blob: f37cb93768abc634f9afbec174470e25941ac9a6 [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 */
Alexey Dobriyand50112e2017-11-15 17:32:18 -080024/* DEBUG: Perform (expensive) checks on alloc/free */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080025#define SLAB_CONSISTENCY_CHECKS ((slab_flags_t __force)0x00000100U)
Alexey Dobriyand50112e2017-11-15 17:32:18 -080026/* DEBUG: Red zone objs in a cache */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080027#define SLAB_RED_ZONE ((slab_flags_t __force)0x00000400U)
Alexey Dobriyand50112e2017-11-15 17:32:18 -080028/* DEBUG: Poison objects */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080029#define SLAB_POISON ((slab_flags_t __force)0x00000800U)
Alexey Dobriyand50112e2017-11-15 17:32:18 -080030/* Align objs on cache lines */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080031#define SLAB_HWCACHE_ALIGN ((slab_flags_t __force)0x00002000U)
Alexey Dobriyand50112e2017-11-15 17:32:18 -080032/* Use GFP_DMA memory */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080033#define SLAB_CACHE_DMA ((slab_flags_t __force)0x00004000U)
Alexey Dobriyand50112e2017-11-15 17:32:18 -080034/* DEBUG: Store the last owner for bug hunting */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080035#define SLAB_STORE_USER ((slab_flags_t __force)0x00010000U)
Alexey Dobriyand50112e2017-11-15 17:32:18 -080036/* Panic if kmem_cache_create() fails */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080037#define SLAB_PANIC ((slab_flags_t __force)0x00040000U)
Peter Zijlstrad7de4c12008-11-13 20:40:12 +020038/*
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -080039 * SLAB_TYPESAFE_BY_RCU - **WARNING** READ THIS!
Peter Zijlstrad7de4c12008-11-13 20:40:12 +020040 *
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 *
Joonsoo Kim68126702013-10-24 10:07:42 +090064 * 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.
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -080072 *
73 * Note that SLAB_TYPESAFE_BY_RCU was originally named SLAB_DESTROY_BY_RCU.
Peter Zijlstrad7de4c12008-11-13 20:40:12 +020074 */
Alexey Dobriyand50112e2017-11-15 17:32:18 -080075/* Defer freeing slabs to RCU */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080076#define SLAB_TYPESAFE_BY_RCU ((slab_flags_t __force)0x00080000U)
Alexey Dobriyand50112e2017-11-15 17:32:18 -080077/* Spread some memory over cpuset */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080078#define SLAB_MEM_SPREAD ((slab_flags_t __force)0x00100000U)
Alexey Dobriyand50112e2017-11-15 17:32:18 -080079/* Trace allocations and frees */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080080#define SLAB_TRACE ((slab_flags_t __force)0x00200000U)
Linus Torvalds1da177e2005-04-16 15:20:36 -070081
Thomas Gleixner30327ac2008-04-30 00:54:59 -070082/* Flag to prevent checks on free */
83#ifdef CONFIG_DEBUG_OBJECTS
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080084# define SLAB_DEBUG_OBJECTS ((slab_flags_t __force)0x00400000U)
Thomas Gleixner30327ac2008-04-30 00:54:59 -070085#else
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080086# define SLAB_DEBUG_OBJECTS 0
Thomas Gleixner30327ac2008-04-30 00:54:59 -070087#endif
88
Alexey Dobriyand50112e2017-11-15 17:32:18 -080089/* Avoid kmemleak tracing */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080090#define SLAB_NOLEAKTRACE ((slab_flags_t __force)0x00800000U)
Catalin Marinasd5cff632009-06-11 13:22:40 +010091
Vegard Nossum2dff4402008-05-31 15:56:17 +020092/* Don't track use of uninitialized memory */
93#ifdef CONFIG_KMEMCHECK
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080094# define SLAB_NOTRACK ((slab_flags_t __force)0x01000000U)
Vegard Nossum2dff4402008-05-31 15:56:17 +020095#else
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080096# define SLAB_NOTRACK 0
Vegard Nossum2dff4402008-05-31 15:56:17 +020097#endif
Alexey Dobriyand50112e2017-11-15 17:32:18 -080098/* Fault injection mark */
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +030099#ifdef CONFIG_FAILSLAB
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -0800100# define SLAB_FAILSLAB ((slab_flags_t __force)0x02000000U)
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +0300101#else
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -0800102# define SLAB_FAILSLAB 0
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +0300103#endif
Alexey Dobriyand50112e2017-11-15 17:32:18 -0800104/* Account to memcg */
Johannes Weiner127424c2016-01-20 15:02:32 -0800105#if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -0800106# define SLAB_ACCOUNT ((slab_flags_t __force)0x04000000U)
