blob: 231abc8976c52c4f075ac79b9dc83a2ff4611a9e [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
Alexey Dobriyand50112e2017-11-15 17:32:18 -080092/* Fault injection mark */
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +030093#ifdef CONFIG_FAILSLAB
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080094# define SLAB_FAILSLAB ((slab_flags_t __force)0x02000000U)
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +030095#else
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080096# define SLAB_FAILSLAB 0
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +030097#endif
Alexey Dobriyand50112e2017-11-15 17:32:18 -080098/* Account to memcg */
Johannes Weiner127424c2016-01-20 15:02:32 -080099#if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -0800100# define SLAB_ACCOUNT ((slab_flags_t __force)0x04000000U)
Vladimir Davydov230e9fc2016-01-14 15:18:15 -0800101#else
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -0800102# define SLAB_ACCOUNT 0
Vladimir Davydov230e9fc2016-01-14 15:18:15 -0800103#endif
Vegard Nossum2dff4402008-05-31 15:56:17 +0200104
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -0700105#ifdef CONFIG_KASAN
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -0800106#define SLAB_KASAN ((slab_flags_t __force)0x08000000U)
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -0700107#else
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -0800108#define SLAB_KASAN 0
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -0700109#endif
110
Mel Gormane12ba742007-10-16 01:25:52 -0700111/* The following flags affect the page allocator grouping pages by mobility */
Alexey Dobriyand50112e2017-11-15 17:32:18 -0800112/* Objects are reclaimable */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -0800113#define SLAB_RECLAIM_ACCOUNT ((slab_flags_t __force)0x00020000U)
Mel Gormane12ba742007-10-16 01:25:52 -0700114#define SLAB_TEMPORARY SLAB_RECLAIM_ACCOUNT /* Objects are short-lived */
Christoph Lameter2e892f42006-12-13 00:34:23 -0800115/*
Christoph Lameter6cb8f912007-07-17 04:03:22 -0700116 * ZERO_SIZE_PTR will be returned for zero sized kmalloc requests.
117 *
118 * Dereferencing ZERO_SIZE_PTR will lead to a distinct access fault.
119 *
120 * ZERO_SIZE_PTR can be passed to kfree though in the same way that NULL can.
121 * Both make kfree a no-op.
122 */
123#define ZERO_SIZE_PTR ((void *)16)
124
Roland Dreier1d4ec7b2007-07-20 12:13:20 -0700125#define ZERO_OR_NULL_PTR(x) ((unsigned long)(x) <= \
Christoph Lameter6cb8f912007-07-17 04:03:22 -0700126 (unsigned long)ZERO_SIZE_PTR)
127
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000128#include <linux/kmemleak.h>
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800129#include <linux/kasan.h>
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500130
Glauber Costa2633d7a2012-12-18 14:22:34 -0800131struct mem_cgroup;
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500132/*
Christoph Lameter2e892f42006-12-13 00:34:23 -0800133 * struct kmem_cache related prototypes
134 */
135void __init kmem_cache_init(void);
Denis Kirjanovfda90122015-11-05 18:44:59 -0800136bool slab_is_available(void);
Matt Mackall10cef602006-01-08 01:01:45 -0800137
Kees Cook2d891fb2017-11-30 13:04:32 -0800138extern bool usercopy_fallback;
139
David Windsor8eb82842017-06-10 22:50:28 -0400140struct kmem_cache *kmem_cache_create(const char *name, size_t size,
141 size_t align, slab_flags_t flags,
142 void (*ctor)(void *));
143struct kmem_cache *kmem_cache_create_usercopy(const char *name,
144 size_t size, size_t align, slab_flags_t flags,
145 size_t useroffset, size_t usersize,
146 void (*ctor)(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 */
David Windsor8eb82842017-06-10 22:50:28 -0400162#define KMEM_CACHE(__struct, __flags) \
163 kmem_cache_create(#__struct, sizeof(struct __struct), \
164 __alignof__(struct __struct), (__flags), NULL)
165
166/*
167 * To whitelist a single field for copying to/from usercopy, use this
168 * macro instead for KMEM_CACHE() above.
