blob: af5aa65c7c18c120d9f395aad56d26a18786b7ed [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001/* SPDX-License-Identifier: GPL-2.0 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
Christoph Lameter2e892f42006-12-13 00:34:23 -08003 * Written by Mark Hemment, 1996 (markhe@nextd.demon.co.uk).
4 *
Christoph Lametercde53532008-07-04 09:59:22 -07005 * (C) SGI 2006, Christoph Lameter
Christoph Lameter2e892f42006-12-13 00:34:23 -08006 * Cleaned up and restructured to ease the addition of alternative
7 * implementations of SLAB allocators.
Christoph Lameterf1b6eb62013-09-04 16:35:34 +00008 * (C) Linux Foundation 2008-2013
9 * Unified interface for all slab allocators
Linus Torvalds1da177e2005-04-16 15:20:36 -070010 */
11
12#ifndef _LINUX_SLAB_H
13#define _LINUX_SLAB_H
14
Andrew Morton1b1cec42006-12-06 20:33:22 -080015#include <linux/gfp.h>
Andrew Morton1b1cec42006-12-06 20:33:22 -080016#include <linux/types.h>
Glauber Costa1f458cb2012-12-18 14:22:50 -080017#include <linux/workqueue.h>
18
Linus Torvalds1da177e2005-04-16 15:20:36 -070019
Christoph Lameter2e892f42006-12-13 00:34:23 -080020/*
21 * Flags to pass to kmem_cache_create().
David Rientjes124dee02015-04-14 15:44:28 -070022 * The ones marked DEBUG are only valid if CONFIG_DEBUG_SLAB is set.
Linus Torvalds1da177e2005-04-16 15:20:36 -070023 */
Laura Abbottbecfda62016-03-15 14:55:06 -070024#define SLAB_CONSISTENCY_CHECKS 0x00000100UL /* DEBUG: Perform (expensive) checks on alloc/free */
Christoph Lameter55935a32006-12-13 00:34:24 -080025#define SLAB_RED_ZONE 0x00000400UL /* DEBUG: Red zone objs in a cache */
26#define SLAB_POISON 0x00000800UL /* DEBUG: Poison objects */
27#define SLAB_HWCACHE_ALIGN 0x00002000UL /* Align objs on cache lines */
Christoph Lameter2e892f42006-12-13 00:34:23 -080028#define SLAB_CACHE_DMA 0x00004000UL /* Use GFP_DMA memory */
Christoph Lameter2e892f42006-12-13 00:34:23 -080029#define SLAB_STORE_USER 0x00010000UL /* DEBUG: Store the last owner for bug hunting */
Christoph Lameter2e892f42006-12-13 00:34:23 -080030#define SLAB_PANIC 0x00040000UL /* Panic if kmem_cache_create() fails */
Peter Zijlstrad7de4c12008-11-13 20:40:12 +020031/*
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -080032 * SLAB_TYPESAFE_BY_RCU - **WARNING** READ THIS!
Peter Zijlstrad7de4c12008-11-13 20:40:12 +020033 *
34 * This delays freeing the SLAB page by a grace period, it does _NOT_
35 * delay object freeing. This means that if you do kmem_cache_free()
36 * that memory location is free to be reused at any time. Thus it may
37 * be possible to see another object there in the same RCU grace period.
38 *
39 * This feature only ensures the memory location backing the object
40 * stays valid, the trick to using this is relying on an independent
41 * object validation pass. Something like:
42 *
43 * rcu_read_lock()
44 * again:
45 * obj = lockless_lookup(key);
46 * if (obj) {
47 * if (!try_get_ref(obj)) // might fail for free objects
48 * goto again;
49 *
50 * if (obj->key != key) { // not the object we expected
51 * put_ref(obj);
52 * goto again;
53 * }
54 * }
55 * rcu_read_unlock();
56 *
Joonsoo Kim68126702013-10-24 10:07:42 +090057 * This is useful if we need to approach a kernel structure obliquely,
58 * from its address obtained without the usual locking. We can lock
59 * the structure to stabilize it and check it's still at the given address,
60 * only if we can be sure that the memory has not been meanwhile reused
61 * for some other kind of object (which our subsystem's lock might corrupt).
62 *
63 * rcu_read_lock before reading the address, then rcu_read_unlock after
64 * taking the spinlock within the structure expected at that address.
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -080065 *
66 * Note that SLAB_TYPESAFE_BY_RCU was originally named SLAB_DESTROY_BY_RCU.