Vladimir Davydov230e9fc2016-01-14 15:18:15 -0800107#else
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -0800108# define SLAB_ACCOUNT 0
Vladimir Davydov230e9fc2016-01-14 15:18:15 -0800109#endif
Vegard Nossum2dff4402008-05-31 15:56:17 +0200110
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -0700111#ifdef CONFIG_KASAN
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -0800112#define SLAB_KASAN ((slab_flags_t __force)0x08000000U)
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -0700113#else
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -0800114#define SLAB_KASAN 0
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -0700115#endif
116
Mel Gormane12ba742007-10-16 01:25:52 -0700117/* The following flags affect the page allocator grouping pages by mobility */
Alexey Dobriyand50112e2017-11-15 17:32:18 -0800118/* Objects are reclaimable */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -0800119#define SLAB_RECLAIM_ACCOUNT ((slab_flags_t __force)0x00020000U)
Mel Gormane12ba742007-10-16 01:25:52 -0700120#define SLAB_TEMPORARY SLAB_RECLAIM_ACCOUNT /* Objects are short-lived */
Christoph Lameter2e892f42006-12-13 00:34:23 -0800121/*
Christoph Lameter6cb8f912007-07-17 04:03:22 -0700122 * ZERO_SIZE_PTR will be returned for zero sized kmalloc requests.
123 *
124 * Dereferencing ZERO_SIZE_PTR will lead to a distinct access fault.
125 *
126 * ZERO_SIZE_PTR can be passed to kfree though in the same way that NULL can.
127 * Both make kfree a no-op.
128 */
129#define ZERO_SIZE_PTR ((void *)16)
130
Roland Dreier1d4ec7b2007-07-20 12:13:20 -0700131#define ZERO_OR_NULL_PTR(x) ((unsigned long)(x) <= \
Christoph Lameter6cb8f912007-07-17 04:03:22 -0700132 (unsigned long)ZERO_SIZE_PTR)
133
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000134#include <linux/kmemleak.h>
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800135#include <linux/kasan.h>
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500136
Glauber Costa2633d7a2012-12-18 14:22:34 -0800137struct mem_cgroup;
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500138/*
Christoph Lameter2e892f42006-12-13 00:34:23 -0800139 * struct kmem_cache related prototypes
140 */
141void __init kmem_cache_init(void);
Denis Kirjanovfda90122015-11-05 18:44:59 -0800142bool slab_is_available(void);
Matt Mackall10cef602006-01-08 01:01:45 -0800143
Christoph Lameter2e892f42006-12-13 00:34:23 -0800144struct kmem_cache *kmem_cache_create(const char *, size_t, size_t,
Alexey Dobriyand50112e2017-11-15 17:32:18 -0800145 slab_flags_t,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700146 void (*)(void *));
Christoph Lameter2e892f42006-12-13 00:34:23 -0800147void kmem_cache_destroy(struct kmem_cache *);
148int kmem_cache_shrink(struct kmem_cache *);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -0800149
150void memcg_create_kmem_cache(struct mem_cgroup *, struct kmem_cache *);
151void memcg_deactivate_kmem_caches(struct mem_cgroup *);
152void memcg_destroy_kmem_caches(struct mem_cgroup *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153
Christoph Lameter0a31bd52007-05-06 14:49:57 -0700154/*
155 * Please use this macro to create slab caches. Simply specify the
156 * name of the structure and maybe some flags that are listed above.
157 *
158 * The alignment of the struct determines object alignment. If you
159 * f.e. add ____cacheline_aligned_in_smp to the struct declaration
160 * then the objects will be properly aligned in SMP configurations.
161 */
162#define KMEM_CACHE(__struct, __flags) kmem_cache_create(#__struct,\
163 sizeof(struct __struct), __alignof__(struct __struct),\
Paul Mundt20c2df82007-07-20 10:11:58 +0900164 (__flags), NULL)
Christoph Lameter0a31bd52007-05-06 14:49:57 -0700165
Christoph Lameter2e892f42006-12-13 00:34:23 -0800166/*
Christoph Lameter34504662013-01-10 19:00:53 +0000167 * Common kmalloc functions provided by all allocators
168 */
169void * __must_check __krealloc(const void *, size_t, gfp_t);
170void * __must_check krealloc(const void *, size_t, gfp_t);
171void kfree(const void *);
172void kzfree(const void *);
173size_t ksize(const void *);
174
Kees Cookf5509cc2016-06-07 11:05:33 -0700175#ifdef CONFIG_HAVE_HARDENED_USERCOPY_ALLOCATOR
176const char *__check_heap_object(const void *ptr, unsigned long n,
177 struct page *page);
178#else
179static inline const char *__check_heap_object(const void *ptr,
180 unsigned long n,
181 struct page *page)
182{
183 return NULL;
184}
185#endif
186
Christoph Lameterc601fd62013-02-05 16:36:47 +0000187/*
188 * Some archs want to perform DMA into kmalloc caches and need a guaranteed
189 * alignment larger than the alignment of a 64-bit integer.
