blob: dd6897f6201077e942b15804245bdd86122b9194 [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>
Roman Gushchinf0a3a242019-07-11 20:56:27 -070019#include <linux/percpu-refcount.h>
Glauber Costa1f458cb2012-12-18 14:22:50 -080020
Linus Torvalds1da177e2005-04-16 15:20:36 -070021
Christoph Lameter2e892f42006-12-13 00:34:23 -080022/*
23 * Flags to pass to kmem_cache_create().
David Rientjes124dee02015-04-14 15:44:28 -070024 * The ones marked DEBUG are only valid if CONFIG_DEBUG_SLAB is set.
Linus Torvalds1da177e2005-04-16 15:20:36 -070025 */
Alexey Dobriyand50112e2017-11-15 17:32:18 -080026/* DEBUG: Perform (expensive) checks on alloc/free */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080027#define SLAB_CONSISTENCY_CHECKS ((slab_flags_t __force)0x00000100U)
Alexey Dobriyand50112e2017-11-15 17:32:18 -080028/* DEBUG: Red zone objs in a cache */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080029#define SLAB_RED_ZONE ((slab_flags_t __force)0x00000400U)
Alexey Dobriyand50112e2017-11-15 17:32:18 -080030/* DEBUG: Poison objects */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080031#define SLAB_POISON ((slab_flags_t __force)0x00000800U)
Alexey Dobriyand50112e2017-11-15 17:32:18 -080032/* Align objs on cache lines */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080033#define SLAB_HWCACHE_ALIGN ((slab_flags_t __force)0x00002000U)
Alexey Dobriyand50112e2017-11-15 17:32:18 -080034/* Use GFP_DMA memory */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080035#define SLAB_CACHE_DMA ((slab_flags_t __force)0x00004000U)
Nicolas Boichat6d6ea1e2019-03-28 20:43:42 -070036/* Use GFP_DMA32 memory */
37#define SLAB_CACHE_DMA32 ((slab_flags_t __force)0x00008000U)
Alexey Dobriyand50112e2017-11-15 17:32:18 -080038/* DEBUG: Store the last owner for bug hunting */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080039#define SLAB_STORE_USER ((slab_flags_t __force)0x00010000U)
Alexey Dobriyand50112e2017-11-15 17:32:18 -080040/* Panic if kmem_cache_create() fails */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080041#define SLAB_PANIC ((slab_flags_t __force)0x00040000U)
Peter Zijlstrad7de4c12008-11-13 20:40:12 +020042/*
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -080043 * SLAB_TYPESAFE_BY_RCU - **WARNING** READ THIS!
Peter Zijlstrad7de4c12008-11-13 20:40:12 +020044 *
45 * This delays freeing the SLAB page by a grace period, it does _NOT_
46 * delay object freeing. This means that if you do kmem_cache_free()
47 * that memory location is free to be reused at any time. Thus it may
48 * be possible to see another object there in the same RCU grace period.
49 *
50 * This feature only ensures the memory location backing the object
51 * stays valid, the trick to using this is relying on an independent
52 * object validation pass. Something like:
53 *
54 * rcu_read_lock()
55 * again:
56 * obj = lockless_lookup(key);
57 * if (obj) {
58 * if (!try_get_ref(obj)) // might fail for free objects
59 * goto again;
60 *
61 * if (obj->key != key) { // not the object we expected
62 * put_ref(obj);
63 * goto again;
64 * }
65 * }
66 * rcu_read_unlock();
67 *
Joonsoo Kim68126702013-10-24 10:07:42 +090068 * This is useful if we need to approach a kernel structure obliquely,
69 * from its address obtained without the usual locking. We can lock
70 * the structure to stabilize it and check it's still at the given address,
71 * only if we can be sure that the memory has not been meanwhile reused
72 * for some other kind of object (which our subsystem's lock might corrupt).
73 *
74 * rcu_read_lock before reading the address, then rcu_read_unlock after
75 * taking the spinlock within the structure expected at that address.
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -080076 *
77 * Note that SLAB_TYPESAFE_BY_RCU was originally named SLAB_DESTROY_BY_RCU.