Peter Zijlstrad7de4c12008-11-13 20:40:12 +020067 */
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -080068#define SLAB_TYPESAFE_BY_RCU 0x00080000UL /* Defer freeing slabs to RCU */
Paul Jackson101a5002006-03-24 03:16:07 -080069#define SLAB_MEM_SPREAD 0x00100000UL /* Spread some memory over cpuset */
Christoph Lameter81819f02007-05-06 14:49:36 -070070#define SLAB_TRACE 0x00200000UL /* Trace allocations and frees */
Linus Torvalds1da177e2005-04-16 15:20:36 -070071
Thomas Gleixner30327ac2008-04-30 00:54:59 -070072/* Flag to prevent checks on free */
73#ifdef CONFIG_DEBUG_OBJECTS
74# define SLAB_DEBUG_OBJECTS 0x00400000UL
75#else
76# define SLAB_DEBUG_OBJECTS 0x00000000UL
77#endif
78
Catalin Marinasd5cff632009-06-11 13:22:40 +010079#define SLAB_NOLEAKTRACE 0x00800000UL /* Avoid kmemleak tracing */
80
Vegard Nossum2dff4402008-05-31 15:56:17 +020081/* Don't track use of uninitialized memory */
82#ifdef CONFIG_KMEMCHECK
83# define SLAB_NOTRACK 0x01000000UL
84#else
85# define SLAB_NOTRACK 0x00000000UL
86#endif
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +030087#ifdef CONFIG_FAILSLAB
88# define SLAB_FAILSLAB 0x02000000UL /* Fault injection mark */
89#else
90# define SLAB_FAILSLAB 0x00000000UL
91#endif
Johannes Weiner127424c2016-01-20 15:02:32 -080092#if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
Vladimir Davydov230e9fc2016-01-14 15:18:15 -080093# define SLAB_ACCOUNT 0x04000000UL /* Account to memcg */
94#else
95# define SLAB_ACCOUNT 0x00000000UL
96#endif
Vegard Nossum2dff4402008-05-31 15:56:17 +020097
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -070098#ifdef CONFIG_KASAN
99#define SLAB_KASAN 0x08000000UL
100#else
101#define SLAB_KASAN 0x00000000UL
102#endif
103
Mel Gormane12ba742007-10-16 01:25:52 -0700104/* The following flags affect the page allocator grouping pages by mobility */
105#define SLAB_RECLAIM_ACCOUNT 0x00020000UL /* Objects are reclaimable */
106#define SLAB_TEMPORARY SLAB_RECLAIM_ACCOUNT /* Objects are short-lived */
Christoph Lameter2e892f42006-12-13 00:34:23 -0800107/*
Christoph Lameter6cb8f912007-07-17 04:03:22 -0700108 * ZERO_SIZE_PTR will be returned for zero sized kmalloc requests.
109 *
110 * Dereferencing ZERO_SIZE_PTR will lead to a distinct access fault.
111 *
112 * ZERO_SIZE_PTR can be passed to kfree though in the same way that NULL can.
113 * Both make kfree a no-op.
114 */
115#define ZERO_SIZE_PTR ((void *)16)
116
Roland Dreier1d4ec7b2007-07-20 12:13:20 -0700117#define ZERO_OR_NULL_PTR(x) ((unsigned long)(x) <= \
Christoph Lameter6cb8f912007-07-17 04:03:22 -0700118 (unsigned long)ZERO_SIZE_PTR)
119
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000120#include <linux/kmemleak.h>
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800121#include <linux/kasan.h>
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500122
Glauber Costa2633d7a2012-12-18 14:22:34 -0800123struct mem_cgroup;
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500124/*
Christoph Lameter2e892f42006-12-13 00:34:23 -0800125 * struct kmem_cache related prototypes
126 */
127void __init kmem_cache_init(void);
Denis Kirjanovfda90122015-11-05 18:44:59 -0800128bool slab_is_available(void);
Matt Mackall10cef602006-01-08 01:01:45 -0800129
Christoph Lameter2e892f42006-12-13 00:34:23 -0800130struct kmem_cache *kmem_cache_create(const char *, size_t, size_t,
Christoph Lameterebe29732006-12-06 20:32:59 -0800131 unsigned long,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700132 void (*)(void *));
Christoph Lameter2e892f42006-12-13 00:34:23 -0800133void kmem_cache_destroy(struct kmem_cache *);
134int kmem_cache_shrink(struct kmem_cache *);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -0800135
136void memcg_create_kmem_cache(struct mem_cgroup *, struct kmem_cache *);
137void memcg_deactivate_kmem_caches(struct mem_cgroup *);
138void memcg_destroy_kmem_caches(struct mem_cgroup *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139
Christoph Lameter0a31bd52007-05-06 14:49:57 -0700140/*
141 * Please use this macro to create slab caches. Simply specify the
142 * name of the structure and maybe some flags that are listed above.
143 *
144 * The alignment of the struct determines object alignment. If you
145 * f.e. add ____cacheline_aligned_in_smp to the struct declaration
146 * then the objects will be properly aligned in SMP configurations.
