.e. add ____cacheline_aligned_in_smp to the struct declaration
145 * then the objects will be properly aligned in SMP configurations.
146 */
147#define KMEM_CACHE(__struct, __flags) kmem_cache_create(#__struct,\
148 sizeof(struct __struct), __alignof__(struct __struct),\
Paul Mundt20c2df82007-07-20 10:11:58 +0900149 (__flags), NULL)
Christoph Lameter0a31bd52007-05-06 14:49:57 -0700150
Christoph Lameter2e892f42006-12-13 00:34:23 -0800151/*
Christoph Lameter34504662013-01-10 19:00:53 +0000152 * Common kmalloc functions provided by all allocators
153 */
154void * __must_check __krealloc(const void *, size_t, gfp_t);
155void * __must_check krealloc(const void *, size_t, gfp_t);
156void kfree(const void *);
157void kzfree(const void *);
158size_t ksize(const void *);
159
Kees Cookf5509cc2016-06-07 11:05:33 -0700160#ifdef CONFIG_HAVE_HARDENED_USERCOPY_ALLOCATOR
161const char *__check_heap_object(const void *ptr, unsigned long n,
162 struct page *page);
163#else
164static inline const char *__check_heap_object(const void *ptr,
165 unsigned long n,
166 struct page *page)
167{
168 return NULL;
169}
170#endif
171
Christoph Lameterc601fd62013-02-05 16:36:47 +0000172/*
173 * Some archs want to perform DMA into kmalloc caches and need a guaranteed
174 * alignment larger than the alignment of a 64-bit integer.
175 * Setting ARCH_KMALLOC_MINALIGN in arch headers allows that.
176 */
177#if defined(ARCH_DMA_MINALIGN) && ARCH_DMA_MINALIGN > 8
178#define ARCH_KMALLOC_MINALIGN ARCH_DMA_MINALIGN
179#define KMALLOC_MIN_SIZE ARCH_DMA_MINALIGN
180#define KMALLOC_SHIFT_LOW ilog2(ARCH_DMA_MINALIGN)
181#else
182#define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
183#endif
184
Christoph Lameter34504662013-01-10 19:00:53 +0000185/*
Rasmus Villemoes94a58c32015-11-20 15:56:48 -0800186 * Setting ARCH_SLAB_MINALIGN in arch headers allows a different alignment.
187 * Intended for arches that get misalignment faults even for 64 bit integer
188 * aligned buffers.
189 */
190#ifndef ARCH_SLAB_MINALIGN
191#define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
192#endif
193
194/*
195 * kmalloc and friends return ARCH_KMALLOC_MINALIGN aligned
196 * pointers. kmem_cache_alloc and friends return ARCH_SLAB_MINALIGN
197 * aligned pointers.
198 */
199#define __assume_kmalloc_alignment __assume_aligned(ARCH_KMALLOC_MINALIGN)
200#define __assume_slab_alignment __assume_aligned(ARCH_SLAB_MINALIGN)
201#define __assume_page_alignment __assume_aligned(PAGE_SIZE)
202
203/*
Christoph Lameter95a05b42013-01-10 19:14:19 +0000204 * Kmalloc array related definitions
205 */
206
207#ifdef CONFIG_SLAB
208/*
209 * The largest kmalloc size supported by the SLAB allocators is
Christoph Lameter0aa817f2007-05-16 22:11:01 -0700210 * 32 megabyte (2^25) or the maximum allocatable page order if that is
211 * less than 32 MB.
212 *
213 * WARNING: Its not easy to increase this value since the allocators have
214 * to do various tricks to work around compiler limitations in order to
215 * ensure proper constant folding.
