blob: 918f374e7156f4739b30c178e5c8beaf5137c4d0 [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>
Kees Cook49b7f892018-05-08 12:52:32 -070016#include <linux/overflow.h>
Andrew Morton1b1cec42006-12-06 20:33:22 -080017#include <linux/types.h>
Glauber Costa1f458cb2012-12-18 14:22:50 -080018#include <linux/workqueue.h>
19
Linus Torvalds1da177e2005-04-16 15:20:36 -070020
Christoph Lameter2e892f42006-12-13 00:34:23 -080021/*
22 * Flags to pass to kmem_cache_create().
David Rientjes124dee02015-04-14 15:44:28 -070023 * The ones marked DEBUG are only valid if CONFIG_DEBUG_SLAB is set.
Linus Torvalds1da177e2005-04-16 15:20:36 -070024 */
Alexey Dobriyand50112e2017-11-15 17:32:18 -080025/* DEBUG: Perform (expensive) checks on alloc/free */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080026#define SLAB_CONSISTENCY_CHECKS ((slab_flags_t __force)0x00000100U)
Alexey Dobriyand50112e2017-11-15 17:32:18 -080027/* DEBUG: Red zone objs in a cache */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080028#define SLAB_RED_ZONE ((slab_flags_t __force)0x00000400U)
Alexey Dobriyand50112e2017-11-15 17:32:18 -080029/* DEBUG: Poison objects */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080030#define SLAB_POISON ((slab_flags_t __force)0x00000800U)
Alexey Dobriyand50112e2017-11-15 17:32:18 -080031/* Align objs on cache lines */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080032#define SLAB_HWCACHE_ALIGN ((slab_flags_t __force)0x00002000U)
Alexey Dobriyand50112e2017-11-15 17:32:18 -080033/* Use GFP_DMA memory */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080034#define SLAB_CACHE_DMA ((slab_flags_t __force)0x00004000U)
Alexey Dobriyand50112e2017-11-15 17:32:18 -080035/* DEBUG: Store the last owner for bug hunting */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080036#define SLAB_STORE_USER ((slab_flags_t __force)0x00010000U)
Alexey Dobriyand50112e2017-11-15 17:32:18 -080037/* Panic if kmem_cache_create() fails */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080038#define SLAB_PANIC ((slab_flags_t __force)0x00040000U)
Peter Zijlstrad7de4c12008-11-13 20:40:12 +020039/*
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -080040 * SLAB_TYPESAFE_BY_RCU - **WARNING** READ THIS!
Peter Zijlstrad7de4c12008-11-13 20:40:12 +020041 *
42 * This delays freeing the SLAB page by a grace period, it does _NOT_
43 * delay object freeing. This means that if you do kmem_cache_free()
44 * that memory location is free to be reused at any time. Thus it may
45 * be possible to see another object there in the same RCU grace period.
46 *
47 * This feature only ensures the memory location backing the object
48 * stays valid, the trick to using this is relying on an independent
49 * object validation pass. Something like:
50 *
51 * rcu_read_lock()
52 * again:
53 * obj = lockless_lookup(key);
54 * if (obj) {
55 * if (!try_get_ref(obj)) // might fail for free objects
56 * goto again;
57 *
58 * if (obj->key != key) { // not the object we expected
59 * put_ref(obj);
60 * goto again;
61 * }
62 * }
63 * rcu_read_unlock();
64 *
Joonsoo Kim68126702013-10-24 10:07:42 +090065 * This is useful if we need to approach a kernel structure obliquely,
66 * from its address obtained without the usual locking. We can lock
67 * the structure to stabilize it and check it's still at the given address,
68 * only if we can be sure that the memory has not been meanwhile reused
69 * for some other kind of object (which our subsystem's lock might corrupt).
70 *
71 * rcu_read_lock before reading the address, then rcu_read_unlock after
72 * taking the spinlock within the structure expected at that address.
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -080073 *
74 * Note that SLAB_TYPESAFE_BY_RCU was originally named SLAB_DESTROY_BY_RCU.