147 */
148#define KMEM_CACHE(__struct, __flags) kmem_cache_create(#__struct,\
149 sizeof(struct __struct), __alignof__(struct __struct),\
Paul Mundt20c2df82007-07-20 10:11:58 +0900150 (__flags), NULL)
Christoph Lameter0a31bd52007-05-06 14:49:57 -0700151
Christoph Lameter2e892f42006-12-13 00:34:23 -0800152/*
Christoph Lameter34504662013-01-10 19:00:53 +0000153 * Common kmalloc functions provided by all allocators
154 */
155void * __must_check __krealloc(const void *, size_t, gfp_t);
156void * __must_check krealloc(const void *, size_t, gfp_t);
157void kfree(const void *);
158void kzfree(const void *);
159size_t ksize(const void *);
160
Kees Cookf5509cc2016-06-07 11:05:33 -0700161#ifdef CONFIG_HAVE_HARDENED_USERCOPY_ALLOCATOR
162const char *__check_heap_object(const void *ptr, unsigned long n,
163 struct page *page);
164#else
165static inline const char *__check_heap_object(const void *ptr,
166 unsigned long n,
167 struct page *page)
168{
169 return NULL;
170}
171#endif
172
Christoph Lameterc601fd62013-02-05 16:36:47 +0000173/*
174 * Some archs want to perform DMA into kmalloc caches and need a guaranteed
175 * alignment larger than the alignment of a 64-bit integer.
176 * Setting ARCH_KMALLOC_MINALIGN in arch headers allows that.
177 */
178#if defined(ARCH_DMA_MINALIGN) && ARCH_DMA_MINALIGN > 8
179#define ARCH_KMALLOC_MINALIGN ARCH_DMA_MINALIGN
180#define KMALLOC_MIN_SIZE ARCH_DMA_MINALIGN
181#define KMALLOC_SHIFT_LOW ilog2(ARCH_DMA_MINALIGN)
182#else
183#define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
184#endif
185
Christoph Lameter34504662013-01-10 19:00:53 +0000186/*
Rasmus Villemoes94a58c32015-11-20 15:56:48 -0800187 * Setting ARCH_SLAB_MINALIGN in arch headers allows a different alignment.
188 * Intended for arches that get misalignment faults even for 64 bit integer
189 * aligned buffers.
190 */
191#ifndef ARCH_SLAB_MINALIGN
192#define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
193#endif
194
195/*
196 * kmalloc and friends return ARCH_KMALLOC_MINALIGN aligned
197 * pointers. kmem_cache_alloc and friends return ARCH_SLAB_MINALIGN
198 * aligned pointers.
199 */
200#define __assume_kmalloc_alignment __assume_aligned(ARCH_KMALLOC_MINALIGN)
201#define __assume_slab_alignment __assume_aligned(ARCH_SLAB_MINALIGN)
202#define __assume_page_alignment __assume_aligned(PAGE_SIZE)
203
204/*
Christoph Lameter95a05b42013-01-10 19:14:19 +0000205 * Kmalloc array related definitions
206 */
207
208#ifdef CONFIG_SLAB
209/*
210 * The largest kmalloc size supported by the SLAB allocators is
Christoph Lameter0aa817f2007-05-16 22:11:01 -0700211 * 32 megabyte (2^25) or the maximum allocatable page order if that is
212 * less than 32 MB.
213 *
214 * WARNING: Its not easy to increase this value since the allocators have
215 * to do various tricks to work around compiler limitations in order to
216 * ensure proper constant folding.
217 */
Christoph Lameterdebee072007-06-23 17:16:43 -0700218#define KMALLOC_SHIFT_HIGH ((MAX_ORDER + PAGE_SHIFT - 1) <= 25 ? \
219 (MAX_ORDER + PAGE_SHIFT - 1) : 25)
Christoph Lameter95a05b42013-01-10 19:14:19 +0000220#define KMALLOC_SHIFT_MAX KMALLOC_SHIFT_HIGH
Christoph Lameterc601fd62013-02-05 16:36:47 +0000221#ifndef KMALLOC_SHIFT_LOW
Christoph Lameter95a05b42013-01-10 19:14:19 +0000222#define KMALLOC_SHIFT_LOW 5
Christoph Lameterc601fd62013-02-05 16:36:47 +0000223#endif
Christoph Lameter069e2b352013-06-14 19:55:13 +0000224#endif
225
226#ifdef CONFIG_SLUB
Christoph Lameter95a05b42013-01-10 19:14:19 +0000227/*
Dave Hansen433a91f2014-01-28 14:24:50 -0800228 * SLUB directly allocates requests fitting in to an order-1 page
229 * (PAGE_SIZE*2). Larger requests are passed to the page allocator.
