linux/include/linux/slab.h
<<
>>
Prefs
   1/* SPDX-License-Identifier: GPL-2.0 */
   2/*
   3 * Written by Mark Hemment, 1996 (markhe@nextd.demon.co.uk).
   4 *
   5 * (C) SGI 2006, Christoph Lameter
   6 *      Cleaned up and restructured to ease the addition of alternative
   7 *      implementations of SLAB allocators.
   8 * (C) Linux Foundation 2008-2013
   9 *      Unified interface for all slab allocators
  10 */
  11
  12#ifndef _LINUX_SLAB_H
  13#define _LINUX_SLAB_H
  14
  15#include <linux/gfp.h>
  16#include <linux/overflow.h>
  17#include <linux/types.h>
  18#include <linux/workqueue.h>
  19
  20
  21/*
  22 * Flags to pass to kmem_cache_create().
  23 * The ones marked DEBUG are only valid if CONFIG_DEBUG_SLAB is set.
  24 */
  25/* DEBUG: Perform (expensive) checks on alloc/free */
  26#define SLAB_CONSISTENCY_CHECKS ((slab_flags_t __force)0x00000100U)
  27/* DEBUG: Red zone objs in a cache */
  28#define SLAB_RED_ZONE           ((slab_flags_t __force)0x00000400U)
  29/* DEBUG: Poison objects */
  30#define SLAB_POISON             ((slab_flags_t __force)0x00000800U)
  31/* Align objs on cache lines */
  32#define SLAB_HWCACHE_ALIGN      ((slab_flags_t __force)0x00002000U)
  33/* Use GFP_DMA memory */
  34#define SLAB_CACHE_DMA          ((slab_flags_t __force)0x00004000U)
  35/* Use GFP_DMA32 memory */
  36#define SLAB_CACHE_DMA32        ((slab_flags_t __force)0x00008000U)
  37/* DEBUG: Store the last owner for bug hunting */
  38#define SLAB_STORE_USER         ((slab_flags_t __force)0x00010000U)
  39/* Panic if kmem_cache_create() fails */
  40#define SLAB_PANIC              ((slab_flags_t __force)0x00040000U)
  41/*
  42 * SLAB_TYPESAFE_BY_RCU - **WARNING** READ THIS!
  43 *
  44 * This delays freeing the SLAB page by a grace period, it does _NOT_
  45 * delay object freeing. This means that if you do kmem_cache_free()
  46 * that memory location is free to be reused at any time. Thus it may
  47 * be possible to see another object there in the same RCU grace period.
  48 *
  49 * This feature only ensures the memory location backing the object
  50 * stays valid, the trick to using this is relying on an independent
  51 * object validation pass. Something like:
  52 *
  53 *  rcu_read_lock()
  54 * again:
  55 *  obj = lockless_lookup(key);
  56 *  if (obj) {
  57 *    if (!try_get_ref(obj)) // might fail for free objects
  58 *      goto again;
  59 *
  60 *    if (obj->key != key) { // not the object we expected
  61 *      put_ref(obj);
  62 *      goto again;
  63 *    }
  64 *  }
  65 *  rcu_read_unlock();
  66 *
  67 * This is useful if we need to approach a kernel structure obliquely,
  68 * from its address obtained without the usual locking. We can lock
  69 * the structure to stabilize it and check it's still at the given address,
  70 * only if we can be sure that the memory has not been meanwhile reused
  71 * for some other kind of object (which our subsystem's lock might corrupt).
  72 *
  73 * rcu_read_lock before reading the address, then rcu_read_unlock after
  74 * taking the spinlock within the structure expected at that address.
  75 *
  76 * Note that SLAB_TYPESAFE_BY_RCU was originally named SLAB_DESTROY_BY_RCU.