Peter Zijlstrad7de4c12008-11-13 20:40:12 +020078 */
Alexey Dobriyand50112e2017-11-15 17:32:18 -080079/* Defer freeing slabs to RCU */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080080#define SLAB_TYPESAFE_BY_RCU ((slab_flags_t __force)0x00080000U)
Alexey Dobriyand50112e2017-11-15 17:32:18 -080081/* Spread some memory over cpuset */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080082#define SLAB_MEM_SPREAD ((slab_flags_t __force)0x00100000U)
Alexey Dobriyand50112e2017-11-15 17:32:18 -080083/* Trace allocations and frees */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080084#define SLAB_TRACE ((slab_flags_t __force)0x00200000U)
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
Thomas Gleixner30327ac2008-04-30 00:54:59 -070086/* Flag to prevent checks on free */
87#ifdef CONFIG_DEBUG_OBJECTS
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080088# define SLAB_DEBUG_OBJECTS ((slab_flags_t __force)0x00400000U)
Thomas Gleixner30327ac2008-04-30 00:54:59 -070089#else
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080090# define SLAB_DEBUG_OBJECTS 0
Thomas Gleixner30327ac2008-04-30 00:54:59 -070091#endif
92
Alexey Dobriyand50112e2017-11-15 17:32:18 -080093/* Avoid kmemleak tracing */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080094#define SLAB_NOLEAKTRACE ((slab_flags_t __force)0x00800000U)
Catalin Marinasd5cff632009-06-11 13:22:40 +010095
Alexey Dobriyand50112e2017-11-15 17:32:18 -080096/* Fault injection mark */
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +030097#ifdef CONFIG_FAILSLAB
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -080098# define SLAB_FAILSLAB ((slab_flags_t __force)0x02000000U)
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +030099#else
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -0800100# define SLAB_FAILSLAB 0
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +0300101#endif
Alexey Dobriyand50112e2017-11-15 17:32:18 -0800102/* Account to memcg */
Kirill Tkhai84c07d12018-08-17 15:47:25 -0700103#ifdef CONFIG_MEMCG_KMEM
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -0800104# define SLAB_ACCOUNT ((slab_flags_t __force)0x04000000U)
Vladimir Davydov230e9fc2016-01-14 15:18:15 -0800105#else
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -0800106# define SLAB_ACCOUNT 0
Vladimir Davydov230e9fc2016-01-14 15:18:15 -0800107#endif
Vegard Nossum2dff4402008-05-31 15:56:17 +0200108
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -0700109#ifdef CONFIG_KASAN
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -0800110#define SLAB_KASAN ((slab_flags_t __force)0x08000000U)
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -0700111#else
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -0800112#define SLAB_KASAN 0
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -0700113#endif
114
Mel Gormane12ba742007-10-16 01:25:52 -0700115/* The following flags affect the page allocator grouping pages by mobility */
Alexey Dobriyand50112e2017-11-15 17:32:18 -0800116/* Objects are reclaimable */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -0800117#define SLAB_RECLAIM_ACCOUNT ((slab_flags_t __force)0x00020000U)
Mel Gormane12ba742007-10-16 01:25:52 -0700118#define SLAB_TEMPORARY SLAB_RECLAIM_ACCOUNT /* Objects are short-lived */
Waiman Longfcf8a1e2019-07-11 20:56:38 -0700119
120/* Slab deactivation flag */
121#define SLAB_DEACTIVATED ((slab_flags_t __force)0x10000000U)
122
Christoph Lameter2e892f42006-12-13 00:34:23 -0800123/*
Christoph Lameter6cb8f912007-07-17 04:03:22 -0700124 * ZERO_SIZE_PTR will be returned for zero sized kmalloc requests.
125 *
126 * Dereferencing ZERO_SIZE_PTR will lead to a distinct access fault.
127 *
128 * ZERO_SIZE_PTR can be passed to kfree though in the same way that NULL can.
129 * Both make kfree a no-op.
130 */
131#define ZERO_SIZE_PTR ((void *)16)
132
Roland Dreier1d4ec7b2007-07-20 12:13:20 -0700133#define ZERO_OR_NULL_PTR(x) ((unsigned long)(x) <= \
Christoph Lameter6cb8f912007-07-17 04:03:22 -0700134 (unsigned long)ZERO_SIZE_PTR)
135
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800136#include <linux/kasan.h>
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500137
Glauber Costa2633d7a2012-12-18 14:22:34 -0800138struct mem_cgroup;
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500139/*
Christoph Lameter2e892f42006-12-13 00:34:23 -0800140 * struct kmem_cache related prototypes
141 */
142void __init kmem_cache_init(void);
Denis Kirjanovfda90122015-11-05 18:44:59 -0800143bool slab_is_available(void);
Matt Mackall10cef602006-01-08 01:01:45 -0800144
Kees Cook2d891fb2017-11-30 13:04:32 -0800145extern bool usercopy_fallback;
146
Alexey Dobriyanf4957d52018-04-05 16:20:37 -0700147struct kmem_cache *kmem_cache_create(const char *name, unsigned int size,
148 unsigned int align, slab_flags_t flags,
David Windsor8eb82842017-06-10 22:50:28 -0400149 void (*ctor)(void *));
150struct kmem_cache *kmem_cache_create_usercopy(const char *name,
Alexey Dobriyanf4957d52018-04-05 16:20:37 -0700151 unsigned int size, unsigned int align,
152 slab_flags_t flags,
Alexey Dobriyan7bbdb812018-04-05 16:21:31 -0700153 unsigned int useroffset, unsigned int usersize,
David Windsor8eb82842017-06-10 22:50:28 -0400154 void (*ctor)(void *));
Christoph Lameter2e892f42006-12-13 00:34:23 -0800155void kmem_cache_destroy(struct kmem_cache *);
156int kmem_cache_shrink(struct kmem_cache *);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -0800157
Christoph Lameter0a31bd52007-05-06 14:49:57 -0700158/*
159 * Please use this macro to create slab caches. Simply specify the
160 * name of the structure and maybe some flags that are listed above.