169 */
170#define KMEM_CACHE_USERCOPY(__struct, __flags, __field) \
171 kmem_cache_create_usercopy(#__struct, \
172 sizeof(struct __struct), \
173 __alignof__(struct __struct), (__flags), \
174 offsetof(struct __struct, __field), \
175 sizeof_field(struct __struct, __field), NULL)
Christoph Lameter0a31bd52007-05-06 14:49:57 -0700176
Christoph Lameter2e892f42006-12-13 00:34:23 -0800177/*
Christoph Lameter34504662013-01-10 19:00:53 +0000178 * Common kmalloc functions provided by all allocators
179 */
180void * __must_check __krealloc(const void *, size_t, gfp_t);
181void * __must_check krealloc(const void *, size_t, gfp_t);
182void kfree(const void *);
183void kzfree(const void *);
184size_t ksize(const void *);
185
Kees Cookf5509cc2016-06-07 11:05:33 -0700186#ifdef CONFIG_HAVE_HARDENED_USERCOPY_ALLOCATOR
Kees Cookf4e6e282018-01-10 14:48:22 -0800187void __check_heap_object(const void *ptr, unsigned long n, struct page *page,
188 bool to_user);
Kees Cookf5509cc2016-06-07 11:05:33 -0700189#else
Kees Cookf4e6e282018-01-10 14:48:22 -0800190static inline void __check_heap_object(const void *ptr, unsigned long n,
191 struct page *page, bool to_user) { }
Kees Cookf5509cc2016-06-07 11:05:33 -0700192#endif
193
Christoph Lameterc601fd62013-02-05 16:36:47 +0000194/*
195 * Some archs want to perform DMA into kmalloc caches and need a guaranteed
196 * alignment larger than the alignment of a 64-bit integer.
197 * Setting ARCH_KMALLOC_MINALIGN in arch headers allows that.
198 */
199#if defined(ARCH_DMA_MINALIGN) && ARCH_DMA_MINALIGN > 8
200#define ARCH_KMALLOC_MINALIGN ARCH_DMA_MINALIGN
201#define KMALLOC_MIN_SIZE ARCH_DMA_MINALIGN
202#define KMALLOC_SHIFT_LOW ilog2(ARCH_DMA_MINALIGN)
203#else
204#define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
205#endif
206
Christoph Lameter34504662013-01-10 19:00:53 +0000207/*
Rasmus Villemoes94a58c32015-11-20 15:56:48 -0800208 * Setting ARCH_SLAB_MINALIGN in arch headers allows a different alignment.
209 * Intended for arches that get misalignment faults even for 64 bit integer
210 * aligned buffers.
211 */
212#ifndef ARCH_SLAB_MINALIGN
213#define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
214#endif
215
216/*
217 * kmalloc and friends return ARCH_KMALLOC_MINALIGN aligned
218 * pointers. kmem_cache_alloc and friends return ARCH_SLAB_MINALIGN
219 * aligned pointers.
220 */
221#define __assume_kmalloc_alignment __assume_aligned(ARCH_KMALLOC_MINALIGN)
222#define __assume_slab_alignment __assume_aligned(ARCH_SLAB_MINALIGN)
223#define __assume_page_alignment __assume_aligned(PAGE_SIZE)
224
225/*
Christoph Lameter95a05b42013-01-10 19:14:19 +0000226 * Kmalloc array related definitions
227 */
228
229#ifdef CONFIG_SLAB
230/*
231 * The largest kmalloc size supported by the SLAB allocators is
Christoph Lameter0aa817f2007-05-16 22:11:01 -0700232 * 32 megabyte (2^25) or the maximum allocatable page order if that is
233 * less than 32 MB.
234 *
235 * WARNING: Its not easy to increase this value since the allocators have
236 * to do various tricks to work around compiler limitations in order to
237 * ensure proper constant folding.