Peter Zijlstrad7de4c12008-11-13 20:40:12 +020075 */
Alexey Dobriyand50112e2017-11-15 17:32:18 -080076/* Defer freeing slabs to RCU */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080077#define SLAB_TYPESAFE_BY_RCU ((slab_flags_t __force)0x00080000U)
Alexey Dobriyand50112e2017-11-15 17:32:18 -080078/* Spread some memory over cpuset */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080079#define SLAB_MEM_SPREAD ((slab_flags_t __force)0x00100000U)
Alexey Dobriyand50112e2017-11-15 17:32:18 -080080/* Trace allocations and frees */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080081#define SLAB_TRACE ((slab_flags_t __force)0x00200000U)
Linus Torvalds1da177e2005-04-16 15:20:36 -070082
Thomas Gleixner30327ac2008-04-30 00:54:59 -070083/* Flag to prevent checks on free */
84#ifdef CONFIG_DEBUG_OBJECTS
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080085# define SLAB_DEBUG_OBJECTS ((slab_flags_t __force)0x00400000U)
Thomas Gleixner30327ac2008-04-30 00:54:59 -070086#else
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080087# define SLAB_DEBUG_OBJECTS 0
Thomas Gleixner30327ac2008-04-30 00:54:59 -070088#endif
89
Alexey Dobriyand50112e2017-11-15 17:32:18 -080090/* Avoid kmemleak tracing */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080091#define SLAB_NOLEAKTRACE ((slab_flags_t __force)0x00800000U)
Catalin Marinasd5cff632009-06-11 13:22:40 +010092
Alexey Dobriyand50112e2017-11-15 17:32:18 -080093/* Fault injection mark */
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +030094#ifdef CONFIG_FAILSLAB
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080095# define SLAB_FAILSLAB ((slab_flags_t __force)0x02000000U)
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +030096#else
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080097# define SLAB_FAILSLAB 0
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +030098#endif
Alexey Dobriyand50112e2017-11-15 17:32:18 -080099/* Account to memcg */
Kirill Tkhai84c07d12018-08-17 15:47:25 -0700100#ifdef CONFIG_MEMCG_KMEM
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -0800101# define SLAB_ACCOUNT ((slab_flags_t __force)0x04000000U)
Vladimir Davydov230e9fc2016-01-14 15:18:15 -0800102#else
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -0800103# define SLAB_ACCOUNT 0
Vladimir Davydov230e9fc2016-01-14 15:18:15 -0800104#endif
Vegard Nossum2dff4402008-05-31 15:56:17 +0200105
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -0700106#ifdef CONFIG_KASAN
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -0800107#define SLAB_KASAN ((slab_flags_t __force)0x08000000U)
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -0700108#else
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -0800109#define SLAB_KASAN 0
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -0700110#endif
111
Mel Gormane12ba742007-10-16 01:25:52 -0700112/* The following flags affect the page allocator grouping pages by mobility */
Alexey Dobriyand50112e2017-11-15 17:32:18 -0800113/* Objects are reclaimable */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -0800114#define SLAB_RECLAIM_ACCOUNT ((slab_flags_t __force)0x00020000U)
Mel Gormane12ba742007-10-16 01:25:52 -0700115#define SLAB_TEMPORARY SLAB_RECLAIM_ACCOUNT /* Objects are short-lived */
Christoph Lameter2e892f42006-12-13 00:34:23 -0800116/*
Christoph Lameter6cb8f912007-07-17 04:03:22 -0700117 * ZERO_SIZE_PTR will be returned for zero sized kmalloc requests.
118 *
119 * Dereferencing ZERO_SIZE_PTR will lead to a distinct access fault.
120 *
121 * ZERO_SIZE_PTR can be passed to kfree though in the same way that NULL can.
122 * Both make kfree a no-op.
123 */
124#define ZERO_SIZE_PTR ((void *)16)
125
Roland Dreier1d4ec7b2007-07-20 12:13:20 -0700126#define ZERO_OR_NULL_PTR(x) ((unsigned long)(x) <= \
Christoph Lameter6cb8f912007-07-17 04:03:22 -0700127 (unsigned long)ZERO_SIZE_PTR)
128
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
Alexey Dobriyanf4957d52018-04-05 16:20:37 -0700140struct kmem_cache *kmem_cache_create(const char *name, unsigned int size,
141 unsigned int align, slab_flags_t flags,
David Windsor8eb82842017-06-10 22:50:28 -0400142 void (*ctor)(void *));
143struct kmem_cache *kmem_cache_create_usercopy(const char *name,
Alexey Dobriyanf4957d52018-04-05 16:20:37 -0700144 unsigned int size, unsigned int align,
145 slab_flags_t flags,
Alexey Dobriyan7bbdb812018-04-05 16:21:31 -0700146 unsigned int useroffset, unsigned int usersize,
David Windsor8eb82842017-06-10 22:50:28 -0400147 void (*ctor)(void *));
Christoph Lameter2e892f42006-12-13 00:34:23 -0800148void kmem_cache_destroy(struct kmem_cache *);
149int kmem_cache_shrink(struct kmem_cache *);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -0800150
151void memcg_create_kmem_cache(struct mem_cgroup *, struct kmem_cache *);
152void memcg_deactivate_kmem_caches(struct mem_cgroup *);
153void memcg_destroy_kmem_caches(struct mem_cgroup *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154
Christoph Lameter0a31bd52007-05-06 14:49:57 -0700155/*
156 * Please use this macro to create slab caches. Simply specify the
157 * name of the structure and maybe some flags that are listed above.
158 *
159 * The alignment of the struct determines object alignment. If you
160 * f.e. add ____cacheline_aligned_in_smp to the struct declaration
161 * then the objects will be properly aligned in SMP configurations.
162 */
David Windsor8eb82842017-06-10 22:50:28 -0400163#define KMEM_CACHE(__struct, __flags) \
164 kmem_cache_create(#__struct, sizeof(struct __struct), \
165 __alignof__(struct __struct), (__flags), NULL)
166
167/*
168 * To whitelist a single field for copying to/from usercopy, use this
169 * macro instead for KMEM_CACHE() above.