Christoph Lameter95a05b42013-01-10 19:14:19 +0000230 */
231#define KMALLOC_SHIFT_HIGH (PAGE_SHIFT + 1)
Michal Hockobb1107f2017-01-10 16:57:27 -0800232#define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT - 1)
Christoph Lameterc601fd62013-02-05 16:36:47 +0000233#ifndef KMALLOC_SHIFT_LOW
Christoph Lameter95a05b42013-01-10 19:14:19 +0000234#define KMALLOC_SHIFT_LOW 3
235#endif
Christoph Lameterc601fd62013-02-05 16:36:47 +0000236#endif
Christoph Lameter0aa817f2007-05-16 22:11:01 -0700237
Christoph Lameter069e2b352013-06-14 19:55:13 +0000238#ifdef CONFIG_SLOB
239/*
Dave Hansen433a91f2014-01-28 14:24:50 -0800240 * SLOB passes all requests larger than one page to the page allocator.
Christoph Lameter069e2b352013-06-14 19:55:13 +0000241 * No kmalloc array is necessary since objects of different sizes can
242 * be allocated from the same page.
243 */
Christoph Lameter069e2b352013-06-14 19:55:13 +0000244#define KMALLOC_SHIFT_HIGH PAGE_SHIFT
Michal Hockobb1107f2017-01-10 16:57:27 -0800245#define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT - 1)
Christoph Lameter069e2b352013-06-14 19:55:13 +0000246#ifndef KMALLOC_SHIFT_LOW
247#define KMALLOC_SHIFT_LOW 3
248#endif
249#endif
250
Christoph Lameter95a05b42013-01-10 19:14:19 +0000251/* Maximum allocatable size */
252#define KMALLOC_MAX_SIZE (1UL << KMALLOC_SHIFT_MAX)
253/* Maximum size for which we actually use a slab cache */
254#define KMALLOC_MAX_CACHE_SIZE (1UL << KMALLOC_SHIFT_HIGH)
255/* Maximum order allocatable via the slab allocagtor */
256#define KMALLOC_MAX_ORDER (KMALLOC_SHIFT_MAX - PAGE_SHIFT)
Christoph Lameter0aa817f2007-05-16 22:11:01 -0700257
Christoph Lameter90810642011-06-23 09:36:12 -0500258/*
Christoph Lameterce6a5022013-01-10 19:14:19 +0000259 * Kmalloc subsystem.
260 */
Christoph Lameterc601fd62013-02-05 16:36:47 +0000261#ifndef KMALLOC_MIN_SIZE
Christoph Lameter95a05b42013-01-10 19:14:19 +0000262#define KMALLOC_MIN_SIZE (1 << KMALLOC_SHIFT_LOW)
Christoph Lameterce6a5022013-01-10 19:14:19 +0000263#endif
Christoph Lameterce6a5022013-01-10 19:14:19 +0000264
Joonsoo Kim24f870d2014-03-12 17:06:19 +0900265/*
266 * This restriction comes from byte sized index implementation.
267 * Page size is normally 2^12 bytes and, in this case, if we want to use
268 * byte sized index which can represent 2^8 entries, the size of the object
269 * should be equal or greater to 2^12 / 2^8 = 2^4 = 16.
270 * If minimum size of kmalloc is less than 16, we use it as minimum object
271 * size and give up to use byte sized index.
272 */
273#define SLAB_OBJ_MIN_SIZE (KMALLOC_MIN_SIZE < 16 ? \
274 (KMALLOC_MIN_SIZE) : 16)
275
Christoph Lameter069e2b352013-06-14 19:55:13 +0000276#ifndef CONFIG_SLOB
Christoph Lameter9425c582013-01-10 19:12:17 +0000277extern struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1];
278#ifdef CONFIG_ZONE_DMA
279extern struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1];
280#endif
281
Christoph Lameterce6a5022013-01-10 19:14:19 +0000282/*
283 * Figure out which kmalloc slab an allocation of a certain size
284 * belongs to.