161 *
162 * The alignment of the struct determines object alignment. If you
163 * f.e. add ____cacheline_aligned_in_smp to the struct declaration
164 * then the objects will be properly aligned in SMP configurations.
165 */
David Windsor8eb82842017-06-10 22:50:28 -0400166#define KMEM_CACHE(__struct, __flags) \
167 kmem_cache_create(#__struct, sizeof(struct __struct), \
168 __alignof__(struct __struct), (__flags), NULL)
169
170/*
171 * To whitelist a single field for copying to/from usercopy, use this
172 * macro instead for KMEM_CACHE() above.
173 */
174#define KMEM_CACHE_USERCOPY(__struct, __flags, __field) \
175 kmem_cache_create_usercopy(#__struct, \
176 sizeof(struct __struct), \
177 __alignof__(struct __struct), (__flags), \
178 offsetof(struct __struct, __field), \
179 sizeof_field(struct __struct, __field), NULL)
Christoph Lameter0a31bd52007-05-06 14:49:57 -0700180
Christoph Lameter2e892f42006-12-13 00:34:23 -0800181/*
Christoph Lameter34504662013-01-10 19:00:53 +0000182 * Common kmalloc functions provided by all allocators
183 */
Christoph Lameter34504662013-01-10 19:00:53 +0000184void * __must_check krealloc(const void *, size_t, gfp_t);
185void kfree(const void *);
Waiman Long453431a2020-08-06 23:18:13 -0700186void kfree_sensitive(const void *);
Marco Elver10d1f8c2019-07-11 20:54:14 -0700187size_t __ksize(const void *);
Christoph Lameter34504662013-01-10 19:00:53 +0000188size_t ksize(const void *);
189
Kees Cookf5509cc2016-06-07 11:05:33 -0700190#ifdef CONFIG_HAVE_HARDENED_USERCOPY_ALLOCATOR
Kees Cookf4e6e282018-01-10 14:48:22 -0800191void __check_heap_object(const void *ptr, unsigned long n, struct page *page,
192 bool to_user);
Kees Cookf5509cc2016-06-07 11:05:33 -0700193#else
Kees Cookf4e6e282018-01-10 14:48:22 -0800194static inline void __check_heap_object(const void *ptr, unsigned long n,
195 struct page *page, bool to_user) { }
Kees Cookf5509cc2016-06-07 11:05:33 -0700196#endif
197
Christoph Lameterc601fd62013-02-05 16:36:47 +0000198/*
199 * Some archs want to perform DMA into kmalloc caches and need a guaranteed
200 * alignment larger than the alignment of a 64-bit integer.
201 * Setting ARCH_KMALLOC_MINALIGN in arch headers allows that.
202 */
203#if defined(ARCH_DMA_MINALIGN) && ARCH_DMA_MINALIGN > 8
204#define ARCH_KMALLOC_MINALIGN ARCH_DMA_MINALIGN
205#define KMALLOC_MIN_SIZE ARCH_DMA_MINALIGN
206#define KMALLOC_SHIFT_LOW ilog2(ARCH_DMA_MINALIGN)
207#else
208#define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
209#endif
210
Christoph Lameter34504662013-01-10 19:00:53 +0000211/*
Rasmus Villemoes94a58c32015-11-20 15:56:48 -0800212 * Setting ARCH_SLAB_MINALIGN in arch headers allows a different alignment.
213 * Intended for arches that get misalignment faults even for 64 bit integer
214 * aligned buffers.
215 */
216#ifndef ARCH_SLAB_MINALIGN
217#define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
218#endif
219
220/*
221 * kmalloc and friends return ARCH_KMALLOC_MINALIGN aligned
222 * pointers. kmem_cache_alloc and friends return ARCH_SLAB_MINALIGN
223 * aligned pointers.
224 */
225#define __assume_kmalloc_alignment __assume_aligned(ARCH_KMALLOC_MINALIGN)
226#define __assume_slab_alignment __assume_aligned(ARCH_SLAB_MINALIGN)
227#define __assume_page_alignment __assume_aligned(PAGE_SIZE)
228
229/*
Christoph Lameter95a05b42013-01-10 19:14:19 +0000230 * Kmalloc array related definitions
231 */
232
233#ifdef CONFIG_SLAB
234/*
235 * The largest kmalloc size supported by the SLAB allocators is
Christoph Lameter0aa817f2007-05-16 22:11:01 -0700236 * 32 megabyte (2^25) or the maximum allocatable page order if that is
237 * less than 32 MB.
238 *
239 * WARNING: Its not easy to increase this value since the allocators have
240 * to do various tricks to work around compiler limitations in order to
241 * ensure proper constant folding.