190 * Setting ARCH_KMALLOC_MINALIGN in arch headers allows that.
191 */
192#if defined(ARCH_DMA_MINALIGN) && ARCH_DMA_MINALIGN > 8
193#define ARCH_KMALLOC_MINALIGN ARCH_DMA_MINALIGN
194#define KMALLOC_MIN_SIZE ARCH_DMA_MINALIGN
195#define KMALLOC_SHIFT_LOW ilog2(ARCH_DMA_MINALIGN)
196#else
197#define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
198#endif
199
Christoph Lameter34504662013-01-10 19:00:53 +0000200/*
Rasmus Villemoes94a58c32015-11-20 15:56:48 -0800201 * Setting ARCH_SLAB_MINALIGN in arch headers allows a different alignment.
202 * Intended for arches that get misalignment faults even for 64 bit integer
203 * aligned buffers.
204 */
205#ifndef ARCH_SLAB_MINALIGN
206#define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
207#endif
208
209/*
210 * kmalloc and friends return ARCH_KMALLOC_MINALIGN aligned
211 * pointers. kmem_cache_alloc and friends return ARCH_SLAB_MINALIGN
212 * aligned pointers.
213 */
214#define __assume_kmalloc_alignment __assume_aligned(ARCH_KMALLOC_MINALIGN)
215#define __assume_slab_alignment __assume_aligned(ARCH_SLAB_MINALIGN)
216#define __assume_page_alignment __assume_aligned(PAGE_SIZE)
217
218/*
Christoph Lameter95a05b42013-01-10 19:14:19 +0000219 * Kmalloc array related definitions
220 */
221
222#ifdef CONFIG_SLAB
223/*
224 * The largest kmalloc size supported by the SLAB allocators is
Christoph Lameter0aa817f2007-05-16 22:11:01 -0700225 * 32 megabyte (2^25) or the maximum allocatable page order if that is
226 * less than 32 MB.
227 *
228 * WARNING: Its not easy to increase this value since the allocators have
229 * to do various tricks to work around compiler limitations in order to
230 * ensure proper constant folding.
231 */
Christoph Lameterdebee072007-06-23 17:16:43 -0700232#define KMALLOC_SHIFT_HIGH ((MAX_ORDER + PAGE_SHIFT - 1) <= 25 ? \
233 (MAX_ORDER + PAGE_SHIFT - 1) : 25)
Christoph Lameter95a05b42013-01-10 19:14:19 +0000234#define KMALLOC_SHIFT_MAX KMALLOC_SHIFT_HIGH
Christoph Lameterc601fd62013-02-05 16:36:47 +0000235#ifndef KMALLOC_SHIFT_LOW
Christoph Lameter95a05b42013-01-10 19:14:19 +0000236#define KMALLOC_SHIFT_LOW 5
Christoph Lameterc601fd62013-02-05 16:36:47 +0000237#endif
Christoph Lameter069e2b352013-06-14 19:55:13 +0000238#endif
239
240#ifdef CONFIG_SLUB
Christoph Lameter95a05b42013-01-10 19:14:19 +0000241/*
Dave Hansen433a91f2014-01-28 14:24:50 -0800242 * SLUB directly allocates requests fitting in to an order-1 page
243 * (PAGE_SIZE*2). Larger requests are passed to the page allocator.
Christoph Lameter95a05b42013-01-10 19:14:19 +0000244 */
245#define KMALLOC_SHIFT_HIGH (PAGE_SHIFT + 1)
Michal Hockobb1107f2017-01-10 16:57:27 -0800246#define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT - 1)
Christoph Lameterc601fd62013-02-05 16:36:47 +0000247#ifndef KMALLOC_SHIFT_LOW
Christoph Lameter95a05b42013-01-10 19:14:19 +0000248#define KMALLOC_SHIFT_LOW 3
249#endif
Christoph Lameterc601fd62013-02-05 16:36:47 +0000250#endif
Christoph Lameter0aa817f2007-05-16 22:11:01 -0700251
Christoph Lameter069e2b352013-06-14 19:55:13 +0000252#ifdef CONFIG_SLOB
253/*
Dave Hansen433a91f2014-01-28 14:24:50 -0800254 * SLOB passes all requests larger than one page to the page allocator.