  77 */
  78/* Defer freeing slabs to RCU */
  79#define SLAB_TYPESAFE_BY_RCU    ((slab_flags_t __force)0x00080000U)
  80/* Spread some memory over cpuset */
  81#define SLAB_MEM_SPREAD         ((slab_flags_t __force)0x00100000U)
  82/* Trace allocations and frees */
  83#define SLAB_TRACE              ((slab_flags_t __force)0x00200000U)
  84
  85/* Flag to prevent checks on free */
  86#ifdef CONFIG_DEBUG_OBJECTS
  87# define SLAB_DEBUG_OBJECTS     ((slab_flags_t __force)0x00400000U)
  88#else
  89# define SLAB_DEBUG_OBJECTS     0
  90#endif
  91
  92/* Avoid kmemleak tracing */
  93#define SLAB_NOLEAKTRACE        ((slab_flags_t __force)0x00800000U)
  94
  95/* Fault injection mark */
  96#ifdef CONFIG_FAILSLAB
  97# define SLAB_FAILSLAB          ((slab_flags_t __force)0x02000000U)
  98#else
  99# define SLAB_FAILSLAB          0
 100#endif
 101/* Account to memcg */
 102#ifdef CONFIG_MEMCG_KMEM
 103# define SLAB_ACCOUNT           ((slab_flags_t __force)0x04000000U)
 104#else
 105# define SLAB_ACCOUNT           0
 106#endif
 107
 108#ifdef CONFIG_KASAN
 109#define SLAB_KASAN              ((slab_flags_t __force)0x08000000U)
 110#else
 111#define SLAB_KASAN              0
 112#endif
 113
 114/* The following flags affect the page allocator grouping pages by mobility */
 115/* Objects are reclaimable */
 116#define SLAB_RECLAIM_ACCOUNT    ((slab_flags_t __force)0x00020000U)
 117#define SLAB_TEMPORARY          SLAB_RECLAIM_ACCOUNT    /* Objects are short-lived */
 118/*
 119 * ZERO_SIZE_PTR will be returned for zero sized kmalloc requests.
 120 *
 121 * Dereferencing ZERO_SIZE_PTR will lead to a distinct access fault.
 122 *
 123 * ZERO_SIZE_PTR can be passed to kfree though in the same way that NULL can.
 124 * Both make kfree a no-op.
 125 */
 126#define ZERO_SIZE_PTR ((void *)16)
 127
 128#define ZERO_OR_NULL_PTR(x) ((unsigned long)(x) <= \
 129                                (unsigned long)ZERO_SIZE_PTR)
 130
 131#include <linux/kasan.h>
 132
 133struct mem_cgroup;
 134/*
 135 * struct kmem_cache related prototypes
 136 */
 137void __init kmem_cache_init(void);
 138bool slab_is_available(void);
 139
 140extern bool usercopy_fallback;
 141
 142struct kmem_cache *kmem_cache_create(const char *name, unsigned int size,
 143                        unsigned int align, slab_flags_t flags,
 144                        void (*ctor)(void *));
 145struct kmem_cache *kmem_cache_create_usercopy(const char *name,
 146                        unsigned int size, unsigned int align,
 147                        slab_flags_t flags,
 148                        unsigned int useroffset, unsigned int usersize,
 149                        void (*ctor)(void *));
 150void kmem_cache_destroy(struct kmem_cache *);
 151int kmem_cache_shrink(struct kmem_cache *);
 152
 153void memcg_create_kmem_cache(struct mem_cgroup *, struct kmem_cache *);
 154void memcg_deactivate_kmem_caches(struct mem_cgroup *);
 155void memcg_destroy_kmem_caches(struct mem_cgroup *);
 156
 157/*
 158 * Please use this macro to create slab caches. Simply specify the
 159 * name of the structure and maybe some flags that are listed above.
 160 *
 161 * The alignment of the struct determines object alignment. If you
 162 * f.e. add ____cacheline_aligned_in_smp to the struct declaration
 163 * then the objects will be properly aligned in SMP configurations.
 164 */
 165#define KMEM_CACHE(__struct, __flags)                                   \
 166                kmem_cache_create(#__struct, sizeof(struct __struct),   \
 167                        __alignof__(struct __struct), (__flags), NULL)
 168
 169/*
 170 * To whitelist a single field for copying to/from usercopy, use this
 171 * macro instead for KMEM_CACHE() above.