238 */
Christoph Lameterdebee072007-06-23 17:16:43 -0700239#define KMALLOC_SHIFT_HIGH ((MAX_ORDER + PAGE_SHIFT - 1) <= 25 ? \
240 (MAX_ORDER + PAGE_SHIFT - 1) : 25)
Christoph Lameter95a05b42013-01-10 19:14:19 +0000241#define KMALLOC_SHIFT_MAX KMALLOC_SHIFT_HIGH
Christoph Lameterc601fd62013-02-05 16:36:47 +0000242#ifndef KMALLOC_SHIFT_LOW
Christoph Lameter95a05b42013-01-10 19:14:19 +0000243#define KMALLOC_SHIFT_LOW 5
Christoph Lameterc601fd62013-02-05 16:36:47 +0000244#endif
Christoph Lameter069e2b352013-06-14 19:55:13 +0000245#endif
246
247#ifdef CONFIG_SLUB
Christoph Lameter95a05b42013-01-10 19:14:19 +0000248/*
Dave Hansen433a91f2014-01-28 14:24:50 -0800249 * SLUB directly allocates requests fitting in to an order-1 page
250 * (PAGE_SIZE*2). Larger requests are passed to the page allocator.
Christoph Lameter95a05b42013-01-10 19:14:19 +0000251 */
252#define KMALLOC_SHIFT_HIGH (PAGE_SHIFT + 1)
Michal Hockobb1107f2017-01-10 16:57:27 -0800253#define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT - 1)
Christoph Lameterc601fd62013-02-05 16:36:47 +0000254#ifndef KMALLOC_SHIFT_LOW
Christoph Lameter95a05b42013-01-10 19:14:19 +0000255#define KMALLOC_SHIFT_LOW 3
256#endif
Christoph Lameterc601fd62013-02-05 16:36:47 +0000257#endif
Christoph Lameter0aa817f2007-05-16 22:11:01 -0700258
Christoph Lameter069e2b352013-06-14 19:55:13 +0000259#ifdef CONFIG_SLOB
260/*
Dave Hansen433a91f2014-01-28 14:24:50 -0800261 * SLOB passes all requests larger than one page to the page allocator.
Christoph Lameter069e2b352013-06-14 19:55:13 +0000262 * No kmalloc array is necessary since objects of different sizes can
263 * be allocated from the same page.
264 */
Christoph Lameter069e2b352013-06-14 19:55:13 +0000265#define KMALLOC_SHIFT_HIGH PAGE_SHIFT
Michal Hockobb1107f2017-01-10 16:57:27 -0800266#define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT - 1)
Christoph Lameter069e2b352013-06-14 19:55:13 +0000267#ifndef KMALLOC_SHIFT_LOW
268#define KMALLOC_SHIFT_LOW 3
269#endif
270#endif
271
Christoph Lameter95a05b42013-01-10 19:14:19 +0000272/* Maximum allocatable size */
273#define KMALLOC_MAX_SIZE (1UL << KMALLOC_SHIFT_MAX)
274/* Maximum size for which we actually use a slab cache */
275#define KMALLOC_MAX_CACHE_SIZE (1UL << KMALLOC_SHIFT_HIGH)
276/* Maximum order allocatable via the slab allocagtor */
277#define KMALLOC_MAX_ORDER (KMALLOC_SHIFT_MAX - PAGE_SHIFT)
Christoph Lameter0aa817f2007-05-16 22:11:01 -0700278
Christoph Lameter90810642011-06-23 09:36:12 -0500279/*
Christoph Lameterce6a5022013-01-10 19:14:19 +0000280 * Kmalloc subsystem.
281 */
Christoph Lameterc601fd62013-02-05 16:36:47 +0000282#ifndef KMALLOC_MIN_SIZE
Christoph Lameter95a05b42013-01-10 19:14:19 +0000283#define KMALLOC_MIN_SIZE (1 << KMALLOC_SHIFT_LOW)
Christoph Lameterce6a5022013-01-10 19:14:19 +0000284#endif
Christoph Lameterce6a5022013-01-10 19:14:19 +0000285
Joonsoo Kim24f870d2014-03-12 17:06:19 +0900286/*
287 * This restriction comes from byte sized index implementation.
288 * Page size is normally 2^12 bytes and, in this case, if we want to use
289 * byte sized index which can represent 2^8 entries, the size of the object
290 * should be equal or greater to 2^12 / 2^8 = 2^4 = 16.
291 * If minimum size of kmalloc is less than 16, we use it as minimum object
292 * size and give up to use byte sized index.