Christoph Lameter069e2b352013-06-14 19:55:13 +0000255 * No kmalloc array is necessary since objects of different sizes can
256 * be allocated from the same page.
257 */
Christoph Lameter069e2b352013-06-14 19:55:13 +0000258#define KMALLOC_SHIFT_HIGH PAGE_SHIFT
Michal Hockobb1107f2017-01-10 16:57:27 -0800259#define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT - 1)
Christoph Lameter069e2b352013-06-14 19:55:13 +0000260#ifndef KMALLOC_SHIFT_LOW
261#define KMALLOC_SHIFT_LOW 3
262#endif
263#endif
264
Christoph Lameter95a05b42013-01-10 19:14:19 +0000265/* Maximum allocatable size */
266#define KMALLOC_MAX_SIZE (1UL << KMALLOC_SHIFT_MAX)
267/* Maximum size for which we actually use a slab cache */
268#define KMALLOC_MAX_CACHE_SIZE (1UL << KMALLOC_SHIFT_HIGH)
269/* Maximum order allocatable via the slab allocagtor */
270#define KMALLOC_MAX_ORDER (KMALLOC_SHIFT_MAX - PAGE_SHIFT)
Christoph Lameter0aa817f2007-05-16 22:11:01 -0700271
Christoph Lameter90810642011-06-23 09:36:12 -0500272/*
Christoph Lameterce6a5022013-01-10 19:14:19 +0000273 * Kmalloc subsystem.
274 */
Christoph Lameterc601fd62013-02-05 16:36:47 +0000275#ifndef KMALLOC_MIN_SIZE
Christoph Lameter95a05b42013-01-10 19:14:19 +0000276#define KMALLOC_MIN_SIZE (1 << KMALLOC_SHIFT_LOW)
Christoph Lameterce6a5022013-01-10 19:14:19 +0000277#endif
Christoph Lameterce6a5022013-01-10 19:14:19 +0000278
Joonsoo Kim24f870d2014-03-12 17:06:19 +0900279/*
280 * This restriction comes from byte sized index implementation.
281 * Page size is normally 2^12 bytes and, in this case, if we want to use
282 * byte sized index which can represent 2^8 entries, the size of the object
283 * should be equal or greater to 2^12 / 2^8 = 2^4 = 16.
284 * If minimum size of kmalloc is less than 16, we use it as minimum object
285 * size and give up to use byte sized index.
286 */
287#define SLAB_OBJ_MIN_SIZE (KMALLOC_MIN_SIZE < 16 ? \
288 (KMALLOC_MIN_SIZE) : 16)
289
Christoph Lameter069e2b352013-06-14 19:55:13 +0000290#ifndef CONFIG_SLOB
Christoph Lameter9425c582013-01-10 19:12:17 +0000291extern struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1];
292#ifdef CONFIG_ZONE_DMA
293extern struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1];
294#endif
295
Christoph Lameterce6a5022013-01-10 19:14:19 +0000296/*
297 * Figure out which kmalloc slab an allocation of a certain size
298 * belongs to.
299 * 0 = zero alloc
300 * 1 = 65 .. 96 bytes
Rasmus Villemoes1ed58b62015-06-24 16:55:59 -0700301 * 2 = 129 .. 192 bytes
302 * n = 2^(n-1)+1 .. 2^n
Christoph Lameterce6a5022013-01-10 19:14:19 +0000303 */
304static __always_inline int kmalloc_index(size_t size)
305{
306 if (!size)
307 return 0;
308
309 if (size <= KMALLOC_MIN_SIZE)
310 return KMALLOC_SHIFT_LOW;
311
312 if (KMALLOC_MIN_SIZE <= 32 && size > 64 && size <= 96)
313 return 1;
314 if (KMALLOC_MIN_SIZE <= 64 && size > 128 && size <= 192)
315 return 2;
316 if (size <= 8) return 3;
317 if (size <= 16) return 4;
318 if (size <= 32) return 5;
319 if (size <= 64) return 6;
320 if (size <= 128) return 7;
321 if (size <= 256) return 8;
322 if (size <= 512) return 9;
323 if (size <= 1024) return 10;
324 if (size <= 2 * 1024) return 11;
325 if (size <= 4 * 1024) return 12;
326 if (size <= 8 * 1024) return 13;
327 if (size <= 16 * 1024) return 14;
328 if (size <= 32 * 1024) return 15;
329 if (size <= 64 * 1024) return 16;
330 if (size <= 128 * 1024) return 17;
331 if (size <= 256 * 1024) return 18;
332 if (size <= 512 * 1024) return 19;
333 if (size <= 1024 * 1024) return 20;
334 if (size <= 2 * 1024 * 1024) return 21;
335 if (size <= 4 * 1024 * 1024) return 22;
336 if (size <= 8 * 1024 * 1024) return 23;
337 if (size <= 16 * 1024 * 1024) return 24;
338 if (size <= 32 * 1024 * 1024) return 25;
339 if (size <= 64 * 1024 * 1024) return 26;
340 BUG();
341
342 /* Will never be reached. Needed because the compiler may complain */
343 return -1;
344}
Christoph Lameter069e2b352013-06-14 19:55:13 +0000345#endif /* !CONFIG_SLOB */
Christoph Lameterce6a5022013-01-10 19:14:19 +0000346
Rasmus Villemoes48a270552016-05-19 17:10:55 -0700347void *__kmalloc(size_t size, gfp_t flags) __assume_kmalloc_alignment __malloc;
348void *kmem_cache_alloc(struct kmem_cache *, gfp_t flags) __assume_slab_alignment __malloc;
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -0800349void kmem_cache_free(struct kmem_cache *, void *);