242 */
Christoph Lameterdebee072007-06-23 17:16:43 -0700243#define KMALLOC_SHIFT_HIGH ((MAX_ORDER + PAGE_SHIFT - 1) <= 25 ? \
244 (MAX_ORDER + PAGE_SHIFT - 1) : 25)
Christoph Lameter95a05b42013-01-10 19:14:19 +0000245#define KMALLOC_SHIFT_MAX KMALLOC_SHIFT_HIGH
Christoph Lameterc601fd62013-02-05 16:36:47 +0000246#ifndef KMALLOC_SHIFT_LOW
Christoph Lameter95a05b42013-01-10 19:14:19 +0000247#define KMALLOC_SHIFT_LOW 5
Christoph Lameterc601fd62013-02-05 16:36:47 +0000248#endif
Christoph Lameter069e2b352013-06-14 19:55:13 +0000249#endif
250
251#ifdef CONFIG_SLUB
Christoph Lameter95a05b42013-01-10 19:14:19 +0000252/*
Dave Hansen433a91f2014-01-28 14:24:50 -0800253 * SLUB directly allocates requests fitting in to an order-1 page
254 * (PAGE_SIZE*2). Larger requests are passed to the page allocator.
Christoph Lameter95a05b42013-01-10 19:14:19 +0000255 */
256#define KMALLOC_SHIFT_HIGH (PAGE_SHIFT + 1)
Michal Hockobb1107f2017-01-10 16:57:27 -0800257#define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT - 1)
Christoph Lameterc601fd62013-02-05 16:36:47 +0000258#ifndef KMALLOC_SHIFT_LOW
Christoph Lameter95a05b42013-01-10 19:14:19 +0000259#define KMALLOC_SHIFT_LOW 3
260#endif
Christoph Lameterc601fd62013-02-05 16:36:47 +0000261#endif
Christoph Lameter0aa817f2007-05-16 22:11:01 -0700262
Christoph Lameter069e2b352013-06-14 19:55:13 +0000263#ifdef CONFIG_SLOB
264/*
Dave Hansen433a91f2014-01-28 14:24:50 -0800265 * SLOB passes all requests larger than one page to the page allocator.
Christoph Lameter069e2b352013-06-14 19:55:13 +0000266 * No kmalloc array is necessary since objects of different sizes can
267 * be allocated from the same page.
268 */
Christoph Lameter069e2b352013-06-14 19:55:13 +0000269#define KMALLOC_SHIFT_HIGH PAGE_SHIFT
Michal Hockobb1107f2017-01-10 16:57:27 -0800270#define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT - 1)
Christoph Lameter069e2b352013-06-14 19:55:13 +0000271#ifndef KMALLOC_SHIFT_LOW
272#define KMALLOC_SHIFT_LOW 3
273#endif
274#endif
275
Christoph Lameter95a05b42013-01-10 19:14:19 +0000276/* Maximum allocatable size */
277#define KMALLOC_MAX_SIZE (1UL << KMALLOC_SHIFT_MAX)
278/* Maximum size for which we actually use a slab cache */
279#define KMALLOC_MAX_CACHE_SIZE (1UL << KMALLOC_SHIFT_HIGH)
tangjianqiangd7cff4d2020-10-13 16:48:37 -0700280/* Maximum order allocatable via the slab allocator */
Christoph Lameter95a05b42013-01-10 19:14:19 +0000281#define KMALLOC_MAX_ORDER (KMALLOC_SHIFT_MAX - PAGE_SHIFT)
Christoph Lameter0aa817f2007-05-16 22:11:01 -0700282
Christoph Lameter90810642011-06-23 09:36:12 -0500283/*
Christoph Lameterce6a5022013-01-10 19:14:19 +0000284 * Kmalloc subsystem.
285 */
Christoph Lameterc601fd62013-02-05 16:36:47 +0000286#ifndef KMALLOC_MIN_SIZE
Christoph Lameter95a05b42013-01-10 19:14:19 +0000287#define KMALLOC_MIN_SIZE (1 << KMALLOC_SHIFT_LOW)
Christoph Lameterce6a5022013-01-10 19:14:19 +0000288#endif
Christoph Lameterce6a5022013-01-10 19:14:19 +0000289
Joonsoo Kim24f870d2014-03-12 17:06:19 +0900290/*
291 * This restriction comes from byte sized index implementation.
292 * Page size is normally 2^12 bytes and, in this case, if we want to use
293 * byte sized index which can represent 2^8 entries, the size of the object
294 * should be equal or greater to 2^12 / 2^8 = 2^4 = 16.
295 * If minimum size of kmalloc is less than 16, we use it as minimum object
296 * size and give up to use byte sized index.
297 */
298#define SLAB_OBJ_MIN_SIZE (KMALLOC_MIN_SIZE < 16 ? \
299 (KMALLOC_MIN_SIZE) : 16)
300
Vlastimil Babka12915232018-10-26 15:05:38 -0700301/*
302 * Whenever changing this, take care of that kmalloc_type() and
303 * create_kmalloc_caches() still work as intended.