216 */
Christoph Lameterdebee072007-06-23 17:16:43 -0700217#define KMALLOC_SHIFT_HIGH ((MAX_ORDER + PAGE_SHIFT - 1) <= 25 ? \
218 (MAX_ORDER + PAGE_SHIFT - 1) : 25)
Christoph Lameter95a05b42013-01-10 19:14:19 +0000219#define KMALLOC_SHIFT_MAX KMALLOC_SHIFT_HIGH
Christoph Lameterc601fd62013-02-05 16:36:47 +0000220#ifndef KMALLOC_SHIFT_LOW
Christoph Lameter95a05b42013-01-10 19:14:19 +0000221#define KMALLOC_SHIFT_LOW 5
Christoph Lameterc601fd62013-02-05 16:36:47 +0000222#endif
Christoph Lameter069e2b352013-06-14 19:55:13 +0000223#endif
224
225#ifdef CONFIG_SLUB
Christoph Lameter95a05b42013-01-10 19:14:19 +0000226/*
Dave Hansen433a91f2014-01-28 14:24:50 -0800227 * SLUB directly allocates requests fitting in to an order-1 page
228 * (PAGE_SIZE*2). Larger requests are passed to the page allocator.
Christoph Lameter95a05b42013-01-10 19:14:19 +0000229 */
230#define KMALLOC_SHIFT_HIGH (PAGE_SHIFT + 1)
Michal Hockobb1107f2017-01-10 16:57:27 -0800231#define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT - 1)
Christoph Lameterc601fd62013-02-05 16:36:47 +0000232#ifndef KMALLOC_SHIFT_LOW
Christoph Lameter95a05b42013-01-10 19:14:19 +0000233#define KMALLOC_SHIFT_LOW 3
234#endif
Christoph Lameterc601fd62013-02-05 16:36:47 +0000235#endif
Christoph Lameter0aa817f2007-05-16 22:11:01 -0700236
Christoph Lameter069e2b352013-06-14 19:55:13 +0000237#ifdef CONFIG_SLOB
238/*
Dave Hansen433a91f2014-01-28 14:24:50 -0800239 * SLOB passes all requests larger than one page to the page allocator.
Christoph Lameter069e2b352013-06-14 19:55:13 +0000240 * No kmalloc array is necessary since objects of different sizes can
241 * be allocated from the same page.
242 */
Christoph Lameter069e2b352013-06-14 19:55:13 +0000243#define KMALLOC_SHIFT_HIGH PAGE_SHIFT
Michal Hockobb1107f2017-01-10 16:57:27 -0800244#define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT - 1)
Christoph Lameter069e2b352013-06-14 19:55:13 +0000245#ifndef KMALLOC_SHIFT_LOW
246#define KMALLOC_SHIFT_LOW 3
247#endif
248#endif
249
Christoph Lameter95a05b42013-01-10 19:14:19 +0000250/* Maximum allocatable size */
251#define KMALLOC_MAX_SIZE (1UL << KMALLOC_SHIFT_MAX)
252/* Maximum size for which we actually use a slab cache */
253#define KMALLOC_MAX_CACHE_SIZE (1UL << KMALLOC_SHIFT_HIGH)
254/* Maximum order allocatable via the slab allocagtor */
255#define KMALLOC_MAX_ORDER (KMALLOC_SHIFT_MAX - PAGE_SHIFT)
Christoph Lameter0aa817f2007-05-16 22:11:01 -0700256
Christoph Lameter90810642011-06-23 09:36:12 -0500257/*
Christoph Lameterce6a5022013-01-10 19:14:19 +0000258 * Kmalloc subsystem.
259 */
Christoph Lameterc601fd62013-02-05 16:36:47 +0000260#ifndef KMALLOC_MIN_SIZE
Christoph Lameter95a05b42013-01-10 19:14:19 +0000261#define KMALLOC_MIN_SIZE (1 << KMALLOC_SHIFT_LOW)
Christoph Lameterce6a5022013-01-10 19:14:19 +0000262#endif
Christoph Lameterce6a5022013-01-10 19:14:19 +0000263
Joonsoo Kim24f870d2014-03-12 17:06:19 +0900264/*
265 * This restriction comes from byte sized index implementation.
266 * Page size is normally 2^12 bytes and, in this case, if we want to use
267 * byte sized index which can represent 2^8 entries, the size of the object
268 * should be equal or greater to 2^12 / 2^8 = 2^4 = 16.
269 * If minimum size of kmalloc is less than 16, we use it as minimum object
270 * size and give up to use byte sized index.