170 */
171#define KMEM_CACHE_USERCOPY(__struct, __flags, __field) \
172 kmem_cache_create_usercopy(#__struct, \
173 sizeof(struct __struct), \
174 __alignof__(struct __struct), (__flags), \
175 offsetof(struct __struct, __field), \
176 sizeof_field(struct __struct, __field), NULL)
Christoph Lameter0a31bd52007-05-06 14:49:57 -0700177
Christoph Lameter2e892f42006-12-13 00:34:23 -0800178/*
Christoph Lameter34504662013-01-10 19:00:53 +0000179 * Common kmalloc functions provided by all allocators
180 */
181void * __must_check __krealloc(const void *, size_t, gfp_t);
182void * __must_check krealloc(const void *, size_t, gfp_t);
183void kfree(const void *);
184void kzfree(const void *);
185size_t ksize(const void *);
186
Kees Cookf5509cc2016-06-07 11:05:33 -0700187#ifdef CONFIG_HAVE_HARDENED_USERCOPY_ALLOCATOR
Kees Cookf4e6e282018-01-10 14:48:22 -0800188void __check_heap_object(const void *ptr, unsigned long n, struct page *page,
189 bool to_user);
Kees Cookf5509cc2016-06-07 11:05:33 -0700190#else
Kees Cookf4e6e282018-01-10 14:48:22 -0800191static inline void __check_heap_object(const void *ptr, unsigned long n,
192 struct page *page, bool to_user) { }
Kees Cookf5509cc2016-06-07 11:05:33 -0700193#endif
194
Christoph Lameterc601fd62013-02-05 16:36:47 +0000195/*
196 * Some archs want to perform DMA into kmalloc caches and need a guaranteed
197 * alignment larger than the alignment of a 64-bit integer.
198 * Setting ARCH_KMALLOC_MINALIGN in arch headers allows that.
199 */
200#if defined(ARCH_DMA_MINALIGN) && ARCH_DMA_MINALIGN > 8
201#define ARCH_KMALLOC_MINALIGN ARCH_DMA_MINALIGN
202#define KMALLOC_MIN_SIZE ARCH_DMA_MINALIGN
203#define KMALLOC_SHIFT_LOW ilog2(ARCH_DMA_MINALIGN)
204#else
205#define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
206#endif
207
Christoph Lameter34504662013-01-10 19:00:53 +0000208/*
Rasmus Villemoes94a58c32015-11-20 15:56:48 -0800209 * Setting ARCH_SLAB_MINALIGN in arch headers allows a different alignment.
210 * Intended for arches that get misalignment faults even for 64 bit integer
211 * aligned buffers.
212 */
213#ifndef ARCH_SLAB_MINALIGN
214#define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
215#endif
216
217/*
218 * kmalloc and friends return ARCH_KMALLOC_MINALIGN aligned
219 * pointers. kmem_cache_alloc and friends return ARCH_SLAB_MINALIGN
220 * aligned pointers.
221 */
222#define __assume_kmalloc_alignment __assume_aligned(ARCH_KMALLOC_MINALIGN)
223#define __assume_slab_alignment __assume_aligned(ARCH_SLAB_MINALIGN)
224#define __assume_page_alignment __assume_aligned(PAGE_SIZE)
225
226/*
Christoph Lameter95a05b42013-01-10 19:14:19 +0000227 * Kmalloc array related definitions
228 */
229
230#ifdef CONFIG_SLAB
231/*
232 * The largest kmalloc size supported by the SLAB allocators is
Christoph Lameter0aa817f2007-05-16 22:11:01 -0700233 * 32 megabyte (2^25) or the maximum allocatable page order if that is
234 * less than 32 MB.
235 *
236 * WARNING: Its not easy to increase this value since the allocators have
237 * to do various tricks to work around compiler limitations in order to
238 * ensure proper constant folding.
239 */
Christoph Lameterdebee072007-06-23 17:16:43 -0700240#define KMALLOC_SHIFT_HIGH ((MAX_ORDER + PAGE_SHIFT - 1) <= 25 ? \
241 (MAX_ORDER + PAGE_SHIFT - 1) : 25)
Christoph Lameter95a05b42013-01-10 19:14:19 +0000242#define KMALLOC_SHIFT_MAX KMALLOC_SHIFT_HIGH
Christoph Lameterc601fd62013-02-05 16:36:47 +0000243#ifndef KMALLOC_SHIFT_LOW
Christoph Lameter95a05b42013-01-10 19:14:19 +0000244#define KMALLOC_SHIFT_LOW 5
Christoph Lameterc601fd62013-02-05 16:36:47 +0000245#endif
Christoph Lameter069e2b352013-06-14 19:55:13 +0000246#endif
247
248#ifdef CONFIG_SLUB
Christoph Lameter95a05b42013-01-10 19:14:19 +0000249/*
Dave Hansen433a91f2014-01-28 14:24:50 -0800250 * SLUB directly allocates requests fitting in to an order-1 page
251 * (PAGE_SIZE*2). Larger requests are passed to the page allocator.
Christoph Lameter95a05b42013-01-10 19:14:19 +0000252 */
253#define KMALLOC_SHIFT_HIGH (PAGE_SHIFT + 1)
Michal Hockobb1107f2017-01-10 16:57:27 -0800254#define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT - 1)
Christoph Lameterc601fd62013-02-05 16:36:47 +0000255#ifndef KMALLOC_SHIFT_LOW
Christoph Lameter95a05b42013-01-10 19:14:19 +0000256#define KMALLOC_SHIFT_LOW 3
257#endif
Christoph Lameterc601fd62013-02-05 16:36:47 +0000258#endif
Christoph Lameter0aa817f2007-05-16 22:11:01 -0700259
Christoph Lameter069e2b352013-06-14 19:55:13 +0000260#ifdef CONFIG_SLOB
261/*
Dave Hansen433a91f2014-01-28 14:24:50 -0800262 * SLOB passes all requests larger than one page to the page allocator.