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000350
Christoph Lameter484748f2015-09-04 15:45:34 -0700351/*
Jesper Dangaard Brouer9f706d62016-03-15 14:54:03 -0700352 * Bulk allocation and freeing operations. These are accelerated in an
Christoph Lameter484748f2015-09-04 15:45:34 -0700353 * allocator specific way to avoid taking locks repeatedly or building
354 * metadata structures unnecessarily.
355 *
356 * Note that interrupts must be enabled when calling these functions.
357 */
358void kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
Jesper Dangaard Brouer865762a2015-11-20 15:57:58 -0800359int kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
Christoph Lameter484748f2015-09-04 15:45:34 -0700360
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -0700361/*
362 * Caller must not use kfree_bulk() on memory not originally allocated
363 * by kmalloc(), because the SLOB allocator cannot handle this.
364 */
365static __always_inline void kfree_bulk(size_t size, void **p)
366{
367 kmem_cache_free_bulk(NULL, size, p);
368}
369
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000370#ifdef CONFIG_NUMA
Rasmus Villemoes48a270552016-05-19 17:10:55 -0700371void *__kmalloc_node(size_t size, gfp_t flags, int node) __assume_kmalloc_alignment __malloc;
372void *kmem_cache_alloc_node(struct kmem_cache *, gfp_t flags, int node) __assume_slab_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000373#else
374static __always_inline void *__kmalloc_node(size_t size, gfp_t flags, int node)
375{
376 return __kmalloc(size, flags);
377}
378
379static __always_inline void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t flags, int node)
380{
381 return kmem_cache_alloc(s, flags);
382}
383#endif
384
385#ifdef CONFIG_TRACING
Rasmus Villemoes48a270552016-05-19 17:10:55 -0700386extern void *kmem_cache_alloc_trace(struct kmem_cache *, gfp_t, size_t) __assume_slab_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000387
388#ifdef CONFIG_NUMA
389extern void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
390 gfp_t gfpflags,
Rasmus Villemoes48a270552016-05-19 17:10:55 -0700391 int node, size_t size) __assume_slab_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000392#else
393static __always_inline void *
394kmem_cache_alloc_node_trace(struct kmem_cache *s,
395 gfp_t gfpflags,
396 int node, size_t size)
397{
398 return kmem_cache_alloc_trace(s, gfpflags, size);
399}
400#endif /* CONFIG_NUMA */
401
402#else /* CONFIG_TRACING */
403static __always_inline void *kmem_cache_alloc_trace(struct kmem_cache *s,
404 gfp_t flags, size_t size)
405{
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800406 void *ret = kmem_cache_alloc(s, flags);
407
Alexander Potapenko505f5dc2016-03-25 14:22:02 -0700408 kasan_kmalloc(s, ret, size, flags);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800409 return ret;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000410}
411
412static __always_inline void *
413kmem_cache_alloc_node_trace(struct kmem_cache *s,
414 gfp_t gfpflags,
415 int node, size_t size)
416{
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800417 void *ret = kmem_cache_alloc_node(s, gfpflags, node);
418
Alexander Potapenko505f5dc2016-03-25 14:22:02 -0700419 kasan_kmalloc(s, ret, size, gfpflags);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800420 return ret;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000421}
422#endif /* CONFIG_TRACING */
423
Rasmus Villemoes48a270552016-05-19 17:10:55 -0700424extern void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) __assume_page_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000425
426#ifdef CONFIG_TRACING
Rasmus Villemoes48a270552016-05-19 17:10:55 -0700427extern void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) __assume_page_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000428#else
429static __always_inline void *
430kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
431{
432 return kmalloc_order(size, flags, order);
433}
Christoph Lameterce6a5022013-01-10 19:14:19 +0000434#endif
435
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000436static __always_inline void *kmalloc_large(size_t size, gfp_t flags)
437{
438 unsigned int order = get_order(size);
439 return kmalloc_order_trace(size, flags, order);
440}
441
442/**
443 * kmalloc - allocate memory
444 * @size: how many bytes of memory are required.