 172 */
 173#define KMEM_CACHE_USERCOPY(__struct, __flags, __field)                 \
 174                kmem_cache_create_usercopy(#__struct,                   \
 175                        sizeof(struct __struct),                        \
 176                        __alignof__(struct __struct), (__flags),        \
 177                        offsetof(struct __struct, __field),             \
 178                        sizeof_field(struct __struct, __field), NULL)
 179
 180/*
 181 * Common kmalloc functions provided by all allocators
 182 */
 183void * __must_check __krealloc(const void *, size_t, gfp_t);
 184void * __must_check krealloc(const void *, size_t, gfp_t);
 185void kfree(const void *);
 186void kzfree(const void *);
 187size_t ksize(const void *);
 188
 189#ifdef CONFIG_HAVE_HARDENED_USERCOPY_ALLOCATOR
 190void __check_heap_object(const void *ptr, unsigned long n, struct page *page,
 191                        bool to_user);
 192#else
 193static inline void __check_heap_object(const void *ptr, unsigned long n,
 194                                       struct page *page, bool to_user) { }
 195#endif
 196
 197/*
 198 * Some archs want to perform DMA into kmalloc caches and need a guaranteed
 199 * alignment larger than the alignment of a 64-bit integer.
 200 * Setting ARCH_KMALLOC_MINALIGN in arch headers allows that.
 201 */
 202#if defined(ARCH_DMA_MINALIGN) && ARCH_DMA_MINALIGN > 8
 203#define ARCH_KMALLOC_MINALIGN ARCH_DMA_MINALIGN
 204#define KMALLOC_MIN_SIZE ARCH_DMA_MINALIGN
 205#define KMALLOC_SHIFT_LOW ilog2(ARCH_DMA_MINALIGN)
 206#else
 207#define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
 208#endif
 209
 210/*
 211 * Setting ARCH_SLAB_MINALIGN in arch headers allows a different alignment.
 212 * Intended for arches that get misalignment faults even for 64 bit integer
 213 * aligned buffers.
 214 */
 215#ifndef ARCH_SLAB_MINALIGN
 216#define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
 217#endif
 218
 219/*
 220 * kmalloc and friends return ARCH_KMALLOC_MINALIGN aligned
 221 * pointers. kmem_cache_alloc and friends return ARCH_SLAB_MINALIGN
 222 * aligned pointers.
 223 */
 224#define __assume_kmalloc_alignment __assume_aligned(ARCH_KMALLOC_MINALIGN)
 225#define __assume_slab_alignment __assume_aligned(ARCH_SLAB_MINALIGN)
 226#define __assume_page_alignment __assume_aligned(PAGE_SIZE)
 227
 228/*
 229 * Kmalloc array related definitions
 230 */
 231
 232#ifdef CONFIG_SLAB
 233/*
 234 * The largest kmalloc size supported by the SLAB allocators is
 235 * 32 megabyte (2^25) or the maximum allocatable page order if that is
 236 * less than 32 MB.
 237 *
 238 * WARNING: Its not easy to increase this value since the allocators have
 239 * to do various tricks to work around compiler limitations in order to
 240 * ensure proper constant folding.
 241 */
 242#define KMALLOC_SHIFT_HIGH      ((MAX_ORDER + PAGE_SHIFT - 1) <= 25 ? \
 243                                (MAX_ORDER + PAGE_SHIFT - 1) : 25)
 244#define KMALLOC_SHIFT_MAX       KMALLOC_SHIFT_HIGH
 245#ifndef KMALLOC_SHIFT_LOW
 246#define KMALLOC_SHIFT_LOW       5
 247#endif
 248#endif
 249
 250#ifdef CONFIG_SLUB
 251/*
 252 * SLUB directly allocates requests fitting in to an order-1 page
 253 * (PAGE_SIZE*2).  Larger requests are passed to the page allocator.