304 */
Vlastimil Babkacc252ea2018-10-26 15:05:34 -0700305enum kmalloc_cache_type {
306 KMALLOC_NORMAL = 0,
Vlastimil Babka12915232018-10-26 15:05:38 -0700307 KMALLOC_RECLAIM,
Christoph Lameter9425c582013-01-10 19:12:17 +0000308#ifdef CONFIG_ZONE_DMA
Vlastimil Babkacc252ea2018-10-26 15:05:34 -0700309 KMALLOC_DMA,
Christoph Lameter9425c582013-01-10 19:12:17 +0000310#endif
Vlastimil Babkacc252ea2018-10-26 15:05:34 -0700311 NR_KMALLOC_TYPES
312};
313
314#ifndef CONFIG_SLOB
315extern struct kmem_cache *
316kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1];
317
318static __always_inline enum kmalloc_cache_type kmalloc_type(gfp_t flags)
319{
Vlastimil Babkacc252ea2018-10-26 15:05:34 -0700320#ifdef CONFIG_ZONE_DMA
Vlastimil Babka4e45f712018-12-28 00:33:17 -0800321 /*
322 * The most common case is KMALLOC_NORMAL, so test for it
323 * with a single branch for both flags.
324 */
325 if (likely((flags & (__GFP_DMA | __GFP_RECLAIMABLE)) == 0))
326 return KMALLOC_NORMAL;
Vlastimil Babka12915232018-10-26 15:05:38 -0700327
328 /*
Vlastimil Babka4e45f712018-12-28 00:33:17 -0800329 * At least one of the flags has to be set. If both are, __GFP_DMA
330 * is more important.
Vlastimil Babka12915232018-10-26 15:05:38 -0700331 */
Vlastimil Babka4e45f712018-12-28 00:33:17 -0800332 return flags & __GFP_DMA ? KMALLOC_DMA : KMALLOC_RECLAIM;
333#else
334 return flags & __GFP_RECLAIMABLE ? KMALLOC_RECLAIM : KMALLOC_NORMAL;
335#endif
Vlastimil Babkacc252ea2018-10-26 15:05:34 -0700336}
Christoph Lameter9425c582013-01-10 19:12:17 +0000337
Christoph Lameterce6a5022013-01-10 19:14:19 +0000338/*
339 * Figure out which kmalloc slab an allocation of a certain size
340 * belongs to.
341 * 0 = zero alloc
342 * 1 = 65 .. 96 bytes
Rasmus Villemoes1ed58b62015-06-24 16:55:59 -0700343 * 2 = 129 .. 192 bytes
344 * n = 2^(n-1)+1 .. 2^n
Christoph Lameterce6a5022013-01-10 19:14:19 +0000345 */
Alexey Dobriyan36071a22018-04-05 16:20:22 -0700346static __always_inline unsigned int kmalloc_index(size_t size)
Christoph Lameterce6a5022013-01-10 19:14:19 +0000347{
348 if (!size)
349 return 0;
350
351 if (size <= KMALLOC_MIN_SIZE)
352 return KMALLOC_SHIFT_LOW;
353
354 if (KMALLOC_MIN_SIZE <= 32 && size > 64 && size <= 96)
355 return 1;
356 if (KMALLOC_MIN_SIZE <= 64 && size > 128 && size <= 192)
357 return 2;
358 if (size <= 8) return 3;
359 if (size <= 16) return 4;
360 if (size <= 32) return 5;
361 if (size <= 64) return 6;
362 if (size <= 128) return 7;
363 if (size <= 256) return 8;
364 if (size <= 512) return 9;
365 if (size <= 1024) return 10;
366 if (size <= 2 * 1024) return 11;
367 if (size <= 4 * 1024) return 12;
368 if (size <= 8 * 1024) return 13;
369 if (size <= 16 * 1024) return 14;
370 if (size <= 32 * 1024) return 15;
371 if (size <= 64 * 1024) return 16;
372 if (size <= 128 * 1024) return 17;
373 if (size <= 256 * 1024) return 18;
374 if (size <= 512 * 1024) return 19;
375 if (size <= 1024 * 1024) return 20;
376 if (size <= 2 * 1024 * 1024) return 21;
377 if (size <= 4 * 1024 * 1024) return 22;
378 if (size <= 8 * 1024 * 1024) return 23;
379 if (size <= 16 * 1024 * 1024) return 24;
380 if (size <= 32 * 1024 * 1024) return 25;
381 if (size <= 64 * 1024 * 1024) return 26;
382 BUG();
383
384 /* Will never be reached. Needed because the compiler may complain */
385 return -1;
386}
Christoph Lameter069e2b352013-06-14 19:55:13 +0000387#endif /* !CONFIG_SLOB */
Christoph Lameterce6a5022013-01-10 19:14:19 +0000388
Rasmus Villemoes48a270552016-05-19 17:10:55 -0700389void *__kmalloc(size_t size, gfp_t flags) __assume_kmalloc_alignment __malloc;
390void *kmem_cache_alloc(struct kmem_cache *, gfp_t flags) __assume_slab_alignment __malloc;
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -0800391void kmem_cache_free(struct kmem_cache *, void *);
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000392
Christoph Lameter484748f2015-09-04 15:45:34 -0700393/*
Jesper Dangaard Brouer9f706d62016-03-15 14:54:03 -0700394 * Bulk allocation and freeing operations. These are accelerated in an
Christoph Lameter484748f2015-09-04 15:45:34 -0700395 * allocator specific way to avoid taking locks repeatedly or building
396 * metadata structures unnecessarily.