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800445 * @flags: the type of memory to allocate.
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000446 *
447 * kmalloc is the normal method of allocating memory
448 * for objects smaller than page size in the kernel.
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800449 *
450 * The @flags argument may be one of:
451 *
452 * %GFP_USER - Allocate memory on behalf of user. May sleep.
453 *
454 * %GFP_KERNEL - Allocate normal kernel ram. May sleep.
455 *
456 * %GFP_ATOMIC - Allocation will not sleep. May use emergency pools.
457 * For example, use this inside interrupt handlers.
458 *
459 * %GFP_HIGHUSER - Allocate pages from high memory.
460 *
461 * %GFP_NOIO - Do not do any I/O at all while trying to get memory.
462 *
463 * %GFP_NOFS - Do not make any fs calls while trying to get memory.
464 *
465 * %GFP_NOWAIT - Allocation will not sleep.
466 *
Johannes Weinere97ca8e52014-03-10 15:49:43 -0700467 * %__GFP_THISNODE - Allocate node-local memory only.
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800468 *
469 * %GFP_DMA - Allocation suitable for DMA.
470 * Should only be used for kmalloc() caches. Otherwise, use a
471 * slab created with SLAB_DMA.
472 *
473 * Also it is possible to set different flags by OR'ing
474 * in one or more of the following additional @flags:
475 *
476 * %__GFP_COLD - Request cache-cold pages instead of
477 * trying to return cache-warm pages.
478 *
479 * %__GFP_HIGH - This allocation has high priority and may use emergency pools.
480 *
481 * %__GFP_NOFAIL - Indicate that this allocation is in no way allowed to fail
482 * (think twice before using).
483 *
484 * %__GFP_NORETRY - If memory is not immediately available,
485 * then give up at once.
486 *
487 * %__GFP_NOWARN - If allocation fails, don't issue any warnings.
488 *
Michal Hockodcda9b02017-07-12 14:36:45 -0700489 * %__GFP_RETRY_MAYFAIL - Try really hard to succeed the allocation but fail
490 * eventually.
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800491 *
492 * There are other flags available as well, but these are not intended
493 * for general use, and so are not documented here. For a full list of
494 * potential flags, always refer to linux/gfp.h.
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000495 */
496static __always_inline void *kmalloc(size_t size, gfp_t flags)
497{
498 if (__builtin_constant_p(size)) {
499 if (size > KMALLOC_MAX_CACHE_SIZE)
500 return kmalloc_large(size, flags);
501#ifndef CONFIG_SLOB
502 if (!(flags & GFP_DMA)) {
503 int index = kmalloc_index(size);
504
505 if (!index)
506 return ZERO_SIZE_PTR;
507
508 return kmem_cache_alloc_trace(kmalloc_caches[index],
509 flags, size);
510 }
511#endif
512 }
513 return __kmalloc(size, flags);
514}
515
Christoph Lameterce6a5022013-01-10 19:14:19 +0000516/*
517 * Determine size used for the nth kmalloc cache.
518 * return size or 0 if a kmalloc cache for that
519 * size does not exist
520 */
521static __always_inline int kmalloc_size(int n)
522{
Christoph Lameter069e2b352013-06-14 19:55:13 +0000523#ifndef CONFIG_SLOB
Christoph Lameterce6a5022013-01-10 19:14:19 +0000524 if (n > 2)
525 return 1 << n;
526
527 if (n == 1 && KMALLOC_MIN_SIZE <= 32)
528 return 96;
529
530 if (n == 2 && KMALLOC_MIN_SIZE <= 64)
531 return 192;
Christoph Lameter069e2b352013-06-14 19:55:13 +0000532#endif
Christoph Lameterce6a5022013-01-10 19:14:19 +0000533 return 0;
534}
Christoph Lameterce6a5022013-01-10 19:14:19 +0000535
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000536static __always_inline void *kmalloc_node(size_t size, gfp_t flags, int node)
537{
538#ifndef CONFIG_SLOB
539 if (__builtin_constant_p(size) &&
Christoph Lameter23774a22013-09-04 19:58:08 +0000540 size <= KMALLOC_MAX_CACHE_SIZE && !(flags & GFP_DMA)) {
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000541 int i = kmalloc_index(size);
542
543 if (!i)
544 return ZERO_SIZE_PTR;
545
546 return kmem_cache_alloc_node_trace(kmalloc_caches[i],
547 flags, node, size);
548 }
549#endif
550 return __kmalloc_node(size, flags, node);
551}
552
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800553struct memcg_cache_array {
554 struct rcu_head rcu;
555 struct kmem_cache *entries[0];
556};
557
Christoph Lameter0aa817f2007-05-16 22:11:01 -0700558/*
Glauber Costaba6c4962012-12-18 14:22:27 -0800559 * This is the main placeholder for memcg-related information in kmem caches.