293 */
294#define SLAB_OBJ_MIN_SIZE (KMALLOC_MIN_SIZE < 16 ? \
295 (KMALLOC_MIN_SIZE) : 16)
296
Christoph Lameter069e2b352013-06-14 19:55:13 +0000297#ifndef CONFIG_SLOB
Christoph Lameter9425c582013-01-10 19:12:17 +0000298extern struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1];
299#ifdef CONFIG_ZONE_DMA
300extern struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1];
301#endif
302
Christoph Lameterce6a5022013-01-10 19:14:19 +0000303/*
304 * Figure out which kmalloc slab an allocation of a certain size
305 * belongs to.
306 * 0 = zero alloc
307 * 1 = 65 .. 96 bytes
Rasmus Villemoes1ed58b62015-06-24 16:55:59 -0700308 * 2 = 129 .. 192 bytes
309 * n = 2^(n-1)+1 .. 2^n
Christoph Lameterce6a5022013-01-10 19:14:19 +0000310 */
311static __always_inline int kmalloc_index(size_t size)
312{
313 if (!size)
314 return 0;
315
316 if (size <= KMALLOC_MIN_SIZE)
317 return KMALLOC_SHIFT_LOW;
318
319 if (KMALLOC_MIN_SIZE <= 32 && size > 64 && size <= 96)
320 return 1;
321 if (KMALLOC_MIN_SIZE <= 64 && size > 128 && size <= 192)
322 return 2;
323 if (size <= 8) return 3;
324 if (size <= 16) return 4;
325 if (size <= 32) return 5;
326 if (size <= 64) return 6;
327 if (size <= 128) return 7;
328 if (size <= 256) return 8;
329 if (size <= 512) return 9;
330 if (size <= 1024) return 10;
331 if (size <= 2 * 1024) return 11;
332 if (size <= 4 * 1024) return 12;
333 if (size <= 8 * 1024) return 13;
334 if (size <= 16 * 1024) return 14;
335 if (size <= 32 * 1024) return 15;
336 if (size <= 64 * 1024) return 16;
337 if (size <= 128 * 1024) return 17;
338 if (size <= 256 * 1024) return 18;
339 if (size <= 512 * 1024) return 19;
340 if (size <= 1024 * 1024) return 20;
341 if (size <= 2 * 1024 * 1024) return 21;
342 if (size <= 4 * 1024 * 1024) return 22;
343 if (size <= 8 * 1024 * 1024) return 23;
344 if (size <= 16 * 1024 * 1024) return 24;
345 if (size <= 32 * 1024 * 1024) return 25;
346 if (size <= 64 * 1024 * 1024) return 26;
347 BUG();
348
349 /* Will never be reached. Needed because the compiler may complain */
350 return -1;
351}
Christoph Lameter069e2b352013-06-14 19:55:13 +0000352#endif /* !CONFIG_SLOB */
Christoph Lameterce6a5022013-01-10 19:14:19 +0000353
Rasmus Villemoes48a27052016-05-19 17:10:55 -0700354void *__kmalloc(size_t size, gfp_t flags) __assume_kmalloc_alignment __malloc;
355void *kmem_cache_alloc(struct kmem_cache *, gfp_t flags) __assume_slab_alignment __malloc;
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -0800356void kmem_cache_free(struct kmem_cache *, void *);
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000357
Christoph Lameter484748f2015-09-04 15:45:34 -0700358/*
Jesper Dangaard Brouer9f706d62016-03-15 14:54:03 -0700359 * Bulk allocation and freeing operations. These are accelerated in an
Christoph Lameter484748f2015-09-04 15:45:34 -0700360 * allocator specific way to avoid taking locks repeatedly or building
361 * metadata structures unnecessarily.
362 *
363 * Note that interrupts must be enabled when calling these functions.
364 */
365void kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
Jesper Dangaard Brouer865762a2015-11-20 15:57:58 -0800366int kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
Christoph Lameter484748f2015-09-04 15:45:34 -0700367
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -0700368/*
369 * Caller must not use kfree_bulk() on memory not originally allocated
370 * by kmalloc(), because the SLOB allocator cannot handle this.