271 */
272#define SLAB_OBJ_MIN_SIZE (KMALLOC_MIN_SIZE < 16 ? \
273 (KMALLOC_MIN_SIZE) : 16)
274
Christoph Lameter069e2b352013-06-14 19:55:13 +0000275#ifndef CONFIG_SLOB
Christoph Lameter9425c582013-01-10 19:12:17 +0000276extern struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1];
277#ifdef CONFIG_ZONE_DMA
278extern struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1];
279#endif
280
Christoph Lameterce6a5022013-01-10 19:14:19 +0000281/*
282 * Figure out which kmalloc slab an allocation of a certain size
283 * belongs to.
284 * 0 = zero alloc
285 * 1 = 65 .. 96 bytes
Rasmus Villemoes1ed58b62015-06-24 16:55:59 -0700286 * 2 = 129 .. 192 bytes
287 * n = 2^(n-1)+1 .. 2^n
Christoph Lameterce6a5022013-01-10 19:14:19 +0000288 */
289static __always_inline int kmalloc_index(size_t size)
290{
291 if (!size)
292 return 0;
293
294 if (size <= KMALLOC_MIN_SIZE)
295 return KMALLOC_SHIFT_LOW;
296
297 if (KMALLOC_MIN_SIZE <= 32 && size > 64 && size <= 96)
298 return 1;
299 if (KMALLOC_MIN_SIZE <= 64 && size > 128 && size <= 192)
300 return 2;
301 if (size <= 8) return 3;
302 if (size <= 16) return 4;
303 if (size <= 32) return 5;
304 if (size <= 64) return 6;
305 if (size <= 128) return 7;
306 if (size <= 256) return 8;
307 if (size <= 512) return 9;
308 if (size <= 1024) return 10;
309 if (size <= 2 * 1024) return 11;
310 if (size <= 4 * 1024) return 12;
311 if (size <= 8 * 1024) return 13;
312 if (size <= 16 * 1024) return 14;
313 if (size <= 32 * 1024) return 15;
314 if (size <= 64 * 1024) return 16;
315 if (size <= 128 * 1024) return 17;
316 if (size <= 256 * 1024) return 18;
317 if (size <= 512 * 1024) return 19;
318 if (size <= 1024 * 1024) return 20;
319 if (size <= 2 * 1024 * 1024) return 21;
320 if (size <= 4 * 1024 * 1024) return 22;
321 if (size <= 8 * 1024 * 1024) return 23;
322 if (size <= 16 * 1024 * 1024) return 24;
323 if (size <= 32 * 1024 * 1024) return 25;
324 if (size <= 64 * 1024 * 1024) return 26;
325 BUG();
326
327 /* Will never be reached. Needed because the compiler may complain */
328 return -1;
329}
Christoph Lameter069e2b352013-06-14 19:55:13 +0000330#endif /* !CONFIG_SLOB */
Christoph Lameterce6a5022013-01-10 19:14:19 +0000331
Rasmus Villemoes48a270552016-05-19 17:10:55 -0700332void *__kmalloc(size_t size, gfp_t flags) __assume_kmalloc_alignment __malloc;
333void *kmem_cache_alloc(struct kmem_cache *, gfp_t flags) __assume_slab_alignment __malloc;
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -0800334void kmem_cache_free(struct kmem_cache *, void *);
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000335
Christoph Lameter484748f2015-09-04 15:45:34 -0700336/*
Jesper Dangaard Brouer9f706d62016-03-15 14:54:03 -0700337 * Bulk allocation and freeing operations. These are accelerated in an
Christoph Lameter484748f2015-09-04 15:45:34 -0700338 * allocator specific way to avoid taking locks repeatedly or building
339 * metadata structures unnecessarily.
340 *
341 * Note that interrupts must be enabled when calling these functions.
342 */
343void kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
Jesper Dangaard Brouer865762a2015-11-20 15:57:58 -0800344int kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
Christoph Lameter484748f2015-09-04 15:45:34 -0700345
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -0700346/*
347 * Caller must not use kfree_bulk() on memory not originally allocated
348 * by kmalloc(), because the SLOB allocator cannot handle this.