Glauber Costaba6c4962012-12-18 14:22:27 -0800560 * Both the root cache and the child caches will have it. For the root cache,
561 * this will hold a dynamically allocated array large enough to hold
Vladimir Davydovf8570262014-01-23 15:53:06 -0800562 * information about the currently limited memcgs in the system. To allow the
563 * array to be accessed without taking any locks, on relocation we free the old
564 * version only after a grace period.
Glauber Costaba6c4962012-12-18 14:22:27 -0800565 *
Tejun Heo9eeadc82017-02-22 15:41:17 -0800566 * Root and child caches hold different metadata.
Glauber Costaba6c4962012-12-18 14:22:27 -0800567 *
Tejun Heo9eeadc82017-02-22 15:41:17 -0800568 * @root_cache: Common to root and child caches. NULL for root, pointer to
569 * the root cache for children.
Vladimir Davydov426589f2015-02-12 14:59:23 -0800570 *
Tejun Heo9eeadc82017-02-22 15:41:17 -0800571 * The following fields are specific to root caches.
572 *
573 * @memcg_caches: kmemcg ID indexed table of child caches. This table is
574 * used to index child cachces during allocation and cleared
575 * early during shutdown.
576 *
Tejun Heo510ded32017-02-22 15:41:24 -0800577 * @root_caches_node: List node for slab_root_caches list.
578 *
Tejun Heo9eeadc82017-02-22 15:41:17 -0800579 * @children: List of all child caches. While the child caches are also
580 * reachable through @memcg_caches, a child cache remains on
581 * this list until it is actually destroyed.
582 *
583 * The following fields are specific to child caches.
584 *
585 * @memcg: Pointer to the memcg this cache belongs to.
586 *
587 * @children_node: List node for @root_cache->children list.
Tejun Heobc2791f2017-02-22 15:41:21 -0800588 *
589 * @kmem_caches_node: List node for @memcg->kmem_caches list.
Glauber Costaba6c4962012-12-18 14:22:27 -0800590 */
591struct memcg_cache_params {
Tejun Heo9eeadc82017-02-22 15:41:17 -0800592 struct kmem_cache *root_cache;
Glauber Costaba6c4962012-12-18 14:22:27 -0800593 union {
Tejun Heo9eeadc82017-02-22 15:41:17 -0800594 struct {
595 struct memcg_cache_array __rcu *memcg_caches;
Tejun Heo510ded32017-02-22 15:41:24 -0800596 struct list_head __root_caches_node;
Tejun Heo9eeadc82017-02-22 15:41:17 -0800597 struct list_head children;
598 };
Glauber Costa2633d7a2012-12-18 14:22:34 -0800599 struct {
600 struct mem_cgroup *memcg;
Tejun Heo9eeadc82017-02-22 15:41:17 -0800601 struct list_head children_node;
Tejun Heobc2791f2017-02-22 15:41:21 -0800602 struct list_head kmem_caches_node;
Tejun Heo01fb58b2017-02-22 15:41:30 -0800603
604 void (*deact_fn)(struct kmem_cache *);
605 union {
606 struct rcu_head deact_rcu_head;
607 struct work_struct deact_work;
608 };
Glauber Costa2633d7a2012-12-18 14:22:34 -0800609 };
Glauber Costaba6c4962012-12-18 14:22:27 -0800610 };
611};
612
Glauber Costa2633d7a2012-12-18 14:22:34 -0800613int memcg_update_all_caches(int num_memcgs);
614
Christoph Lameter2e892f42006-12-13 00:34:23 -0800615/**
Michael Opdenackere7efa612013-06-25 18:16:55 +0200616 * kmalloc_array - allocate memory for an array.
617 * @n: number of elements.
618 * @size: element size.
619 * @flags: the type of memory to allocate (see kmalloc).