Christoph Lameter069e2b352013-06-14 19:55:13 +0000263 * No kmalloc array is necessary since objects of different sizes can
264 * be allocated from the same page.
265 */
Christoph Lameter069e2b352013-06-14 19:55:13 +0000266#define KMALLOC_SHIFT_HIGH PAGE_SHIFT
Michal Hockobb1107f2017-01-10 16:57:27 -0800267#define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT - 1)
Christoph Lameter069e2b352013-06-14 19:55:13 +0000268#ifndef KMALLOC_SHIFT_LOW
269#define KMALLOC_SHIFT_LOW 3
270#endif
271#endif
272
Christoph Lameter95a05b42013-01-10 19:14:19 +0000273/* Maximum allocatable size */
274#define KMALLOC_MAX_SIZE (1UL << KMALLOC_SHIFT_MAX)
275/* Maximum size for which we actually use a slab cache */
276#define KMALLOC_MAX_CACHE_SIZE (1UL << KMALLOC_SHIFT_HIGH)
277/* Maximum order allocatable via the slab allocagtor */
278#define KMALLOC_MAX_ORDER (KMALLOC_SHIFT_MAX - PAGE_SHIFT)
Christoph Lameter0aa817f2007-05-16 22:11:01 -0700279
Christoph Lameter90810642011-06-23 09:36:12 -0500280/*
Christoph Lameterce6a5022013-01-10 19:14:19 +0000281 * Kmalloc subsystem.
282 */
Christoph Lameterc601fd62013-02-05 16:36:47 +0000283#ifndef KMALLOC_MIN_SIZE
Christoph Lameter95a05b42013-01-10 19:14:19 +0000284#define KMALLOC_MIN_SIZE (1 << KMALLOC_SHIFT_LOW)
Christoph Lameterce6a5022013-01-10 19:14:19 +0000285#endif
Christoph Lameterce6a5022013-01-10 19:14:19 +0000286
Joonsoo Kim24f870d2014-03-12 17:06:19 +0900287/*
288 * This restriction comes from byte sized index implementation.
289 * Page size is normally 2^12 bytes and, in this case, if we want to use
290 * byte sized index which can represent 2^8 entries, the size of the object
291 * should be equal or greater to 2^12 / 2^8 = 2^4 = 16.
292 * If minimum size of kmalloc is less than 16, we use it as minimum object
293 * size and give up to use byte sized index.
294 */
295#define SLAB_OBJ_MIN_SIZE (KMALLOC_MIN_SIZE < 16 ? \
296 (KMALLOC_MIN_SIZE) : 16)
297
Vlastimil Babka12915232018-10-26 15:05:38 -0700298/*
299 * Whenever changing this, take care of that kmalloc_type() and
300 * create_kmalloc_caches() still work as intended.
301 */
Vlastimil Babkacc252ea2018-10-26 15:05:34 -0700302enum kmalloc_cache_type {
303 KMALLOC_NORMAL = 0,
Vlastimil Babka12915232018-10-26 15:05:38 -0700304 KMALLOC_RECLAIM,
Christoph Lameter9425c582013-01-10 19:12:17 +0000305#ifdef CONFIG_ZONE_DMA
Vlastimil Babkacc252ea2018-10-26 15:05:34 -0700306 KMALLOC_DMA,
Christoph Lameter9425c582013-01-10 19:12:17 +0000307#endif
Vlastimil Babkacc252ea2018-10-26 15:05:34 -0700308 NR_KMALLOC_TYPES
309};
310
311#ifndef CONFIG_SLOB
312extern struct kmem_cache *
313kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1];
314
315static __always_inline enum kmalloc_cache_type kmalloc_type(gfp_t flags)
316{
317 int is_dma = 0;
Vlastimil Babka12915232018-10-26 15:05:38 -0700318 int type_dma = 0;
319 int is_reclaimable;
Vlastimil Babkacc252ea2018-10-26 15:05:34 -0700320
321#ifdef CONFIG_ZONE_DMA
322 is_dma = !!(flags & __GFP_DMA);
Vlastimil Babka12915232018-10-26 15:05:38 -0700323 type_dma = is_dma * KMALLOC_DMA;
Vlastimil Babkacc252ea2018-10-26 15:05:34 -0700324#endif
325
Vlastimil Babka12915232018-10-26 15:05:38 -0700326 is_reclaimable = !!(flags & __GFP_RECLAIMABLE);
327
328 /*
329 * If an allocation is both __GFP_DMA and __GFP_RECLAIMABLE, return
330 * KMALLOC_DMA and effectively ignore __GFP_RECLAIMABLE
331 */
332 return type_dma + (is_reclaimable & !is_dma) * KMALLOC_RECLAIM;
Vlastimil Babkacc252ea2018-10-26 15:05:34 -0700333}
Christoph Lameter9425c582013-01-10 19:12:17 +0000334
Christoph Lameterce6a5022013-01-10 19:14:19 +0000335/*
336 * Figure out which kmalloc slab an allocation of a certain size
337 * belongs to.