397 *
398 * Note that interrupts must be enabled when calling these functions.
399 */
400void kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
Jesper Dangaard Brouer865762a2015-11-20 15:57:58 -0800401int kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
Christoph Lameter484748f2015-09-04 15:45:34 -0700402
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -0700403/*
404 * Caller must not use kfree_bulk() on memory not originally allocated
405 * by kmalloc(), because the SLOB allocator cannot handle this.
406 */
407static __always_inline void kfree_bulk(size_t size, void **p)
408{
409 kmem_cache_free_bulk(NULL, size, p);
410}
411
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000412#ifdef CONFIG_NUMA
Rasmus Villemoes48a270552016-05-19 17:10:55 -0700413void *__kmalloc_node(size_t size, gfp_t flags, int node) __assume_kmalloc_alignment __malloc;
414void *kmem_cache_alloc_node(struct kmem_cache *, gfp_t flags, int node) __assume_slab_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000415#else
416static __always_inline void *__kmalloc_node(size_t size, gfp_t flags, int node)
417{
418 return __kmalloc(size, flags);
419}
420
421static __always_inline void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t flags, int node)
422{
423 return kmem_cache_alloc(s, flags);
424}
425#endif
426
427#ifdef CONFIG_TRACING
Rasmus Villemoes48a270552016-05-19 17:10:55 -0700428extern void *kmem_cache_alloc_trace(struct kmem_cache *, gfp_t, size_t) __assume_slab_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000429
430#ifdef CONFIG_NUMA
431extern void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
432 gfp_t gfpflags,
Rasmus Villemoes48a270552016-05-19 17:10:55 -0700433 int node, size_t size) __assume_slab_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000434#else
435static __always_inline void *
436kmem_cache_alloc_node_trace(struct kmem_cache *s,
437 gfp_t gfpflags,
438 int node, size_t size)
439{
440 return kmem_cache_alloc_trace(s, gfpflags, size);
441}
442#endif /* CONFIG_NUMA */
443
444#else /* CONFIG_TRACING */
445static __always_inline void *kmem_cache_alloc_trace(struct kmem_cache *s,
446 gfp_t flags, size_t size)
447{
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800448 void *ret = kmem_cache_alloc(s, flags);
449
Andrey Konovalov01165232018-12-28 00:29:37 -0800450 ret = kasan_kmalloc(s, ret, size, flags);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800451 return ret;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000452}
453
454static __always_inline void *
455kmem_cache_alloc_node_trace(struct kmem_cache *s,
456 gfp_t gfpflags,
457 int node, size_t size)
458{
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800459 void *ret = kmem_cache_alloc_node(s, gfpflags, node);
460
Andrey Konovalov01165232018-12-28 00:29:37 -0800461 ret = kasan_kmalloc(s, ret, size, gfpflags);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800462 return ret;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000463}
464#endif /* CONFIG_TRACING */
465
Rasmus Villemoes48a270552016-05-19 17:10:55 -0700466extern void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) __assume_page_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000467
468#ifdef CONFIG_TRACING
Rasmus Villemoes48a270552016-05-19 17:10:55 -0700469extern void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) __assume_page_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000470#else
471static __always_inline void *
472kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
473{
474 return kmalloc_order(size, flags, order);
475}
Christoph Lameterce6a5022013-01-10 19:14:19 +0000476#endif
477
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000478static __always_inline void *kmalloc_large(size_t size, gfp_t flags)
479{
480 unsigned int order = get_order(size);
481 return kmalloc_order_trace(size, flags, order);
482}
483
484/**
485 * kmalloc - allocate memory
486 * @size: how many bytes of memory are required.
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800487 * @flags: the type of memory to allocate.
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000488 *
489 * kmalloc is the normal method of allocating memory
490 * for objects smaller than page size in the kernel.
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800491 *
Vlastimil Babka59bb4792019-10-06 17:58:45 -0700492 * The allocated object address is aligned to at least ARCH_KMALLOC_MINALIGN
493 * bytes. For @size of power of two bytes, the alignment is also guaranteed
494 * to be at least to the size.