 254 */
 255#define KMALLOC_SHIFT_HIGH      (PAGE_SHIFT + 1)
 256#define KMALLOC_SHIFT_MAX       (MAX_ORDER + PAGE_SHIFT - 1)
 257#ifndef KMALLOC_SHIFT_LOW
 258#define KMALLOC_SHIFT_LOW       3
 259#endif
 260#endif
 261
 262#ifdef CONFIG_SLOB
 263/*
 264 * SLOB passes all requests larger than one page to the page allocator.
 265 * No kmalloc array is necessary since objects of different sizes can
 266 * be allocated from the same page.
 267 */
 268#define KMALLOC_SHIFT_HIGH      PAGE_SHIFT
 269#define KMALLOC_SHIFT_MAX       (MAX_ORDER + PAGE_SHIFT - 1)
 270#ifndef KMALLOC_SHIFT_LOW
 271#define KMALLOC_SHIFT_LOW       3
 272#endif
 273#endif
 274
 275/* Maximum allocatable size */
 276#define KMALLOC_MAX_SIZE        (1UL << KMALLOC_SHIFT_MAX)
 277/* Maximum size for which we actually use a slab cache */
 278#define KMALLOC_MAX_CACHE_SIZE  (1UL << KMALLOC_SHIFT_HIGH)
 279/* Maximum order allocatable via the slab allocagtor */
 280#define KMALLOC_MAX_ORDER       (KMALLOC_SHIFT_MAX - PAGE_SHIFT)
 281
 282/*
 283 * Kmalloc subsystem.
 284 */
 285#ifndef KMALLOC_MIN_SIZE
 286#define KMALLOC_MIN_SIZE (1 << KMALLOC_SHIFT_LOW)
 287#endif
 288
 289/*
 290 * This restriction comes from byte sized index implementation.
 291 * Page size is normally 2^12 bytes and, in this case, if we want to use
 292 * byte sized index which can represent 2^8 entries, the size of the object
 293 * should be equal or greater to 2^12 / 2^8 = 2^4 = 16.
 294 * If minimum size of kmalloc is less than 16, we use it as minimum object
 295 * size and give up to use byte sized index.
 296 */
 297#define SLAB_OBJ_MIN_SIZE      (KMALLOC_MIN_SIZE < 16 ? \
 298                               (KMALLOC_MIN_SIZE) : 16)
 299
 300/*
 301 * Whenever changing this, take care of that kmalloc_type() and
 302 * create_kmalloc_caches() still work as intended.
 303 */
 304enum kmalloc_cache_type {
 305        KMALLOC_NORMAL = 0,
 306        KMALLOC_RECLAIM,
 307#ifdef CONFIG_ZONE_DMA
 308        KMALLOC_DMA,
 309#endif
 310        NR_KMALLOC_TYPES
 311};
 312
 313#ifndef CONFIG_SLOB
 314extern struct kmem_cache *
 315kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1];
 316
 317static __always_inline enum kmalloc_cache_type kmalloc_type(gfp_t flags)
 318{
 319#ifdef CONFIG_ZONE_DMA
 320        /*
 321         * The most common case is KMALLOC_NORMAL, so test for it
 322         * with a single branch for both flags.
 323         */
 324        if (likely((flags & (__GFP_DMA | __GFP_RECLAIMABLE)) == 0))
 325                return KMALLOC_NORMAL;
 326
 327        /*
 328         * At least one of the flags has to be set. If both are, __GFP_DMA
 329         * is more important.
 330         */
 331        return flags & __GFP_DMA ? KMALLOC_DMA : KMALLOC_RECLAIM;
 332#else
 333        return flags & __GFP_RECLAIMABLE ? KMALLOC_RECLAIM : KMALLOC_NORMAL;
 334#endif
 335}
 336
 337/*
 338 * Figure out which kmalloc slab an allocation of a certain size
 339 * belongs to.