349 */
350static __always_inline void kfree_bulk(size_t size, void **p)
351{
352 kmem_cache_free_bulk(NULL, size, p);
353}
354
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000355#ifdef CONFIG_NUMA
Rasmus Villemoes48a270552016-05-19 17:10:55 -0700356void *__kmalloc_node(size_t size, gfp_t flags, int node) __assume_kmalloc_alignment __malloc;
357void *kmem_cache_alloc_node(struct kmem_cache *, gfp_t flags, int node) __assume_slab_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000358#else
359static __always_inline void *__kmalloc_node(size_t size, gfp_t flags, int node)
360{
361 return __kmalloc(size, flags);
362}
363
364static __always_inline void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t flags, int node)
365{
366 return kmem_cache_alloc(s, flags);
367}
368#endif
369
370#ifdef CONFIG_TRACING
Rasmus Villemoes48a270552016-05-19 17:10:55 -0700371extern void *kmem_cache_alloc_trace(struct kmem_cache *, gfp_t, size_t) __assume_slab_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000372
373#ifdef CONFIG_NUMA
374extern void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
375 gfp_t gfpflags,
Rasmus Villemoes48a270552016-05-19 17:10:55 -0700376 int node, size_t size) __assume_slab_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000377#else
378static __always_inline void *
379kmem_cache_alloc_node_trace(struct kmem_cache *s,
380 gfp_t gfpflags,
381 int node, size_t size)
382{
383 return kmem_cache_alloc_trace(s, gfpflags, size);
384}
385#endif /* CONFIG_NUMA */
386
387#else /* CONFIG_TRACING */
388static __always_inline void *kmem_cache_alloc_trace(struct kmem_cache *s,
389 gfp_t flags, size_t size)
390{
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800391 void *ret = kmem_cache_alloc(s, flags);
392
Alexander Potapenko505f5dc2016-03-25 14:22:02 -0700393 kasan_kmalloc(s, ret, size, flags);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800394 return ret;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000395}
396
397static __always_inline void *
398kmem_cache_alloc_node_trace(struct kmem_cache *s,
399 gfp_t gfpflags,
400 int node, size_t size)
401{
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800402 void *ret = kmem_cache_alloc_node(s, gfpflags, node);
403
Alexander Potapenko505f5dc2016-03-25 14:22:02 -0700404 kasan_kmalloc(s, ret, size, gfpflags);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800405 return ret;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000406}
407#endif /* CONFIG_TRACING */
408
Rasmus Villemoes48a270552016-05-19 17:10:55 -0700409extern void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) __assume_page_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000410
411#ifdef CONFIG_TRACING
Rasmus Villemoes48a270552016-05-19 17:10:55 -0700412extern void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) __assume_page_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000413#else
414static __always_inline void *
415kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
416{
417 return kmalloc_order(size, flags, order);
418}
Christoph Lameterce6a5022013-01-10 19:14:19 +0000419#endif
420
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000421static __always_inline void *kmalloc_large(size_t size, gfp_t flags)
422{
423 unsigned int order = get_order(size);
424 return kmalloc_order_trace(size, flags, order);
425}
426
427/**
428 * kmalloc - allocate memory
429 * @size: how many bytes of memory are required.
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800430 * @flags: the type of memory to allocate.
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000431 *
432 * kmalloc is the normal method of allocating memory
433 * for objects smaller than page size in the kernel.
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800434 *
435 * The @flags argument may be one of:
436 *
437 * %GFP_USER - Allocate memory on behalf of user. May sleep.
438 *
439 * %GFP_KERNEL - Allocate normal kernel ram. May sleep.
440 *
441 * %GFP_ATOMIC - Allocation will not sleep. May use emergency pools.
442 * For example, use this inside interrupt handlers.
443 *
444 * %GFP_HIGHUSER - Allocate pages from high memory.
445 *
446 * %GFP_NOIO - Do not do any I/O at all while trying to get memory.
447 *
448 * %GFP_NOFS - Do not make any fs calls while trying to get memory.
449 *
450 * %GFP_NOWAIT - Allocation will not sleep.