285 * 0 = zero alloc
286 * 1 = 65 .. 96 bytes
Rasmus Villemoes1ed58b62015-06-24 16:55:59 -0700287 * 2 = 129 .. 192 bytes
288 * n = 2^(n-1)+1 .. 2^n
Christoph Lameterce6a5022013-01-10 19:14:19 +0000289 */
290static __always_inline int kmalloc_index(size_t size)
291{
292 if (!size)
293 return 0;
294
295 if (size <= KMALLOC_MIN_SIZE)
296 return KMALLOC_SHIFT_LOW;
297
298 if (KMALLOC_MIN_SIZE <= 32 && size > 64 && size <= 96)
299 return 1;
300 if (KMALLOC_MIN_SIZE <= 64 && size > 128 && size <= 192)
301 return 2;
302 if (size <= 8) return 3;
303 if (size <= 16) return 4;
304 if (size <= 32) return 5;
305 if (size <= 64) return 6;
306 if (size <= 128) return 7;
307 if (size <= 256) return 8;
308 if (size <= 512) return 9;
309 if (size <= 1024) return 10;
310 if (size <= 2 * 1024) return 11;
311 if (size <= 4 * 1024) return 12;
312 if (size <= 8 * 1024) return 13;
313 if (size <= 16 * 1024) return 14;
314 if (size <= 32 * 1024) return 15;
315 if (size <= 64 * 1024) return 16;
316 if (size <= 128 * 1024) return 17;
317 if (size <= 256 * 1024) return 18;
318 if (size <= 512 * 1024) return 19;
319 if (size <= 1024 * 1024) return 20;
320 if (size <= 2 * 1024 * 1024) return 21;
321 if (size <= 4 * 1024 * 1024) return 22;
322 if (size <= 8 * 1024 * 1024) return 23;
323 if (size <= 16 * 1024 * 1024) return 24;
324 if (size <= 32 * 1024 * 1024) return 25;
325 if (size <= 64 * 1024 * 1024) return 26;
326 BUG();
327
328 /* Will never be reached. Needed because the compiler may complain */
329 return -1;
330}
Christoph Lameter069e2b352013-06-14 19:55:13 +0000331#endif /* !CONFIG_SLOB */
Christoph Lameterce6a5022013-01-10 19:14:19 +0000332
Rasmus Villemoes48a270552016-05-19 17:10:55 -0700333void *__kmalloc(size_t size, gfp_t flags) __assume_kmalloc_alignment __malloc;
334void *kmem_cache_alloc(struct kmem_cache *, gfp_t flags) __assume_slab_alignment __malloc;
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -0800335void kmem_cache_free(struct kmem_cache *, void *);
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000336
Christoph Lameter484748f2015-09-04 15:45:34 -0700337/*
Jesper Dangaard Brouer9f706d62016-03-15 14:54:03 -0700338 * Bulk allocation and freeing operations. These are accelerated in an
Christoph Lameter484748f2015-09-04 15:45:34 -0700339 * allocator specific way to avoid taking locks repeatedly or building
340 * metadata structures unnecessarily.
341 *
342 * Note that interrupts must be enabled when calling these functions.
343 */
344void kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
Jesper Dangaard Brouer865762a2015-11-20 15:57:58 -0800345int kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
Christoph Lameter484748f2015-09-04 15:45:34 -0700346
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -0700347/*
348 * Caller must not use kfree_bulk() on memory not originally allocated
349 * by kmalloc(), because the SLOB allocator cannot handle this.
350 */
351static __always_inline void kfree_bulk(size_t size, void **p)
352{
353 kmem_cache_free_bulk(NULL, size, p);
354}
355
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000356#ifdef CONFIG_NUMA
Rasmus Villemoes48a270552016-05-19 17:10:55 -0700357void *__kmalloc_node(size_t size, gfp_t flags, int node) __assume_kmalloc_alignment __malloc;
358void *kmem_cache_alloc_node(struct kmem_cache *, gfp_t flags, int node) __assume_slab_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000359#else
360static __always_inline void *__kmalloc_node(size_t size, gfp_t flags, int node)
361{
362 return __kmalloc(size, flags);
363}
364
365static __always_inline void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t flags, int node)
366{
367 return kmem_cache_alloc(s, flags);
368}
369#endif
370
371#ifdef CONFIG_TRACING
Rasmus Villemoes48a270552016-05-19 17:10:55 -0700372extern void *kmem_cache_alloc_trace(struct kmem_cache *, gfp_t, size_t) __assume_slab_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000373
374#ifdef CONFIG_NUMA
375extern void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
376 gfp_t gfpflags,
Rasmus Villemoes48a270552016-05-19 17:10:55 -0700377 int node, size_t size) __assume_slab_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000378#else
379static __always_inline void *
380kmem_cache_alloc_node_trace(struct kmem_cache *s,
381 gfp_t gfpflags,
382 int node, size_t size)
383{
384 return kmem_cache_alloc_trace(s, gfpflags, size);
385}
386#endif /* CONFIG_NUMA */
387
388#else /* CONFIG_TRACING */
389static __always_inline void *kmem_cache_alloc_trace(struct kmem_cache *s,
390 gfp_t flags, size_t size)
391{
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800392 void *ret = kmem_cache_alloc(s, flags);
393
Alexander Potapenko505f5dc2016-03-25 14:22:02 -0700394 kasan_kmalloc(s, ret, size, flags);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800395 return ret;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000396}
397
398static __always_inline void *
399kmem_cache_alloc_node_trace(struct kmem_cache *s,
400 gfp_t gfpflags,
401 int node, size_t size)
402{
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800403 void *ret = kmem_cache_alloc_node(s, gfpflags, node);
404
Alexander Potapenko505f5dc2016-03-25 14:22:02 -0700405 kasan_kmalloc(s, ret, size, gfpflags);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800406 return ret;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000407}
408#endif /* CONFIG_TRACING */
409
Rasmus Villemoes48a270552016-05-19 17:10:55 -0700410extern void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) __assume_page_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000411
412#ifdef CONFIG_TRACING
Rasmus Villemoes48a270552016-05-19 17:10:55 -0700413extern void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) __assume_page_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000414#else
415static __always_inline void *
416kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
417{
418 return kmalloc_order(size, flags, order);
419}
Christoph Lameterce6a5022013-01-10 19:14:19 +0000420#endif
421
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000422static __always_inline void *kmalloc_large(size_t size, gfp_t flags)
423{
424 unsigned int order = get_order(size);
425 return kmalloc_order_trace(size, flags, order);
426}
427
428/**
429 * kmalloc - allocate memory
430 * @size: how many bytes of memory are required.