 340 * 0 = zero alloc
 341 * 1 =  65 .. 96 bytes
 342 * 2 = 129 .. 192 bytes
 343 * n = 2^(n-1)+1 .. 2^n
 344 */
 345static __always_inline unsigned int kmalloc_index(size_t size)
 346{
 347        if (!size)
 348                return 0;
 349
 350        if (size <= KMALLOC_MIN_SIZE)
 351                return KMALLOC_SHIFT_LOW;
 352
 353        if (KMALLOC_MIN_SIZE <= 32 && size > 64 && size <= 96)
 354                return 1;
 355        if (KMALLOC_MIN_SIZE <= 64 && size > 128 && size <= 192)
 356                return 2;
 357        if (size <=          8) return 3;
 358        if (size <=         16) return 4;
 359        if (size <=         32) return 5;
 360        if (size <=         64) return 6;
 361        if (size <=        128) return 7;
 362        if (size <=        256) return 8;
 363        if (size <=        512) return 9;
 364        if (size <=       1024) return 10;
 365        if (size <=   2 * 1024) return 11;
 366        if (size <=   4 * 1024) return 12;
 367        if (size <=   8 * 1024) return 13;
 368        if (size <=  16 * 1024) return 14;
 369        if (size <=  32 * 1024) return 15;
 370        if (size <=  64 * 1024) return 16;
 371        if (size <= 128 * 1024) return 17;
 372        if (size <= 256 * 1024) return 18;
 373        if (size <= 512 * 1024) return 19;
 374        if (size <= 1024 * 1024) return 20;
 375        if (size <=  2 * 1024 * 1024) return 21;
 376        if (size <=  4 * 1024 * 1024) return 22;
 377        if (size <=  8 * 1024 * 1024) return 23;
 378        if (size <=  16 * 1024 * 1024) return 24;
 379        if (size <=  32 * 1024 * 1024) return 25;
 380        if (size <=  64 * 1024 * 1024) return 26;
 381        BUG();
 382
 383        /* Will never be reached. Needed because the compiler may complain */
 384        return -1;
 385}
 386#endif /* !CONFIG_SLOB */
 387
 388void *__kmalloc(size_t size, gfp_t flags) __assume_kmalloc_alignment __malloc;
 389void *kmem_cache_alloc(struct kmem_cache *, gfp_t flags) __assume_slab_alignment __malloc;
 390void kmem_cache_free(struct kmem_cache *, void *);
 391
 392/*
 393 * Bulk allocation and freeing operations. These are accelerated in an
 394 * allocator specific way to avoid taking locks repeatedly or building
 395 * metadata structures unnecessarily.
 396 *
 397 * Note that interrupts must be enabled when calling these functions.
 398 */
 399void kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
 400int kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
 401
 402/*
 403 * Caller must not use kfree_bulk() on memory not originally allocated
 404 * by kmalloc(), because the SLOB allocator cannot handle this.
 405 */
 406static __always_inline void kfree_bulk(size_t size, void **p)
 407{
 408        kmem_cache_free_bulk(NULL, size, p);
 409}
 410
 411#ifdef CONFIG_NUMA
 412void *__kmalloc_node(size_t size, gfp_t flags, int node) __assume_kmalloc_alignment __malloc;
 413void *kmem_cache_alloc_node(struct kmem_cache *, gfp_t flags, int node) __assume_slab_alignment __malloc;
 414#else
 415static __always_inline void *__kmalloc_node(size_t size, gfp_t flags, int node)
 416{
 417        return __kmalloc(size, flags);
 418}
 419
 420static __always_inline void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t flags, int node)
 421{
 422        return kmem_cache_alloc(s, flags);
 423}
 424#endif
 425
 426#ifdef CONFIG_TRACING
 427extern void *kmem_cache_alloc_trace(struct kmem_cache *, gfp_t, size_t) __assume_slab_alignment __malloc;
 428
 429#ifdef CONFIG_NUMA
 430extern void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
 431                                           gfp_t gfpflags,
 432                                           int node, size_t size) __assume_slab_alignment __malloc;
 433#else
 434static __always_inline void *
 435kmem_cache_alloc_node_trace(struct kmem_cache *s,
 436                              gfp_t gfpflags,