451 *
Johannes Weinere97ca8e52014-03-10 15:49:43 -0700452 * %__GFP_THISNODE - Allocate node-local memory only.
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800453 *
454 * %GFP_DMA - Allocation suitable for DMA.
455 * Should only be used for kmalloc() caches. Otherwise, use a
456 * slab created with SLAB_DMA.
457 *
458 * Also it is possible to set different flags by OR'ing
459 * in one or more of the following additional @flags:
460 *
461 * %__GFP_COLD - Request cache-cold pages instead of
462 * trying to return cache-warm pages.
463 *
464 * %__GFP_HIGH - This allocation has high priority and may use emergency pools.
465 *
466 * %__GFP_NOFAIL - Indicate that this allocation is in no way allowed to fail
467 * (think twice before using).
468 *
469 * %__GFP_NORETRY - If memory is not immediately available,
470 * then give up at once.
471 *
472 * %__GFP_NOWARN - If allocation fails, don't issue any warnings.
473 *
474 * %__GFP_REPEAT - If allocation fails initially, try once more before failing.
475 *
476 * There are other flags available as well, but these are not intended
477 * for general use, and so are not documented here. For a full list of
478 * potential flags, always refer to linux/gfp.h.
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000479 */
480static __always_inline void *kmalloc(size_t size, gfp_t flags)
481{
482 if (__builtin_constant_p(size)) {
483 if (size > KMALLOC_MAX_CACHE_SIZE)
484 return kmalloc_large(size, flags);
485#ifndef CONFIG_SLOB
486 if (!(flags & GFP_DMA)) {
487 int index = kmalloc_index(size);
488
489 if (!index)
490 return ZERO_SIZE_PTR;
491
492 return kmem_cache_alloc_trace(kmalloc_caches[index],
493 flags, size);
494 }
495#endif
496 }
497 return __kmalloc(size, flags);
498}
499
Christoph Lameterce6a5022013-01-10 19:14:19 +0000500/*
501 * Determine size used for the nth kmalloc cache.
502 * return size or 0 if a kmalloc cache for that
503 * size does not exist
504 */
505static __always_inline int kmalloc_size(int n)
506{
Christoph Lameter069e2b352013-06-14 19:55:13 +0000507#ifndef CONFIG_SLOB
Christoph Lameterce6a5022013-01-10 19:14:19 +0000508 if (n > 2)
509 return 1 << n;
510
511 if (n == 1 && KMALLOC_MIN_SIZE <= 32)
512 return 96;
513
514 if (n == 2 && KMALLOC_MIN_SIZE <= 64)
515 return 192;
Christoph Lameter069e2b352013-06-14 19:55:13 +0000516#endif
Christoph Lameterce6a5022013-01-10 19:14:19 +0000517 return 0;
518}
Christoph Lameterce6a5022013-01-10 19:14:19 +0000519
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000520static __always_inline void *kmalloc_node(size_t size, gfp_t flags, int node)
521{
522#ifndef CONFIG_SLOB
523 if (__builtin_constant_p(size) &&
Christoph Lameter23774a22013-09-04 19:58:08 +0000524 size <= KMALLOC_MAX_CACHE_SIZE && !(flags & GFP_DMA)) {
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000525 int i = kmalloc_index(size);
526
527 if (!i)
528 return ZERO_SIZE_PTR;
529
530 return kmem_cache_alloc_node_trace(kmalloc_caches[i],
531 flags, node, size);
532 }
533#endif
534 return __kmalloc_node(size, flags, node);
535}
536
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800537struct memcg_cache_array {
538 struct rcu_head rcu;
539 struct kmem_cache *entries[0];
540};
541
Christoph Lameter0aa817f2007-05-16 22:11:01 -0700542/*
Glauber Costaba6c4962012-12-18 14:22:27 -0800543 * This is the main placeholder for memcg-related information in kmem caches.
Glauber Costaba6c4962012-12-18 14:22:27 -0800544 * Both the root cache and the child caches will have it. For the root cache,
545 * this will hold a dynamically allocated array large enough to hold
Vladimir Davydovf8570262014-01-23 15:53:06 -0800546 * information about the currently limited memcgs in the system. To allow the
547 * array to be accessed without taking any locks, on relocation we free the old
548 * version only after a grace period.