338 * 0 = zero alloc
339 * 1 = 65 .. 96 bytes
Rasmus Villemoes1ed58b62015-06-24 16:55:59 -0700340 * 2 = 129 .. 192 bytes
341 * n = 2^(n-1)+1 .. 2^n
Christoph Lameterce6a5022013-01-10 19:14:19 +0000342 */
Alexey Dobriyan36071a22018-04-05 16:20:22 -0700343static __always_inline unsigned int kmalloc_index(size_t size)
Christoph Lameterce6a5022013-01-10 19:14:19 +0000344{
345 if (!size)
346 return 0;
347
348 if (size <= KMALLOC_MIN_SIZE)
349 return KMALLOC_SHIFT_LOW;
350
351 if (KMALLOC_MIN_SIZE <= 32 && size > 64 && size <= 96)
352 return 1;
353 if (KMALLOC_MIN_SIZE <= 64 && size > 128 && size <= 192)
354 return 2;
355 if (size <= 8) return 3;
356 if (size <= 16) return 4;
357 if (size <= 32) return 5;
358 if (size <= 64) return 6;
359 if (size <= 128) return 7;
360 if (size <= 256) return 8;
361 if (size <= 512) return 9;
362 if (size <= 1024) return 10;
363 if (size <= 2 * 1024) return 11;
364 if (size <= 4 * 1024) return 12;
365 if (size <= 8 * 1024) return 13;
366 if (size <= 16 * 1024) return 14;
367 if (size <= 32 * 1024) return 15;
368 if (size <= 64 * 1024) return 16;
369 if (size <= 128 * 1024) return 17;
370 if (size <= 256 * 1024) return 18;
371 if (size <= 512 * 1024) return 19;
372 if (size <= 1024 * 1024) return 20;
373 if (size <= 2 * 1024 * 1024) return 21;
374 if (size <= 4 * 1024 * 1024) return 22;
375 if (size <= 8 * 1024 * 1024) return 23;
376 if (size <= 16 * 1024 * 1024) return 24;
377 if (size <= 32 * 1024 * 1024) return 25;
378 if (size <= 64 * 1024 * 1024) return 26;
379 BUG();
380
381 /* Will never be reached. Needed because the compiler may complain */
382 return -1;
383}
Christoph Lameter069e2b352013-06-14 19:55:13 +0000384#endif /* !CONFIG_SLOB */
Christoph Lameterce6a5022013-01-10 19:14:19 +0000385
Rasmus Villemoes48a270552016-05-19 17:10:55 -0700386void *__kmalloc(size_t size, gfp_t flags) __assume_kmalloc_alignment __malloc;
387void *kmem_cache_alloc(struct kmem_cache *, gfp_t flags) __assume_slab_alignment __malloc;
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -0800388void kmem_cache_free(struct kmem_cache *, void *);
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000389
Christoph Lameter484748f2015-09-04 15:45:34 -0700390/*
Jesper Dangaard Brouer9f706d62016-03-15 14:54:03 -0700391 * Bulk allocation and freeing operations. These are accelerated in an
Christoph Lameter484748f2015-09-04 15:45:34 -0700392 * allocator specific way to avoid taking locks repeatedly or building
393 * metadata structures unnecessarily.
394 *
395 * Note that interrupts must be enabled when calling these functions.
396 */
397void kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
Jesper Dangaard Brouer865762a2015-11-20 15:57:58 -0800398int kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
Christoph Lameter484748f2015-09-04 15:45:34 -0700399
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -0700400/*
401 * Caller must not use kfree_bulk() on memory not originally allocated
402 * by kmalloc(), because the SLOB allocator cannot handle this.
403 */
404static __always_inline void kfree_bulk(size_t size, void **p)
405{
406 kmem_cache_free_bulk(NULL, size, p);
407}
408
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000409#ifdef CONFIG_NUMA
Rasmus Villemoes48a270552016-05-19 17:10:55 -0700410void *__kmalloc_node(size_t size, gfp_t flags, int node) __assume_kmalloc_alignment __malloc;
411void *kmem_cache_alloc_node(struct kmem_cache *, gfp_t flags, int node) __assume_slab_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000412#else
413static __always_inline void *__kmalloc_node(size_t size, gfp_t flags, int node)
414{
415 return __kmalloc(size, flags);
416}
417
418static __always_inline void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t flags, int node)
419{
420 return kmem_cache_alloc(s, flags);
421}
422#endif
423
424#ifdef CONFIG_TRACING
Rasmus Villemoes48a270552016-05-19 17:10:55 -0700425extern void *kmem_cache_alloc_trace(struct kmem_cache *, gfp_t, size_t) __assume_slab_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000426
427#ifdef CONFIG_NUMA
428extern void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
429 gfp_t gfpflags,
Rasmus Villemoes48a270552016-05-19 17:10:55 -0700430 int node, size_t size) __assume_slab_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000431#else
432static __always_inline void *
433kmem_cache_alloc_node_trace(struct kmem_cache *s,
434 gfp_t gfpflags,
435 int node, size_t size)
436{
437 return kmem_cache_alloc_trace(s, gfpflags, size);
438}
439#endif /* CONFIG_NUMA */
440
441#else /* CONFIG_TRACING */
442static __always_inline void *kmem_cache_alloc_trace(struct kmem_cache *s,
443 gfp_t flags, size_t size)
444{
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800445 void *ret = kmem_cache_alloc(s, flags);
446
Alexander Potapenko505f5dc2016-03-25 14:22:02 -0700447 kasan_kmalloc(s, ret, size, flags);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800448 return ret;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000449}
450
451static __always_inline void *
452kmem_cache_alloc_node_trace(struct kmem_cache *s,
453 gfp_t gfpflags,
454 int node, size_t size)
455{
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800456 void *ret = kmem_cache_alloc_node(s, gfpflags, node);
457
Alexander Potapenko505f5dc2016-03-25 14:22:02 -0700458 kasan_kmalloc(s, ret, size, gfpflags);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800459 return ret;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000460}
461#endif /* CONFIG_TRACING */
462
Rasmus Villemoes48a270552016-05-19 17:10:55 -0700463extern void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) __assume_page_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000464
465#ifdef CONFIG_TRACING
Rasmus Villemoes48a270552016-05-19 17:10:55 -0700466extern void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) __assume_page_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000467#else
468static __always_inline void *
469kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
470{
471 return kmalloc_order(size, flags, order);
472}
Christoph Lameterce6a5022013-01-10 19:14:19 +0000473#endif
474
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000475static __always_inline void *kmalloc_large(size_t size, gfp_t flags)
476{
477 unsigned int order = get_order(size);
478 return kmalloc_order_trace(size, flags, order);
479}
480
481/**
482 * kmalloc - allocate memory
483 * @size: how many bytes of memory are required.