 437                              int node, size_t size)
 438{
 439        return kmem_cache_alloc_trace(s, gfpflags, size);
 440}
 441#endif /* CONFIG_NUMA */
 442
 443#else /* CONFIG_TRACING */
 444static __always_inline void *kmem_cache_alloc_trace(struct kmem_cache *s,
 445                gfp_t flags, size_t size)
 446{
 447        void *ret = kmem_cache_alloc(s, flags);
 448
 449        ret = kasan_kmalloc(s, ret, size, flags);
 450        return ret;
 451}
 452
 453static __always_inline void *
 454kmem_cache_alloc_node_trace(struct kmem_cache *s,
 455                              gfp_t gfpflags,
 456                              int node, size_t size)
 457{
 458        void *ret = kmem_cache_alloc_node(s, gfpflags, node);
 459
 460        ret = kasan_kmalloc(s, ret, size, gfpflags);
 461        return ret;
 462}
 463#endif /* CONFIG_TRACING */
 464
 465extern void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) __assume_page_alignment __malloc;
 466
 467#ifdef CONFIG_TRACING
 468extern void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) __assume_page_alignment __malloc;
 469#else
 470static __always_inline void *
 471kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
 472{
 473        return kmalloc_order(size, flags, order);
 474}
 475#endif
 476
 477static __always_inline void *kmalloc_large(size_t size, gfp_t flags)
 478{
 479        unsigned int order = get_order(size);
 480        return kmalloc_order_trace(size, flags, order);
 481}
 482
 483/**
 484 * kmalloc - allocate memory
 485 * @size: how many bytes of memory are required.
 486 * @flags: the type of memory to allocate.
 487 *
 488 * kmalloc is the normal method of allocating memory
 489 * for objects smaller than page size in the kernel.
 490 *
 491 * The @flags argument may be one of the GFP flags defined at
 492 * include/linux/gfp.h and described at
 493 * :ref:`Documentation/core-api/mm-api.rst <mm-api-gfp-flags>`
 494 *
 495 * The recommended usage of the @flags is described at
 496 * :ref:`Documentation/core-api/memory-allocation.rst <memory-allocation>`
 497 *
 498 * Below is a brief outline of the most useful GFP flags
 499 *
 500 * %GFP_KERNEL
 501 *      Allocate normal kernel ram. May sleep.
 502 *
 503 * %GFP_NOWAIT
 504 *      Allocation will not sleep.
 505 *
 506 * %GFP_ATOMIC
 507 *      Allocation will not sleep.  May use emergency pools.
 508 *
 509 * %GFP_HIGHUSER
 510 *      Allocate memory from high memory on behalf of user.
 511 *
 512 * Also it is possible to set different flags by OR'ing
 513 * in one or more of the following additional @flags:
 514 *
 515 * %__GFP_HIGH
 516 *      This allocation has high priority and may use emergency pools.
 517 *
 518 * %__GFP_NOFAIL
 519 *      Indicate that this allocation is in no way allowed to fail
 520 *      (think twice before using).
 521 *
 522 * %__GFP_NORETRY
 523 *      If memory is not immediately available,
 524 *      then give up at once.
 525 *
 526 * %__GFP_NOWARN
 527 *      If allocation fails, don't issue any warnings.
 528 *
 529 * %__GFP_RETRY_MAYFAIL
 530 *      Try really hard to succeed the allocation but fail
 531 *      eventually.
 532 */
 533static __always_inline void *kmalloc(size_t size, gfp_t flags)
 534{
 535        if (__builtin_constant_p(size)) {
 536#ifndef CONFIG_SLOB
 537                unsigned int index;
 538#endif
 539                if (size > KMALLOC_MAX_CACHE_SIZE)
 540                        return kmalloc_large(size, flags);
 541#ifndef CONFIG_SLOB
 542                index = kmalloc_index(size);
 543
 544                if (!index)
 545                        return ZERO_SIZE_PTR;
 546
 547                return kmem_cache_alloc_trace(
 548                                kmalloc_caches[kmalloc_type(flags)][index],
 549                                flags, size);
 550#endif
 551        }
 552        return __kmalloc(size, flags);
 553}
 554
 555/*
 556 * Determine size used for the nth kmalloc cache.