371 */
372static __always_inline void kfree_bulk(size_t size, void **p)
373{
374 kmem_cache_free_bulk(NULL, size, p);
375}
376
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000377#ifdef CONFIG_NUMA
Rasmus Villemoes48a27052016-05-19 17:10:55 -0700378void *__kmalloc_node(size_t size, gfp_t flags, int node) __assume_kmalloc_alignment __malloc;
379void *kmem_cache_alloc_node(struct kmem_cache *, gfp_t flags, int node) __assume_slab_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000380#else
381static __always_inline void *__kmalloc_node(size_t size, gfp_t flags, int node)
382{
383 return __kmalloc(size, flags);
384}
385
386static __always_inline void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t flags, int node)
387{
388 return kmem_cache_alloc(s, flags);
389}
390#endif
391
392#ifdef CONFIG_TRACING
Rasmus Villemoes48a27052016-05-19 17:10:55 -0700393extern void *kmem_cache_alloc_trace(struct kmem_cache *, gfp_t, size_t) __assume_slab_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000394
395#ifdef CONFIG_NUMA
396extern void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
397 gfp_t gfpflags,
Rasmus Villemoes48a27052016-05-19 17:10:55 -0700398 int node, size_t size) __assume_slab_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000399#else
400static __always_inline void *
401kmem_cache_alloc_node_trace(struct kmem_cache *s,
402 gfp_t gfpflags,
403 int node, size_t size)
404{
405 return kmem_cache_alloc_trace(s, gfpflags, size);
406}
407#endif /* CONFIG_NUMA */
408
409#else /* CONFIG_TRACING */
410static __always_inline void *kmem_cache_alloc_trace(struct kmem_cache *s,
411 gfp_t flags, size_t size)
412{
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800413 void *ret = kmem_cache_alloc(s, flags);
414
Alexander Potapenko505f5dc2016-03-25 14:22:02 -0700415 kasan_kmalloc(s, ret, size, flags);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800416 return ret;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000417}
418
419static __always_inline void *
420kmem_cache_alloc_node_trace(struct kmem_cache *s,
421 gfp_t gfpflags,
422 int node, size_t size)
423{
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800424 void *ret = kmem_cache_alloc_node(s, gfpflags, node);
425
Alexander Potapenko505f5dc2016-03-25 14:22:02 -0700426 kasan_kmalloc(s, ret, size, gfpflags);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800427 return ret;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000428}
429#endif /* CONFIG_TRACING */
430
Rasmus Villemoes48a27052016-05-19 17:10:55 -0700431extern void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) __assume_page_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000432
433#ifdef CONFIG_TRACING
Rasmus Villemoes48a27052016-05-19 17:10:55 -0700434extern void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) __assume_page_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000435#else
436static __always_inline void *
437kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
438{
439 return kmalloc_order(size, flags, order);
440}
Christoph Lameterce6a5022013-01-10 19:14:19 +0000441#endif
442
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000443static __always_inline void *kmalloc_large(size_t size, gfp_t flags)
444{
445 unsigned int order = get_order(size);
446 return kmalloc_order_trace(size, flags, order);
447}
448
449/**
450 * kmalloc - allocate memory
451 * @size: how many bytes of memory are required.
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800452 * @flags: the type of memory to allocate.
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000453 *
454 * kmalloc is the normal method of allocating memory
455 * for objects smaller than page size in the kernel.
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800456 *
457 * The @flags argument may be one of:
458 *
459 * %GFP_USER - Allocate memory on behalf of user. May sleep.
460 *
461 * %GFP_KERNEL - Allocate normal kernel ram. May sleep.
462 *
463 * %GFP_ATOMIC - Allocation will not sleep. May use emergency pools.
464 * For example, use this inside interrupt handlers.
465 *
466 * %GFP_HIGHUSER - Allocate pages from high memory.
467 *
468 * %GFP_NOIO - Do not do any I/O at all while trying to get memory.
469 *
470 * %GFP_NOFS - Do not make any fs calls while trying to get memory.
471 *
472 * %GFP_NOWAIT - Allocation will not sleep.
473 *
Johannes Weinere97ca8e2014-03-10 15:49:43 -0700474 * %__GFP_THISNODE - Allocate node-local memory only.
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800475 *
476 * %GFP_DMA - Allocation suitable for DMA.
477 * Should only be used for kmalloc() caches. Otherwise, use a
478 * slab created with SLAB_DMA.