Paul Drynoff800590f2006-06-23 02:03:48 -0700620 */
Xi Wanga8203722012-03-05 15:14:41 -0800621static inline void *kmalloc_array(size_t n, size_t size, gfp_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622{
Xi Wanga3860c12012-05-31 16:26:04 -0700623 if (size != 0 && n > SIZE_MAX / size)
Paul Mundt6193a2f2007-07-15 23:38:22 -0700624 return NULL;
Alexey Dobriyan91c6a052016-07-26 15:22:08 -0700625 if (__builtin_constant_p(n) && __builtin_constant_p(size))
626 return kmalloc(n * size, flags);
Xi Wanga8203722012-03-05 15:14:41 -0800627 return __kmalloc(n * size, flags);
628}
629
630/**
631 * kcalloc - allocate memory for an array. The memory is set to zero.
632 * @n: number of elements.
633 * @size: element size.
634 * @flags: the type of memory to allocate (see kmalloc).
635 */
636static inline void *kcalloc(size_t n, size_t size, gfp_t flags)
637{
638 return kmalloc_array(n, size, flags | __GFP_ZERO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639}
640
Christoph Hellwig1d2c8ee2006-10-04 02:15:25 -0700641/*
642 * kmalloc_track_caller is a special version of kmalloc that records the
643 * calling function of the routine calling it for slab leak tracking instead
644 * of just the calling function (confusing, eh?).
645 * It's useful when the call to kmalloc comes from a widely-used standard
646 * allocator where we care about the real place the memory allocation
647 * request comes from.
648 */
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300649extern void *__kmalloc_track_caller(size_t, gfp_t, unsigned long);
Christoph Hellwig1d2c8ee2006-10-04 02:15:25 -0700650#define kmalloc_track_caller(size, flags) \
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300651 __kmalloc_track_caller(size, flags, _RET_IP_)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652
Manfred Spraul97e2bde2005-05-01 08:58:38 -0700653#ifdef CONFIG_NUMA
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300654extern void *__kmalloc_node_track_caller(size_t, gfp_t, int, unsigned long);
Christoph Hellwig8b98c162006-12-06 20:32:30 -0800655#define kmalloc_node_track_caller(size, flags, node) \
656 __kmalloc_node_track_caller(size, flags, node, \
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300657 _RET_IP_)
Christoph Lameter2e892f42006-12-13 00:34:23 -0800658
Christoph Hellwig8b98c162006-12-06 20:32:30 -0800659#else /* CONFIG_NUMA */
Christoph Lameter2e892f42006-12-13 00:34:23 -0800660
661#define kmalloc_node_track_caller(size, flags, node) \
662 kmalloc_track_caller(size, flags)
663
Pascal Terjandfcd3612008-11-25 15:08:19 +0100664#endif /* CONFIG_NUMA */
Christoph Hellwig8b98c162006-12-06 20:32:30 -0800665
Christoph Lameter81cda662007-07-17 04:03:29 -0700666/*
667 * Shortcuts
668 */
669static inline void *kmem_cache_zalloc(struct kmem_cache *k, gfp_t flags)
670{
671 return kmem_cache_alloc(k, flags | __GFP_ZERO);
672}
673
674/**
675 * kzalloc - allocate memory. The memory is set to zero.
676 * @size: how many bytes of memory are required.
677 * @flags: the type of memory to allocate (see kmalloc).
678 */
679static inline void *kzalloc(size_t size, gfp_t flags)
680{
681 return kmalloc(size, flags | __GFP_ZERO);
682}
683
Jeff Layton979b0fe2008-06-05 22:47:00 -0700684/**
685 * kzalloc_node - allocate zeroed memory from a particular memory node.
686 * @size: how many bytes of memory are required.
687 * @flags: the type of memory to allocate (see kmalloc).
688 * @node: memory node from which to allocate
689 */
690static inline void *kzalloc_node(size_t size, gfp_t flags, int node)
691{
692 return kmalloc_node(size, flags | __GFP_ZERO, node);
693}
694
Joonsoo Kim07f361b2014-10-09 15:26:00 -0700695unsigned int kmem_cache_size(struct kmem_cache *s);
Pekka Enberg7e85ee02009-06-12 14:03:06 +0300696void __init kmem_cache_init_late(void);
697
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +0200698#if defined(CONFIG_SMP) && defined(CONFIG_SLAB)
699int slab_prepare_cpu(unsigned int cpu);
700int slab_dead_cpu(unsigned int cpu);
701#else
702#define slab_prepare_cpu NULL
703#define slab_dead_cpu NULL
704#endif
705
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706#endif /* _LINUX_SLAB_H */