 557 * return size or 0 if a kmalloc cache for that
 558 * size does not exist
 559 */
 560static __always_inline unsigned int kmalloc_size(unsigned int n)
 561{
 562#ifndef CONFIG_SLOB
 563        if (n > 2)
 564                return 1U << n;
 565
 566        if (n == 1 && KMALLOC_MIN_SIZE <= 32)
 567                return 96;
 568
 569        if (n == 2 && KMALLOC_MIN_SIZE <= 64)
 570                return 192;
 571#endif
 572        return 0;
 573}
 574
 575static __always_inline void *kmalloc_node(size_t size, gfp_t flags, int node)
 576{
 577#ifndef CONFIG_SLOB
 578        if (__builtin_constant_p(size) &&
 579                size <= KMALLOC_MAX_CACHE_SIZE) {
 580                unsigned int i = kmalloc_index(size);
 581
 582                if (!i)
 583                        return ZERO_SIZE_PTR;
 584
 585                return kmem_cache_alloc_node_trace(
 586                                kmalloc_caches[kmalloc_type(flags)][i],
 587                                                flags, node, size);
 588        }
 589#endif
 590        return __kmalloc_node(size, flags, node);
 591}
 592
 593struct memcg_cache_array {
 594        struct rcu_head rcu;
 595        struct kmem_cache *entries[0];
 596};
 597
 598/*
 599 * This is the main placeholder for memcg-related information in kmem caches.
 600 * Both the root cache and the child caches will have it. For the root cache,
 601 * this will hold a dynamically allocated array large enough to hold
 602 * information about the currently limited memcgs in the system. To allow the
 603 * array to be accessed without taking any locks, on relocation we free the old
 604 * version only after a grace period.
 605 *
 606 * Root and child caches hold different metadata.
 607 *
 608 * @root_cache: Common to root and child caches.  NULL for root, pointer to
 609 *              the root cache for children.
 610 *
 611 * The following fields are specific to root caches.
 612 *
 613 * @memcg_caches: kmemcg ID indexed table of child caches.  This table is
 614 *              used to index child cachces during allocation and cleared
 615 *              early during shutdown.
 616 *
 617 * @root_caches_node: List node for slab_root_caches list.
 618 *
 619 * @children:   List of all child caches.  While the child caches are also
 620 *              reachable through @memcg_caches, a child cache remains on
 621 *              this list until it is actually destroyed.
 622 *
 623 * The following fields are specific to child caches.
 624 *
 625 * @memcg:      Pointer to the memcg this cache belongs to.
 626 *
 627 * @children_node: List node for @root_cache->children list.
 628 *
 629 * @kmem_caches_node: List node for @memcg->kmem_caches list.
 630 */
 631struct memcg_cache_params {
 632        struct kmem_cache *root_cache;
 633        union {
 634                struct {
 635                        struct memcg_cache_array __rcu *memcg_caches;
 636                        struct list_head __root_caches_node;
 637                        struct list_head children;
 638                        bool dying;
 639                };
 640                struct {
 641                        struct mem_cgroup *memcg;
 642                        struct list_head children_node;
 643                        struct list_head kmem_caches_node;
 644
 645                        void (*deact_fn)(struct kmem_cache *);
 646                        union {
 647                                struct rcu_head deact_rcu_head;
 648                                struct work_struct deact_work;
 649                        };
 650                };
 651        };
 652};
 653
 654int memcg_update_all_caches(int num_memcgs);
 655
 656/**
 657 * kmalloc_array - allocate memory for an array.
 658 * @n: number of elements.
 659 * @size: element size.
 660 * @flags: the type of memory to allocate (see kmalloc).