495 *
Mike Rapoport01598ba2018-11-11 18:48:44 +0200496 * The @flags argument may be one of the GFP flags defined at
497 * include/linux/gfp.h and described at
498 * :ref:`Documentation/core-api/mm-api.rst <mm-api-gfp-flags>`
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800499 *
Mike Rapoport01598ba2018-11-11 18:48:44 +0200500 * The recommended usage of the @flags is described at
Mauro Carvalho Chehab2370ae42020-04-10 14:32:25 -0700501 * :ref:`Documentation/core-api/memory-allocation.rst <memory_allocation>`
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800502 *
Mike Rapoport01598ba2018-11-11 18:48:44 +0200503 * Below is a brief outline of the most useful GFP flags
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800504 *
Mike Rapoport01598ba2018-11-11 18:48:44 +0200505 * %GFP_KERNEL
506 * Allocate normal kernel ram. May sleep.
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800507 *
Mike Rapoport01598ba2018-11-11 18:48:44 +0200508 * %GFP_NOWAIT
509 * Allocation will not sleep.
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800510 *
Mike Rapoport01598ba2018-11-11 18:48:44 +0200511 * %GFP_ATOMIC
512 * Allocation will not sleep. May use emergency pools.
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800513 *
Mike Rapoport01598ba2018-11-11 18:48:44 +0200514 * %GFP_HIGHUSER
515 * Allocate memory from high memory on behalf of user.
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800516 *
517 * Also it is possible to set different flags by OR'ing
518 * in one or more of the following additional @flags:
519 *
Mike Rapoport01598ba2018-11-11 18:48:44 +0200520 * %__GFP_HIGH
521 * This allocation has high priority and may use emergency pools.
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800522 *
Mike Rapoport01598ba2018-11-11 18:48:44 +0200523 * %__GFP_NOFAIL
524 * Indicate that this allocation is in no way allowed to fail
525 * (think twice before using).
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800526 *
Mike Rapoport01598ba2018-11-11 18:48:44 +0200527 * %__GFP_NORETRY
528 * If memory is not immediately available,
529 * then give up at once.
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800530 *
Mike Rapoport01598ba2018-11-11 18:48:44 +0200531 * %__GFP_NOWARN
532 * If allocation fails, don't issue any warnings.
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800533 *
Mike Rapoport01598ba2018-11-11 18:48:44 +0200534 * %__GFP_RETRY_MAYFAIL
535 * Try really hard to succeed the allocation but fail
536 * eventually.
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000537 */
538static __always_inline void *kmalloc(size_t size, gfp_t flags)
539{
540 if (__builtin_constant_p(size)) {
Vlastimil Babkacc252ea2018-10-26 15:05:34 -0700541#ifndef CONFIG_SLOB
542 unsigned int index;
543#endif
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000544 if (size > KMALLOC_MAX_CACHE_SIZE)
545 return kmalloc_large(size, flags);
546#ifndef CONFIG_SLOB
Vlastimil Babkacc252ea2018-10-26 15:05:34 -0700547 index = kmalloc_index(size);
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000548
Vlastimil Babkacc252ea2018-10-26 15:05:34 -0700549 if (!index)
550 return ZERO_SIZE_PTR;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000551
Vlastimil Babkacc252ea2018-10-26 15:05:34 -0700552 return kmem_cache_alloc_trace(
553 kmalloc_caches[kmalloc_type(flags)][index],
554 flags, size);
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000555#endif
556 }
557 return __kmalloc(size, flags);
558}
559
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000560static __always_inline void *kmalloc_node(size_t size, gfp_t flags, int node)
561{
562#ifndef CONFIG_SLOB
563 if (__builtin_constant_p(size) &&
Vlastimil Babkacc252ea2018-10-26 15:05:34 -0700564 size <= KMALLOC_MAX_CACHE_SIZE) {
Alexey Dobriyan36071a22018-04-05 16:20:22 -0700565 unsigned int i = kmalloc_index(size);
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000566
567 if (!i)
568 return ZERO_SIZE_PTR;
569
Vlastimil Babkacc252ea2018-10-26 15:05:34 -0700570 return kmem_cache_alloc_node_trace(
571 kmalloc_caches[kmalloc_type(flags)][i],
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000572 flags, node, size);
573 }
574#endif
575 return __kmalloc_node(size, flags, node);
576}
577
Christoph Lameter2e892f42006-12-13 00:34:23 -0800578/**
Michael Opdenackere7efa612013-06-25 18:16:55 +0200579 * kmalloc_array - allocate memory for an array.
580 * @n: number of elements.
581 * @size: element size.
582 * @flags: the type of memory to allocate (see kmalloc).
Paul Drynoff800590f2006-06-23 02:03:48 -0700583 */
Xi Wanga8203722012-03-05 15:14:41 -0800584static inline void *kmalloc_array(size_t n, size_t size, gfp_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585{
Kees Cook49b7f892018-05-08 12:52:32 -0700586 size_t bytes;
587
588 if (unlikely(check_mul_overflow(n, size, &bytes)))
Paul Mundt6193a2f2007-07-15 23:38:22 -0700589 return NULL;
Alexey Dobriyan91c6a052016-07-26 15:22:08 -0700590 if (__builtin_constant_p(n) && __builtin_constant_p(size))
Kees Cook49b7f892018-05-08 12:52:32 -0700591 return kmalloc(bytes, flags);
592 return __kmalloc(bytes, flags);
Xi Wanga8203722012-03-05 15:14:41 -0800593}
594
595/**
596 * kcalloc - allocate memory for an array. The memory is set to zero.