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800484 * @flags: the type of memory to allocate.
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000485 *
486 * kmalloc is the normal method of allocating memory
487 * for objects smaller than page size in the kernel.
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800488 *
489 * The @flags argument may be one of:
490 *
491 * %GFP_USER - Allocate memory on behalf of user. May sleep.
492 *
493 * %GFP_KERNEL - Allocate normal kernel ram. May sleep.
494 *
495 * %GFP_ATOMIC - Allocation will not sleep. May use emergency pools.
496 * For example, use this inside interrupt handlers.
497 *
498 * %GFP_HIGHUSER - Allocate pages from high memory.
499 *
500 * %GFP_NOIO - Do not do any I/O at all while trying to get memory.
501 *
502 * %GFP_NOFS - Do not make any fs calls while trying to get memory.
503 *
504 * %GFP_NOWAIT - Allocation will not sleep.
505 *
Johannes Weinere97ca8e52014-03-10 15:49:43 -0700506 * %__GFP_THISNODE - Allocate node-local memory only.
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800507 *
508 * %GFP_DMA - Allocation suitable for DMA.
509 * Should only be used for kmalloc() caches. Otherwise, use a
510 * slab created with SLAB_DMA.
511 *
512 * Also it is possible to set different flags by OR'ing
513 * in one or more of the following additional @flags:
514 *
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800515 * %__GFP_HIGH - This allocation has high priority and may use emergency pools.
516 *
517 * %__GFP_NOFAIL - Indicate that this allocation is in no way allowed to fail
518 * (think twice before using).
519 *
520 * %__GFP_NORETRY - If memory is not immediately available,
521 * then give up at once.
522 *
523 * %__GFP_NOWARN - If allocation fails, don't issue any warnings.
524 *
Michal Hockodcda9b02017-07-12 14:36:45 -0700525 * %__GFP_RETRY_MAYFAIL - Try really hard to succeed the allocation but fail
526 * eventually.
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800527 *
528 * There are other flags available as well, but these are not intended
529 * for general use, and so are not documented here. For a full list of
530 * potential flags, always refer to linux/gfp.h.
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000531 */
532static __always_inline void *kmalloc(size_t size, gfp_t flags)
533{
534 if (__builtin_constant_p(size)) {
Vlastimil Babkacc252ea2018-10-26 15:05:34 -0700535#ifndef CONFIG_SLOB
536 unsigned int index;
537#endif
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000538 if (size > KMALLOC_MAX_CACHE_SIZE)
539 return kmalloc_large(size, flags);
540#ifndef CONFIG_SLOB
Vlastimil Babkacc252ea2018-10-26 15:05:34 -0700541 index = kmalloc_index(size);
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000542
Vlastimil Babkacc252ea2018-10-26 15:05:34 -0700543 if (!index)
544 return ZERO_SIZE_PTR;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000545
Vlastimil Babkacc252ea2018-10-26 15:05:34 -0700546 return kmem_cache_alloc_trace(
547 kmalloc_caches[kmalloc_type(flags)][index],
548 flags, size);
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000549#endif
550 }
551 return __kmalloc(size, flags);
552}
553
Christoph Lameterce6a5022013-01-10 19:14:19 +0000554/*
555 * Determine size used for the nth kmalloc cache.
556 * return size or 0 if a kmalloc cache for that
557 * size does not exist
558 */
Alexey Dobriyan0be70322018-04-05 16:20:26 -0700559static __always_inline unsigned int kmalloc_size(unsigned int n)
Christoph Lameterce6a5022013-01-10 19:14:19 +0000560{
Christoph Lameter069e2b352013-06-14 19:55:13 +0000561#ifndef CONFIG_SLOB
Christoph Lameterce6a5022013-01-10 19:14:19 +0000562 if (n > 2)
Alexey Dobriyan0be70322018-04-05 16:20:26 -0700563 return 1U << n;
Christoph Lameterce6a5022013-01-10 19:14:19 +0000564
565 if (n == 1 && KMALLOC_MIN_SIZE <= 32)
566 return 96;
567
568 if (n == 2 && KMALLOC_MIN_SIZE <= 64)
569 return 192;
Christoph Lameter069e2b352013-06-14 19:55:13 +0000570#endif
Christoph Lameterce6a5022013-01-10 19:14:19 +0000571 return 0;
572}
Christoph Lameterce6a5022013-01-10 19:14:19 +0000573
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000574static __always_inline void *kmalloc_node(size_t size, gfp_t flags, int node)
575{
576#ifndef CONFIG_SLOB
577 if (__builtin_constant_p(size) &&
Vlastimil Babkacc252ea2018-10-26 15:05:34 -0700578 size <= KMALLOC_MAX_CACHE_SIZE) {
Alexey Dobriyan36071a22018-04-05 16:20:22 -0700579 unsigned int i = kmalloc_index(size);
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000580
581 if (!i)
582 return ZERO_SIZE_PTR;
583
Vlastimil Babkacc252ea2018-10-26 15:05:34 -0700584 return kmem_cache_alloc_node_trace(
585 kmalloc_caches[kmalloc_type(flags)][i],
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000586 flags, node, size);
587 }
588#endif
589 return __kmalloc_node(size, flags, node);
590}
591
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800592struct memcg_cache_array {
593 struct rcu_head rcu;
594 struct kmem_cache *entries[0];
595};
596
Christoph Lameter0aa817f2007-05-16 22:11:01 -0700597/*
Glauber Costaba6c4962012-12-18 14:22:27 -0800598 * This is the main placeholder for memcg-related information in kmem caches.