479 *
480 * Also it is possible to set different flags by OR'ing
481 * in one or more of the following additional @flags:
482 *
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800483 * %__GFP_HIGH - This allocation has high priority and may use emergency pools.
484 *
485 * %__GFP_NOFAIL - Indicate that this allocation is in no way allowed to fail
486 * (think twice before using).
487 *
488 * %__GFP_NORETRY - If memory is not immediately available,
489 * then give up at once.
490 *
491 * %__GFP_NOWARN - If allocation fails, don't issue any warnings.
492 *
Michal Hockodcda9b02017-07-12 14:36:45 -0700493 * %__GFP_RETRY_MAYFAIL - Try really hard to succeed the allocation but fail
494 * eventually.
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800495 *
496 * There are other flags available as well, but these are not intended
497 * for general use, and so are not documented here. For a full list of
498 * potential flags, always refer to linux/gfp.h.
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000499 */
500static __always_inline void *kmalloc(size_t size, gfp_t flags)
501{
502 if (__builtin_constant_p(size)) {
503 if (size > KMALLOC_MAX_CACHE_SIZE)
504 return kmalloc_large(size, flags);
505#ifndef CONFIG_SLOB
506 if (!(flags & GFP_DMA)) {
507 int index = kmalloc_index(size);
508
509 if (!index)
510 return ZERO_SIZE_PTR;
511
512 return kmem_cache_alloc_trace(kmalloc_caches[index],
513 flags, size);
514 }
515#endif
516 }
517 return __kmalloc(size, flags);
518}
519
Christoph Lameterce6a5022013-01-10 19:14:19 +0000520/*
521 * Determine size used for the nth kmalloc cache.
522 * return size or 0 if a kmalloc cache for that
523 * size does not exist
524 */
525static __always_inline int kmalloc_size(int n)
526{
Christoph Lameter069e2b352013-06-14 19:55:13 +0000527#ifndef CONFIG_SLOB
Christoph Lameterce6a5022013-01-10 19:14:19 +0000528 if (n > 2)
529 return 1 << n;
530
531 if (n == 1 && KMALLOC_MIN_SIZE <= 32)
532 return 96;
533
534 if (n == 2 && KMALLOC_MIN_SIZE <= 64)
535 return 192;
Christoph Lameter069e2b352013-06-14 19:55:13 +0000536#endif
Christoph Lameterce6a5022013-01-10 19:14:19 +0000537 return 0;
538}
Christoph Lameterce6a5022013-01-10 19:14:19 +0000539
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000540static __always_inline void *kmalloc_node(size_t size, gfp_t flags, int node)
541{
542#ifndef CONFIG_SLOB
543 if (__builtin_constant_p(size) &&
Christoph Lameter23774a22013-09-04 19:58:08 +0000544 size <= KMALLOC_MAX_CACHE_SIZE && !(flags & GFP_DMA)) {
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000545 int i = kmalloc_index(size);
546
547 if (!i)
548 return ZERO_SIZE_PTR;
549
550 return kmem_cache_alloc_node_trace(kmalloc_caches[i],
551 flags, node, size);
552 }
553#endif
554 return __kmalloc_node(size, flags, node);
555}
556
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800557struct memcg_cache_array {
558 struct rcu_head rcu;
559 struct kmem_cache *entries[0];
560};
561
Christoph Lameter0aa817f2007-05-16 22:11:01 -0700562/*
Glauber Costaba6c4962012-12-18 14:22:27 -0800563 * This is the main placeholder for memcg-related information in kmem caches.
Glauber Costaba6c4962012-12-18 14:22:27 -0800564 * Both the root cache and the child caches will have it. For the root cache,
565 * this will hold a dynamically allocated array large enough to hold
Vladimir Davydovf8570262014-01-23 15:53:06 -0800566 * information about the currently limited memcgs in the system. To allow the
567 * array to be accessed without taking any locks, on relocation we free the old
568 * version only after a grace period.
Glauber Costaba6c4962012-12-18 14:22:27 -0800569 *
Tejun Heo9eeadc82017-02-22 15:41:17 -0800570 * Root and child caches hold different metadata.