597 * @n: number of elements.
598 * @size: element size.
599 * @flags: the type of memory to allocate (see kmalloc).
600 */
601static inline void *kcalloc(size_t n, size_t size, gfp_t flags)
602{
603 return kmalloc_array(n, size, flags | __GFP_ZERO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604}
605
Christoph Hellwig1d2c8ee2006-10-04 02:15:25 -0700606/*
607 * kmalloc_track_caller is a special version of kmalloc that records the
608 * calling function of the routine calling it for slab leak tracking instead
609 * of just the calling function (confusing, eh?).
610 * It's useful when the call to kmalloc comes from a widely-used standard
611 * allocator where we care about the real place the memory allocation
612 * request comes from.
613 */
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300614extern void *__kmalloc_track_caller(size_t, gfp_t, unsigned long);
Christoph Hellwig1d2c8ee2006-10-04 02:15:25 -0700615#define kmalloc_track_caller(size, flags) \
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300616 __kmalloc_track_caller(size, flags, _RET_IP_)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617
Johannes Thumshirn5799b252017-11-15 17:32:29 -0800618static inline void *kmalloc_array_node(size_t n, size_t size, gfp_t flags,
619 int node)
620{
Kees Cook49b7f892018-05-08 12:52:32 -0700621 size_t bytes;
622
623 if (unlikely(check_mul_overflow(n, size, &bytes)))
Johannes Thumshirn5799b252017-11-15 17:32:29 -0800624 return NULL;
625 if (__builtin_constant_p(n) && __builtin_constant_p(size))
Kees Cook49b7f892018-05-08 12:52:32 -0700626 return kmalloc_node(bytes, flags, node);
627 return __kmalloc_node(bytes, flags, node);
Johannes Thumshirn5799b252017-11-15 17:32:29 -0800628}
629
630static inline void *kcalloc_node(size_t n, size_t size, gfp_t flags, int node)
631{
632 return kmalloc_array_node(n, size, flags | __GFP_ZERO, node);
633}
634
635
Manfred Spraul97e2bde2005-05-01 08:58:38 -0700636#ifdef CONFIG_NUMA
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300637extern void *__kmalloc_node_track_caller(size_t, gfp_t, int, unsigned long);
Christoph Hellwig8b98c162006-12-06 20:32:30 -0800638#define kmalloc_node_track_caller(size, flags, node) \
639 __kmalloc_node_track_caller(size, flags, node, \
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300640 _RET_IP_)
Christoph Lameter2e892f42006-12-13 00:34:23 -0800641
Christoph Hellwig8b98c162006-12-06 20:32:30 -0800642#else /* CONFIG_NUMA */
Christoph Lameter2e892f42006-12-13 00:34:23 -0800643
644#define kmalloc_node_track_caller(size, flags, node) \
645 kmalloc_track_caller(size, flags)
646
Pascal Terjandfcd3612008-11-25 15:08:19 +0100647#endif /* CONFIG_NUMA */
Christoph Hellwig8b98c162006-12-06 20:32:30 -0800648
Christoph Lameter81cda662007-07-17 04:03:29 -0700649/*
650 * Shortcuts
651 */
652static inline void *kmem_cache_zalloc(struct kmem_cache *k, gfp_t flags)
653{
654 return kmem_cache_alloc(k, flags | __GFP_ZERO);
655}
656
657/**
658 * kzalloc - allocate memory. The memory is set to zero.
659 * @size: how many bytes of memory are required.
660 * @flags: the type of memory to allocate (see kmalloc).
661 */
662static inline void *kzalloc(size_t size, gfp_t flags)
663{
664 return kmalloc(size, flags | __GFP_ZERO);
665}
666
Jeff Layton979b0fe2008-06-05 22:47:00 -0700667/**
668 * kzalloc_node - allocate zeroed memory from a particular memory node.
669 * @size: how many bytes of memory are required.
670 * @flags: the type of memory to allocate (see kmalloc).
671 * @node: memory node from which to allocate
672 */
673static inline void *kzalloc_node(size_t size, gfp_t flags, int node)
674{
675 return kmalloc_node(size, flags | __GFP_ZERO, node);
676}
677
Joonsoo Kim07f361b2014-10-09 15:26:00 -0700678unsigned int kmem_cache_size(struct kmem_cache *s);
Pekka Enberg7e85ee02009-06-12 14:03:06 +0300679void __init kmem_cache_init_late(void);
680
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +0200681#if defined(CONFIG_SMP) && defined(CONFIG_SLAB)
682int slab_prepare_cpu(unsigned int cpu);
683int slab_dead_cpu(unsigned int cpu);
684#else
685#define slab_prepare_cpu NULL
686#define slab_dead_cpu NULL
687#endif
688
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689#endif /* _LINUX_SLAB_H */