Glauber Costaba6c4962012-12-18 14:22:27 -0800549 *
Tejun Heo9eeadc82017-02-22 15:41:17 -0800550 * Root and child caches hold different metadata.
Glauber Costaba6c4962012-12-18 14:22:27 -0800551 *
Tejun Heo9eeadc82017-02-22 15:41:17 -0800552 * @root_cache: Common to root and child caches. NULL for root, pointer to
553 * the root cache for children.
Vladimir Davydov426589f2015-02-12 14:59:23 -0800554 *
Tejun Heo9eeadc82017-02-22 15:41:17 -0800555 * The following fields are specific to root caches.
556 *
557 * @memcg_caches: kmemcg ID indexed table of child caches. This table is
558 * used to index child cachces during allocation and cleared
559 * early during shutdown.
560 *
Tejun Heo510ded32017-02-22 15:41:24 -0800561 * @root_caches_node: List node for slab_root_caches list.
562 *
Tejun Heo9eeadc82017-02-22 15:41:17 -0800563 * @children: List of all child caches. While the child caches are also
564 * reachable through @memcg_caches, a child cache remains on
565 * this list until it is actually destroyed.
566 *
567 * The following fields are specific to child caches.
568 *
569 * @memcg: Pointer to the memcg this cache belongs to.
570 *
571 * @children_node: List node for @root_cache->children list.
Tejun Heobc2791f2017-02-22 15:41:21 -0800572 *
573 * @kmem_caches_node: List node for @memcg->kmem_caches list.
Glauber Costaba6c4962012-12-18 14:22:27 -0800574 */
575struct memcg_cache_params {
Tejun Heo9eeadc82017-02-22 15:41:17 -0800576 struct kmem_cache *root_cache;
Glauber Costaba6c4962012-12-18 14:22:27 -0800577 union {
Tejun Heo9eeadc82017-02-22 15:41:17 -0800578 struct {
579 struct memcg_cache_array __rcu *memcg_caches;
Tejun Heo510ded32017-02-22 15:41:24 -0800580 struct list_head __root_caches_node;
Tejun Heo9eeadc82017-02-22 15:41:17 -0800581 struct list_head children;
582 };
Glauber Costa2633d7a2012-12-18 14:22:34 -0800583 struct {
584 struct mem_cgroup *memcg;
Tejun Heo9eeadc82017-02-22 15:41:17 -0800585 struct list_head children_node;
Tejun Heobc2791f2017-02-22 15:41:21 -0800586 struct list_head kmem_caches_node;
Tejun Heo01fb58b2017-02-22 15:41:30 -0800587
588 void (*deact_fn)(struct kmem_cache *);
589 union {
590 struct rcu_head deact_rcu_head;
591 struct work_struct deact_work;
592 };
Glauber Costa2633d7a2012-12-18 14:22:34 -0800593 };
Glauber Costaba6c4962012-12-18 14:22:27 -0800594 };
595};
596
Glauber Costa2633d7a2012-12-18 14:22:34 -0800597int memcg_update_all_caches(int num_memcgs);
598
Christoph Lameter2e892f42006-12-13 00:34:23 -0800599/**
Michael Opdenackere7efa612013-06-25 18:16:55 +0200600 * kmalloc_array - allocate memory for an array.
601 * @n: number of elements.
602 * @size: element size.
603 * @flags: the type of memory to allocate (see kmalloc).
Paul Drynoff800590f2006-06-23 02:03:48 -0700604 */
Xi Wanga8203722012-03-05 15:14:41 -0800605static inline void *kmalloc_array(size_t n, size_t size, gfp_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606{
Xi Wanga3860c12012-05-31 16:26:04 -0700607 if (size != 0 && n > SIZE_MAX / size)
Paul Mundt6193a2f2007-07-15 23:38:22 -0700608 return NULL;
Alexey Dobriyan91c6a052016-07-26 15:22:08 -0700609 if (__builtin_constant_p(n) && __builtin_constant_p(size))
610 return kmalloc(n * size, flags);
Xi Wanga8203722012-03-05 15:14:41 -0800611 return __kmalloc(n * size, flags);
612}
613
614/**
615 * kcalloc - allocate memory for an array. The memory is set to zero.