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800431 * @flags: the type of memory to allocate.
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000432 *
433 * kmalloc is the normal method of allocating memory
434 * for objects smaller than page size in the kernel.
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800435 *
436 * The @flags argument may be one of:
437 *
438 * %GFP_USER - Allocate memory on behalf of user. May sleep.
439 *
440 * %GFP_KERNEL - Allocate normal kernel ram. May sleep.
441 *
442 * %GFP_ATOMIC - Allocation will not sleep. May use emergency pools.
443 * For example, use this inside interrupt handlers.
444 *
445 * %GFP_HIGHUSER - Allocate pages from high memory.
446 *
447 * %GFP_NOIO - Do not do any I/O at all while trying to get memory.
448 *
449 * %GFP_NOFS - Do not make any fs calls while trying to get memory.
450 *
451 * %GFP_NOWAIT - Allocation will not sleep.
452 *
Johannes Weinere97ca8e52014-03-10 15:49:43 -0700453 * %__GFP_THISNODE - Allocate node-local memory only.
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800454 *
455 * %GFP_DMA - Allocation suitable for DMA.
456 * Should only be used for kmalloc() caches. Otherwise, use a
457 * slab created with SLAB_DMA.
458 *
459 * Also it is possible to set different flags by OR'ing
460 * in one or more of the following additional @flags:
461 *
462 * %__GFP_COLD - Request cache-cold pages instead of
463 * trying to return cache-warm pages.
464 *
465 * %__GFP_HIGH - This allocation has high priority and may use emergency pools.
466 *
467 * %__GFP_NOFAIL - Indicate that this allocation is in no way allowed to fail
468 * (think twice before using).
469 *
470 * %__GFP_NORETRY - If memory is not immediately available,
471 * then give up at once.
472 *
473 * %__GFP_NOWARN - If allocation fails, don't issue any warnings.
474 *
Michal Hockodcda9b02017-07-12 14:36:45 -0700475 * %__GFP_RETRY_MAYFAIL - Try really hard to succeed the allocation but fail
476 * eventually.
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800477 *
478 * There are other flags available as well, but these are not intended
479 * for general use, and so are not documented here. For a full list of
480 * potential flags, always refer to linux/gfp.h.
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000481 */
482static __always_inline void *kmalloc(size_t size, gfp_t flags)
483{
484 if (__builtin_constant_p(size)) {
485 if (size > KMALLOC_MAX_CACHE_SIZE)
486 return kmalloc_large(size, flags);
487#ifndef CONFIG_SLOB
488 if (!(flags & GFP_DMA)) {
489 int index = kmalloc_index(size);
490
491 if (!index)
492 return ZERO_SIZE_PTR;
493
494 return kmem_cache_alloc_trace(kmalloc_caches[index],
495 flags, size);
496 }
497#endif
498 }
499 return __kmalloc(size, flags);
500}
501
Christoph Lameterce6a5022013-01-10 19:14:19 +0000502/*
503 * Determine size used for the nth kmalloc cache.