Glauber Costaba6c4962012-12-18 14:22:27 -0800599 * Both the root cache and the child caches will have it. For the root cache,
600 * this will hold a dynamically allocated array large enough to hold
Vladimir Davydovf8570262014-01-23 15:53:06 -0800601 * information about the currently limited memcgs in the system. To allow the
602 * array to be accessed without taking any locks, on relocation we free the old
603 * version only after a grace period.
Glauber Costaba6c4962012-12-18 14:22:27 -0800604 *
Tejun Heo9eeadc82017-02-22 15:41:17 -0800605 * Root and child caches hold different metadata.
Glauber Costaba6c4962012-12-18 14:22:27 -0800606 *
Tejun Heo9eeadc82017-02-22 15:41:17 -0800607 * @root_cache: Common to root and child caches. NULL for root, pointer to
608 * the root cache for children.
Vladimir Davydov426589f2015-02-12 14:59:23 -0800609 *
Tejun Heo9eeadc82017-02-22 15:41:17 -0800610 * The following fields are specific to root caches.
611 *
612 * @memcg_caches: kmemcg ID indexed table of child caches. This table is
613 * used to index child cachces during allocation and cleared
614 * early during shutdown.
615 *
Tejun Heo510ded32017-02-22 15:41:24 -0800616 * @root_caches_node: List node for slab_root_caches list.
617 *
Tejun Heo9eeadc82017-02-22 15:41:17 -0800618 * @children: List of all child caches. While the child caches are also
619 * reachable through @memcg_caches, a child cache remains on
620 * this list until it is actually destroyed.
621 *
622 * The following fields are specific to child caches.
623 *
624 * @memcg: Pointer to the memcg this cache belongs to.
625 *
626 * @children_node: List node for @root_cache->children list.
Tejun Heobc2791f2017-02-22 15:41:21 -0800627 *
628 * @kmem_caches_node: List node for @memcg->kmem_caches list.
Glauber Costaba6c4962012-12-18 14:22:27 -0800629 */
630struct memcg_cache_params {
Tejun Heo9eeadc82017-02-22 15:41:17 -0800631 struct kmem_cache *root_cache;
Glauber Costaba6c4962012-12-18 14:22:27 -0800632 union {
Tejun Heo9eeadc82017-02-22 15:41:17 -0800633 struct {
634 struct memcg_cache_array __rcu *memcg_caches;
Tejun Heo510ded32017-02-22 15:41:24 -0800635 struct list_head __root_caches_node;
Tejun Heo9eeadc82017-02-22 15:41:17 -0800636 struct list_head children;
Shakeel Butt92ee3832018-06-14 15:26:27 -0700637 bool dying;
Tejun Heo9eeadc82017-02-22 15:41:17 -0800638 };
Glauber Costa2633d7a2012-12-18 14:22:34 -0800639 struct {
640 struct mem_cgroup *memcg;
Tejun Heo9eeadc82017-02-22 15:41:17 -0800641 struct list_head children_node;
Tejun Heobc2791f2017-02-22 15:41:21 -0800642 struct list_head kmem_caches_node;
Tejun Heo01fb58b2017-02-22 15:41:30 -0800643
644 void (*deact_fn)(struct kmem_cache *);
645 union {
646 struct rcu_head deact_rcu_head;
647 struct work_struct deact_work;
648 };
Glauber Costa2633d7a2012-12-18 14:22:34 -0800649 };
Glauber Costaba6c4962012-12-18 14:22:27 -0800650 };
651};
652
Glauber Costa2633d7a2012-12-18 14:22:34 -0800653int memcg_update_all_caches(int num_memcgs);
654
Christoph Lameter2e892f42006-12-13 00:34:23 -0800655/**
Michael Opdenackere7efa612013-06-25 18:16:55 +0200656 * kmalloc_array - allocate memory for an array.
657 * @n: number of elements.
658 * @size: element size.
659 * @flags: the type of memory to allocate (see kmalloc).