 661 */
 662static inline void *kmalloc_array(size_t n, size_t size, gfp_t flags)
 663{
 664        size_t bytes;
 665
 666        if (unlikely(check_mul_overflow(n, size, &bytes)))
 667                return NULL;
 668        if (__builtin_constant_p(n) && __builtin_constant_p(size))
 669                return kmalloc(bytes, flags);
 670        return __kmalloc(bytes, flags);
 671}
 672
 673/**
 674 * kcalloc - allocate memory for an array. The memory is set to zero.
 675 * @n: number of elements.
 676 * @size: element size.
 677 * @flags: the type of memory to allocate (see kmalloc).
 678 */
 679static inline void *kcalloc(size_t n, size_t size, gfp_t flags)
 680{
 681        return kmalloc_array(n, size, flags | __GFP_ZERO);
 682}
 683
 684/*
 685 * kmalloc_track_caller is a special version of kmalloc that records the
 686 * calling function of the routine calling it for slab leak tracking instead
 687 * of just the calling function (confusing, eh?).
 688 * It's useful when the call to kmalloc comes from a widely-used standard
 689 * allocator where we care about the real place the memory allocation
 690 * request comes from.
 691 */
 692extern void *__kmalloc_track_caller(size_t, gfp_t, unsigned long);
 693#define kmalloc_track_caller(size, flags) \
 694        __kmalloc_track_caller(size, flags, _RET_IP_)
 695
 696static inline void *kmalloc_array_node(size_t n, size_t size, gfp_t flags,
 697                                       int node)
 698{
 699        size_t bytes;
 700
 701        if (unlikely(check_mul_overflow(n, size, &bytes)))
 702                return NULL;
 703        if (__builtin_constant_p(n) && __builtin_constant_p(size))
 704                return kmalloc_node(bytes, flags, node);
 705        return __kmalloc_node(bytes, flags, node);
 706}
 707
 708static inline void *kcalloc_node(size_t n, size_t size, gfp_t flags, int node)
 709{
 710        return kmalloc_array_node(n, size, flags | __GFP_ZERO, node);
 711}
 712
 713
 714#ifdef CONFIG_NUMA
 715extern void *__kmalloc_node_track_caller(size_t, gfp_t, int, unsigned long);
 716#define kmalloc_node_track_caller(size, flags, node) \
 717        __kmalloc_node_track_caller(size, flags, node, \
 718                        _RET_IP_)
 719
 720#else /* CONFIG_NUMA */
 721
 722#define kmalloc_node_track_caller(size, flags, node) \
 723        kmalloc_track_caller(size, flags)
 724
 725#endif /* CONFIG_NUMA */
 726
 727/*
 728 * Shortcuts
 729 */
 730static inline void *kmem_cache_zalloc(struct kmem_cache *k, gfp_t flags)
 731{
 732        return kmem_cache_alloc(k, flags | __GFP_ZERO);
 733}
 734
 735/**
 736 * kzalloc - allocate memory. The memory is set to zero.
 737 * @size: how many bytes of memory are required.
 738 * @flags: the type of memory to allocate (see kmalloc).
 739 */
 740static inline void *kzalloc(size_t size, gfp_t flags)
 741{
 742        return kmalloc(size, flags | __GFP_ZERO);
 743}
 744
 745/**
 746 * kzalloc_node - allocate zeroed memory from a particular memory node.
 747 * @size: how many bytes of memory are required.
 748 * @flags: the type of memory to allocate (see kmalloc).
 749 * @node: memory node from which to allocate
 750 */
 751static inline void *kzalloc_node(size_t size, gfp_t flags, int node)
 752{
 753        return kmalloc_node(size, flags | __GFP_ZERO, node);
 754}
 755
 756unsigned int kmem_cache_size(struct kmem_cache *s);
 757void __init kmem_cache_init_late(void);
 758
 759#if defined(CONFIG_SMP) && defined(CONFIG_SLAB)
 760int slab_prepare_cpu(unsigned int cpu);
 761int slab_dead_cpu(unsigned int cpu);
 762#else
 763#define slab_prepare_cpu        NULL
 764#define slab_dead_cpu           NULL
 765#endif
 766
 767#endif  /* _LINUX_SLAB_H */
 768