Glauber Costaba6c4962012-12-18 14:22:27 -0800571 *
Tejun Heo9eeadc82017-02-22 15:41:17 -0800572 * @root_cache: Common to root and child caches. NULL for root, pointer to
573 * the root cache for children.
Vladimir Davydov426589f2015-02-12 14:59:23 -0800574 *
Tejun Heo9eeadc82017-02-22 15:41:17 -0800575 * The following fields are specific to root caches.
576 *
577 * @memcg_caches: kmemcg ID indexed table of child caches. This table is
578 * used to index child cachces during allocation and cleared
579 * early during shutdown.
580 *
Tejun Heo510ded32017-02-22 15:41:24 -0800581 * @root_caches_node: List node for slab_root_caches list.
582 *
Tejun Heo9eeadc82017-02-22 15:41:17 -0800583 * @children: List of all child caches. While the child caches are also
584 * reachable through @memcg_caches, a child cache remains on
585 * this list until it is actually destroyed.
586 *
587 * The following fields are specific to child caches.
588 *
589 * @memcg: Pointer to the memcg this cache belongs to.
590 *
591 * @children_node: List node for @root_cache->children list.
Tejun Heobc2791f2017-02-22 15:41:21 -0800592 *
593 * @kmem_caches_node: List node for @memcg->kmem_caches list.
Glauber Costaba6c4962012-12-18 14:22:27 -0800594 */
595struct memcg_cache_params {
Tejun Heo9eeadc82017-02-22 15:41:17 -0800596 struct kmem_cache *root_cache;
Glauber Costaba6c4962012-12-18 14:22:27 -0800597 union {
Tejun Heo9eeadc82017-02-22 15:41:17 -0800598 struct {
599 struct memcg_cache_array __rcu *memcg_caches;
Tejun Heo510ded32017-02-22 15:41:24 -0800600 struct list_head __root_caches_node;
Tejun Heo9eeadc82017-02-22 15:41:17 -0800601 struct list_head children;
602 };
Glauber Costa2633d7a2012-12-18 14:22:34 -0800603 struct {
604 struct mem_cgroup *memcg;
Tejun Heo9eeadc82017-02-22 15:41:17 -0800605 struct list_head children_node;
Tejun Heobc2791f2017-02-22 15:41:21 -0800606 struct list_head kmem_caches_node;
Tejun Heo01fb58b2017-02-22 15:41:30 -0800607
608 void (*deact_fn)(struct kmem_cache *);
609 union {
610 struct rcu_head deact_rcu_head;
611 struct work_struct deact_work;
612 };
Glauber Costa2633d7a2012-12-18 14:22:34 -0800613 };
Glauber Costaba6c4962012-12-18 14:22:27 -0800614 };
615};
616
Glauber Costa2633d7a2012-12-18 14:22:34 -0800617int memcg_update_all_caches(int num_memcgs);
618
Christoph Lameter2e892f42006-12-13 00:34:23 -0800619/**
Michael Opdenackere7efa612013-06-25 18:16:55 +0200620 * kmalloc_array - allocate memory for an array.
621 * @n: number of elements.
622 * @size: element size.
623 * @flags: the type of memory to allocate (see kmalloc).
Paul Drynoff800590f2006-06-23 02:03:48 -0700624 */
Xi Wanga8203722012-03-05 15:14:41 -0800625static inline void *kmalloc_array(size_t n, size_t size, gfp_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626{
Xi Wanga3860c12012-05-31 16:26:04 -0700627 if (size != 0 && n > SIZE_MAX / size)
Paul Mundt6193a2f2007-07-15 23:38:22 -0700628 return NULL;
Alexey Dobriyan91c6a052016-07-26 15:22:08 -0700629 if (__builtin_constant_p(n) && __builtin_constant_p(size))
630 return kmalloc(n * size, flags);
Xi Wanga8203722012-03-05 15:14:41 -0800631 return __kmalloc(n * size, flags);
632}
633
634/**
635 * kcalloc - allocate memory for an array. The memory is set to zero.
636 * @n: number of elements.
637 * @size: element size.
638 * @flags: the type of memory to allocate (see kmalloc).