Paul Drynoff800590f2006-06-23 02:03:48 -0700660 */
Xi Wanga8203722012-03-05 15:14:41 -0800661static inline void *kmalloc_array(size_t n, size_t size, gfp_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662{
Kees Cook49b7f892018-05-08 12:52:32 -0700663 size_t bytes;
664
665 if (unlikely(check_mul_overflow(n, size, &bytes)))
Paul Mundt6193a2f2007-07-15 23:38:22 -0700666 return NULL;
Alexey Dobriyan91c6a052016-07-26 15:22:08 -0700667 if (__builtin_constant_p(n) && __builtin_constant_p(size))
Kees Cook49b7f892018-05-08 12:52:32 -0700668 return kmalloc(bytes, flags);
669 return __kmalloc(bytes, flags);
Xi Wanga8203722012-03-05 15:14:41 -0800670}
671
672/**
673 * kcalloc - allocate memory for an array. The memory is set to zero.
674 * @n: number of elements.
675 * @size: element size.
676 * @flags: the type of memory to allocate (see kmalloc).
677 */
678static inline void *kcalloc(size_t n, size_t size, gfp_t flags)
679{
680 return kmalloc_array(n, size, flags | __GFP_ZERO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681}
682
Christoph Hellwig1d2c8ee2006-10-04 02:15:25 -0700683/*
684 * kmalloc_track_caller is a special version of kmalloc that records the
685 * calling function of the routine calling it for slab leak tracking instead
686 * of just the calling function (confusing, eh?).
687 * It's useful when the call to kmalloc comes from a widely-used standard
688 * allocator where we care about the real place the memory allocation
689 * request comes from.
690 */
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300691extern void *__kmalloc_track_caller(size_t, gfp_t, unsigned long);
Christoph Hellwig1d2c8ee2006-10-04 02:15:25 -0700692#define kmalloc_track_caller(size, flags) \
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300693 __kmalloc_track_caller(size, flags, _RET_IP_)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694
Johannes Thumshirn5799b252017-11-15 17:32:29 -0800695static inline void *kmalloc_array_node(size_t n, size_t size, gfp_t flags,
696 int node)
697{
Kees Cook49b7f892018-05-08 12:52:32 -0700698 size_t bytes;
699
700 if (unlikely(check_mul_overflow(n, size, &bytes)))
Johannes Thumshirn5799b252017-11-15 17:32:29 -0800701 return NULL;
702 if (__builtin_constant_p(n) && __builtin_constant_p(size))
Kees Cook49b7f892018-05-08 12:52:32 -0700703 return kmalloc_node(bytes, flags, node);
704 return __kmalloc_node(bytes, flags, node);
Johannes Thumshirn5799b252017-11-15 17:32:29 -0800705}
706
707static inline void *kcalloc_node(size_t n, size_t size, gfp_t flags, int node)
708{
709 return kmalloc_array_node(n, size, flags | __GFP_ZERO, node);
710}
711
712
Manfred Spraul97e2bde2005-05-01 08:58:38 -0700713#ifdef CONFIG_NUMA
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300714extern void *__kmalloc_node_track_caller(size_t, gfp_t, int, unsigned long);
Christoph Hellwig8b98c162006-12-06 20:32:30 -0800715#define kmalloc_node_track_caller(size, flags, node) \
716 __kmalloc_node_track_caller(size, flags, node, \
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300717 _RET_IP_)
Christoph Lameter2e892f42006-12-13 00:34:23 -0800718
Christoph Hellwig8b98c162006-12-06 20:32:30 -0800719#else /* CONFIG_NUMA */
Christoph Lameter2e892f42006-12-13 00:34:23 -0800720
721#define kmalloc_node_track_caller(size, flags, node) \
722 kmalloc_track_caller(size, flags)
723
Pascal Terjandfcd3612008-11-25 15:08:19 +0100724#endif /* CONFIG_NUMA */
Christoph Hellwig8b98c162006-12-06 20:32:30 -0800725
Christoph Lameter81cda662007-07-17 04:03:29 -0700726/*
727 * Shortcuts
728 */
729static inline void *kmem_cache_zalloc(struct kmem_cache *k, gfp_t flags)
730{
731 return kmem_cache_alloc(k, flags | __GFP_ZERO);
732}
733
734/**
735 * kzalloc - allocate memory. The memory is set to zero.
736 * @size: how many bytes of memory are required.
737 * @flags: the type of memory to allocate (see kmalloc).
738 */
739static inline void *kzalloc(size_t size, gfp_t flags)
740{
741 return kmalloc(size, flags | __GFP_ZERO);
742}
743
Jeff Layton979b0fe2008-06-05 22:47:00 -0700744/**
745 * kzalloc_node - allocate zeroed memory from a particular memory node.
746 * @size: how many bytes of memory are required.
747 * @flags: the type of memory to allocate (see kmalloc).
748 * @node: memory node from which to allocate
749 */
750static inline void *kzalloc_node(size_t size, gfp_t flags, int node)
751{
752 return kmalloc_node(size, flags | __GFP_ZERO, node);
753}
754
Joonsoo Kim07f361b2014-10-09 15:26:00 -0700755unsigned int kmem_cache_size(struct kmem_cache *s);
Pekka Enberg7e85ee02009-06-12 14:03:06 +0300756void __init kmem_cache_init_late(void);
757
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +0200758#if defined(CONFIG_SMP) && defined(CONFIG_SLAB)
759int slab_prepare_cpu(unsigned int cpu);
760int slab_dead_cpu(unsigned int cpu);
761#else
762#define slab_prepare_cpu NULL
763#define slab_dead_cpu NULL
764#endif
765
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766#endif /* _LINUX_SLAB_H */