616 * @n: number of elements.
617 * @size: element size.
618 * @flags: the type of memory to allocate (see kmalloc).
619 */
620static inline void *kcalloc(size_t n, size_t size, gfp_t flags)
621{
622 return kmalloc_array(n, size, flags | __GFP_ZERO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623}
624
Christoph Hellwig1d2c8ee2006-10-04 02:15:25 -0700625/*
626 * kmalloc_track_caller is a special version of kmalloc that records the
627 * calling function of the routine calling it for slab leak tracking instead
628 * of just the calling function (confusing, eh?).
629 * It's useful when the call to kmalloc comes from a widely-used standard
630 * allocator where we care about the real place the memory allocation
631 * request comes from.
632 */
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300633extern void *__kmalloc_track_caller(size_t, gfp_t, unsigned long);
Christoph Hellwig1d2c8ee2006-10-04 02:15:25 -0700634#define kmalloc_track_caller(size, flags) \
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300635 __kmalloc_track_caller(size, flags, _RET_IP_)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636
Manfred Spraul97e2bde2005-05-01 08:58:38 -0700637#ifdef CONFIG_NUMA
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300638extern void *__kmalloc_node_track_caller(size_t, gfp_t, int, unsigned long);
Christoph Hellwig8b98c162006-12-06 20:32:30 -0800639#define kmalloc_node_track_caller(size, flags, node) \
640 __kmalloc_node_track_caller(size, flags, node, \
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300641 _RET_IP_)
Christoph Lameter2e892f42006-12-13 00:34:23 -0800642
Christoph Hellwig8b98c162006-12-06 20:32:30 -0800643#else /* CONFIG_NUMA */
Christoph Lameter2e892f42006-12-13 00:34:23 -0800644
645#define kmalloc_node_track_caller(size, flags, node) \
646 kmalloc_track_caller(size, flags)
647
Pascal Terjandfcd3612008-11-25 15:08:19 +0100648#endif /* CONFIG_NUMA */
Christoph Hellwig8b98c162006-12-06 20:32:30 -0800649
Christoph Lameter81cda662007-07-17 04:03:29 -0700650/*
651 * Shortcuts
652 */
653static inline void *kmem_cache_zalloc(struct kmem_cache *k, gfp_t flags)
654{
655 return kmem_cache_alloc(k, flags | __GFP_ZERO);
656}
657
658/**
659 * kzalloc - allocate memory. The memory is set to zero.
660 * @size: how many bytes of memory are required.
661 * @flags: the type of memory to allocate (see kmalloc).
662 */
663static inline void *kzalloc(size_t size, gfp_t flags)
664{
665 return kmalloc(size, flags | __GFP_ZERO);
666}
667
Jeff Layton979b0fe2008-06-05 22:47:00 -0700668/**
669 * kzalloc_node - allocate zeroed memory from a particular memory node.
670 * @size: how many bytes of memory are required.
671 * @flags: the type of memory to allocate (see kmalloc).
672 * @node: memory node from which to allocate
673 */
674static inline void *kzalloc_node(size_t size, gfp_t flags, int node)
675{
676 return kmalloc_node(size, flags | __GFP_ZERO, node);
677}
678
Joonsoo Kim07f361b2014-10-09 15:26:00 -0700679unsigned int kmem_cache_size(struct kmem_cache *s);
Pekka Enberg7e85ee02009-06-12 14:03:06 +0300680void __init kmem_cache_init_late(void);
681
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +0200682#if defined(CONFIG_SMP) && defined(CONFIG_SLAB)
683int slab_prepare_cpu(unsigned int cpu);
684int slab_dead_cpu(unsigned int cpu);
685#else
686#define slab_prepare_cpu NULL
687#define slab_dead_cpu NULL
688#endif
689
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690#endif /* _LINUX_SLAB_H */