639 */
640static inline void *kcalloc(size_t n, size_t size, gfp_t flags)
641{
642 return kmalloc_array(n, size, flags | __GFP_ZERO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643}
644
Christoph Hellwig1d2c8ee2006-10-04 02:15:25 -0700645/*
646 * kmalloc_track_caller is a special version of kmalloc that records the
647 * calling function of the routine calling it for slab leak tracking instead
648 * of just the calling function (confusing, eh?).
649 * It's useful when the call to kmalloc comes from a widely-used standard
650 * allocator where we care about the real place the memory allocation
651 * request comes from.
652 */
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300653extern void *__kmalloc_track_caller(size_t, gfp_t, unsigned long);
Christoph Hellwig1d2c8ee2006-10-04 02:15:25 -0700654#define kmalloc_track_caller(size, flags) \
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300655 __kmalloc_track_caller(size, flags, _RET_IP_)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656
Johannes Thumshirn5799b252017-11-15 17:32:29 -0800657static inline void *kmalloc_array_node(size_t n, size_t size, gfp_t flags,
658 int node)
659{
660 if (size != 0 && n > SIZE_MAX / size)
661 return NULL;
662 if (__builtin_constant_p(n) && __builtin_constant_p(size))
663 return kmalloc_node(n * size, flags, node);
664 return __kmalloc_node(n * size, flags, node);
665}
666
667static inline void *kcalloc_node(size_t n, size_t size, gfp_t flags, int node)
668{
669 return kmalloc_array_node(n, size, flags | __GFP_ZERO, node);
670}
671
672
Manfred Spraul97e2bde2005-05-01 08:58:38 -0700673#ifdef CONFIG_NUMA
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300674extern void *__kmalloc_node_track_caller(size_t, gfp_t, int, unsigned long);
Christoph Hellwig8b98c162006-12-06 20:32:30 -0800675#define kmalloc_node_track_caller(size, flags, node) \
676 __kmalloc_node_track_caller(size, flags, node, \
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300677 _RET_IP_)
Christoph Lameter2e892f42006-12-13 00:34:23 -0800678
Christoph Hellwig8b98c162006-12-06 20:32:30 -0800679#else /* CONFIG_NUMA */
Christoph Lameter2e892f42006-12-13 00:34:23 -0800680
681#define kmalloc_node_track_caller(size, flags, node) \
682 kmalloc_track_caller(size, flags)
683
Pascal Terjandfcd3612008-11-25 15:08:19 +0100684#endif /* CONFIG_NUMA */
Christoph Hellwig8b98c162006-12-06 20:32:30 -0800685
Christoph Lameter81cda662007-07-17 04:03:29 -0700686/*
687 * Shortcuts
688 */
689static inline void *kmem_cache_zalloc(struct kmem_cache *k, gfp_t flags)
690{
691 return kmem_cache_alloc(k, flags | __GFP_ZERO);
692}
693
694/**
695 * kzalloc - allocate memory. The memory is set to zero.
696 * @size: how many bytes of memory are required.
697 * @flags: the type of memory to allocate (see kmalloc).
698 */
699static inline void *kzalloc(size_t size, gfp_t flags)
700{
701 return kmalloc(size, flags | __GFP_ZERO);
702}
703
Jeff Layton979b0fe2008-06-05 22:47:00 -0700704/**
705 * kzalloc_node - allocate zeroed memory from a particular memory node.
706 * @size: how many bytes of memory are required.
707 * @flags: the type of memory to allocate (see kmalloc).
708 * @node: memory node from which to allocate
709 */
710static inline void *kzalloc_node(size_t size, gfp_t flags, int node)
711{
712 return kmalloc_node(size, flags | __GFP_ZERO, node);
713}
714
Joonsoo Kim07f361b2014-10-09 15:26:00 -0700715unsigned int kmem_cache_size(struct kmem_cache *s);
Pekka Enberg7e85ee02009-06-12 14:03:06 +0300716void __init kmem_cache_init_late(void);
717
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +0200718#if defined(CONFIG_SMP) && defined(CONFIG_SLAB)
719int slab_prepare_cpu(unsigned int cpu);
720int slab_dead_cpu(unsigned int cpu);
721#else
722#define slab_prepare_cpu NULL
723#define slab_dead_cpu NULL
724#endif
725
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726#endif /* _LINUX_SLAB_H */