504 * return size or 0 if a kmalloc cache for that
505 * size does not exist
506 */
507static __always_inline int kmalloc_size(int n)
508{
Christoph Lameter069e2b352013-06-14 19:55:13 +0000509#ifndef CONFIG_SLOB
Christoph Lameterce6a5022013-01-10 19:14:19 +0000510 if (n > 2)
511 return 1 << n;
512
513 if (n == 1 && KMALLOC_MIN_SIZE <= 32)
514 return 96;
515
516 if (n == 2 && KMALLOC_MIN_SIZE <= 64)
517 return 192;
Christoph Lameter069e2b352013-06-14 19:55:13 +0000518#endif
Christoph Lameterce6a5022013-01-10 19:14:19 +0000519 return 0;
520}
Christoph Lameterce6a5022013-01-10 19:14:19 +0000521
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000522static __always_inline void *kmalloc_node(size_t size, gfp_t flags, int node)
523{
524#ifndef CONFIG_SLOB
525 if (__builtin_constant_p(size) &&
Christoph Lameter23774a22013-09-04 19:58:08 +0000526 size <= KMALLOC_MAX_CACHE_SIZE && !(flags & GFP_DMA)) {
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000527 int i = kmalloc_index(size);
528
529 if (!i)
530 return ZERO_SIZE_PTR;
531
532 return kmem_cache_alloc_node_trace(kmalloc_caches[i],
533 flags, node, size);
534 }
535#endif
536 return __kmalloc_node(size, flags, node);
537}
538
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800539struct memcg_cache_array {
540 struct rcu_head rcu;
541 struct kmem_cache *entries[0];
542};
543
Christoph Lameter0aa817f2007-05-16 22:11:01 -0700544/*
Glauber Costaba6c4962012-12-18 14:22:27 -0800545 * This is the main placeholder for memcg-related information in kmem caches.
Glauber Costaba6c4962012-12-18 14:22:27 -0800546 * Both the root cache and the child caches will have it. For the root cache,
547 * this will hold a dynamically allocated array large enough to hold
Vladimir Davydovf8570262014-01-23 15:53:06 -0800548 * information about the currently limited memcgs in the system. To allow the
549 * array to be accessed without taking any locks, on relocation we free the old
550 * version only after a grace period.
Glauber Costaba6c4962012-12-18 14:22:27 -0800551 *
Tejun Heo9eeadc82017-02-22 15:41:17 -0800552 * Root and child caches hold different metadata.
Glauber Costaba6c4962012-12-18 14:22:27 -0800553 *
Tejun Heo9eeadc82017-02-22 15:41:17 -0800554 * @root_cache: Common to root and child caches. NULL for root, pointer to
555 * the root cache for children.
Vladimir Davydov426589f2015-02-12 14:59:23 -0800556 *
Tejun Heo9eeadc82017-02-22 15:41:17 -0800557 * The following fields are specific to root caches.
558 *
559 * @memcg_caches: kmemcg ID indexed table of child caches. This table is
560 * used to index child cachces during allocation and cleared
561 * early during shutdown.
562 *
Tejun Heo510ded32017-02-22 15:41:24 -0800563 * @root_caches_node: List node for slab_root_caches list.
564 *
Tejun Heo9eeadc82017-02-22 15:41:17 -0800565 * @children: List of all child caches. While the child caches are also
566 * reachable through @memcg_caches, a child cache remains on
567 * this list until it is actually destroyed.
568 *
569 * The following fields are specific to child caches.
570 *
571 * @memcg: Pointer to the memcg this cache belongs to.
572 *
573 * @children_node: List node for @root_cache->children list.
Tejun Heobc2791f2017-02-22 15:41:21 -0800574 *
575 * @kmem_caches_node: List node for @memcg->kmem_caches list.
Glauber Costaba6c4962012-12-18 14:22:27 -0800576 */
577struct memcg_cache_params {
Tejun Heo9eeadc82017-02-22 15:41:17 -0800578 struct kmem_cache *root_cache;
Glauber Costaba6c4962012-12-18 14:22:27 -0800579 union {
Tejun Heo9eeadc82017-02-22 15:41:17 -0800580 struct {
581 struct memcg_cache_array __rcu *memcg_caches;
Tejun Heo510ded32017-02-22 15:41:24 -0800582 struct list_head __root_caches_node;
Tejun Heo9eeadc82017-02-22 15:41:17 -0800583 struct list_head children;
584 };
Glauber Costa2633d7a2012-12-18 14:22:34 -0800585 struct {
586 struct mem_cgroup *memcg;
Tejun Heo9eeadc82017-02-22 15:41:17 -0800587 struct list_head children_node;
Tejun Heobc2791f2017-02-22 15:41:21 -0800588 struct list_head kmem_caches_node;
Tejun Heo01fb58b2017-02-22 15:41:30 -0800589
590 void (*deact_fn)(struct kmem_cache *);
591 union {
592 struct rcu_head deact_rcu_head;
593 struct work_struct deact_work;
594 };
Glauber Costa2633d7a2012-12-18 14:22:34 -0800595 };
Glauber Costaba6c4962012-12-18 14:22:27 -0800596 };
597};
598
Glauber Costa2633d7a2012-12-18 14:22:34 -0800599int memcg_update_all_caches(int num_memcgs);
600
Christoph Lameter2e892f42006-12-13 00:34:23 -0800601/**
Michael Opdenackere7efa612013-06-25 18:16:55 +0200602 * kmalloc_array - allocate memory for an array.
603 * @n: number of elements.
604 * @size: element size.
605 * @flags: the type of memory to allocate (see kmalloc).
Paul Drynoff800590f2006-06-23 02:03:48 -0700606 */
Xi Wanga8203722012-03-05 15:14:41 -0800607static inline void *kmalloc_array(size_t n, size_t size, gfp_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608{
Xi Wanga3860c12012-05-31 16:26:04 -0700609 if (size != 0 && n > SIZE_MAX / size)
Paul Mundt6193a2f2007-07-15 23:38:22 -0700610 return NULL;
Alexey Dobriyan91c6a052016-07-26 15:22:08 -0700611 if (__builtin_constant_p(n) && __builtin_constant_p(size))
612 return kmalloc(n * size, flags);
Xi Wanga8203722012-03-05 15:14:41 -0800613 return __kmalloc(n * size, flags);
614}
615
616/**
617 * kcalloc - allocate memory for an array. The memory is set to zero.
618 * @n: number of elements.
619 * @size: element size.
620 * @flags: the type of memory to allocate (see kmalloc).
621 */
622static inline void *kcalloc(size_t n, size_t size, gfp_t flags)
623{
624 return kmalloc_array(n, size, flags | __GFP_ZERO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625}
626
Christoph Hellwig1d2c8ee2006-10-04 02:15:25 -0700627/*
628 * kmalloc_track_caller is a special version of kmalloc that records the
629 * calling function of the routine calling it for slab leak tracking instead
630 * of just the calling function (confusing, eh?).
631 * It's useful when the call to kmalloc comes from a widely-used standard
632 * allocator where we care about the real place the memory allocation
633 * request comes from.
634 */
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300635extern void *__kmalloc_track_caller(size_t, gfp_t, unsigned long);
Christoph Hellwig1d2c8ee2006-10-04 02:15:25 -0700636#define kmalloc_track_caller(size, flags) \
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300637 __kmalloc_track_caller(size, flags, _RET_IP_)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638
Manfred Spraul97e2bde2005-05-01 08:58:38 -0700639#ifdef CONFIG_NUMA
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300640extern void *__kmalloc_node_track_caller(size_t, gfp_t, int, unsigned long);
Christoph Hellwig8b98c162006-12-06 20:32:30 -0800641#define kmalloc_node_track_caller(size, flags, node) \
642 __kmalloc_node_track_caller(size, flags, node, \
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300643 _RET_IP_)
Christoph Lameter2e892f42006-12-13 00:34:23 -0800644
Christoph Hellwig8b98c162006-12-06 20:32:30 -0800645#else /* CONFIG_NUMA */
Christoph Lameter2e892f42006-12-13 00:34:23 -0800646
647#define kmalloc_node_track_caller(size, flags, node) \
648 kmalloc_track_caller(size, flags)
649
Pascal Terjandfcd3612008-11-25 15:08:19 +0100650#endif /* CONFIG_NUMA */
Christoph Hellwig8b98c162006-12-06 20:32:30 -0800651
Christoph Lameter81cda662007-07-17 04:03:29 -0700652/*
653 * Shortcuts
654 */
655static inline void *kmem_cache_zalloc(struct kmem_cache *k, gfp_t flags)
656{
657 return kmem_cache_alloc(k, flags | __GFP_ZERO);
658}
659
660/**
661 * kzalloc - allocate memory. The memory is set to zero.
662 * @size: how many bytes of memory are required.
663 * @flags: the type of memory to allocate (see kmalloc).
664 */
665static inline void *kzalloc(size_t size, gfp_t flags)
666{
667 return kmalloc(size, flags | __GFP_ZERO);
668}
669
Jeff Layton979b0fe2008-06-05 22:47:00 -0700670/**
671 * kzalloc_node - allocate zeroed memory from a particular memory node.
672 * @size: how many bytes of memory are required.
673 * @flags: the type of memory to allocate (see kmalloc).
674 * @node: memory node from which to allocate
675 */
676static inline void *kzalloc_node(size_t size, gfp_t flags, int node)
677{
678 return kmalloc_node(size, flags | __GFP_ZERO, node);
679}
680
Joonsoo Kim07f361b2014-10-09 15:26:00 -0700681unsigned int kmem_cache_size(struct kmem_cache *s);
Pekka Enberg7e85ee02009-06-12 14:03:06 +0300682void __init kmem_cache_init_late(void);
683
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +0200684#if defined(CONFIG_SMP) && defined(CONFIG_SLAB)
685int slab_prepare_cpu(unsigned int cpu);
686int slab_dead_cpu(unsigned int cpu);
687#else
688#define slab_prepare_cpu NULL
689#define slab_dead_cpu NULL
690#endif
691
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692#endif /* _LINUX_SLAB_H */