include/linux/slab.h

   1 /*
   2  * Written by Mark Hemment, 1996 (markhe@nextd.demon.co.uk).
   3  *
   4  * (C) SGI 2006, Christoph Lameter
   5  *      Cleaned up and restructured to ease the addition of alternative
   6  *      implementations of SLAB allocators.
   7  */
   8
   9 #ifndef _LINUX_SLAB_H
  10 #define _LINUX_SLAB_H
  11
  12 #include <linux/gfp.h>
  13 #include <linux/types.h>
  14 #include <linux/workqueue.h>
  15
  16
  17 /*
  18  * Flags to pass to kmem_cache_create().
  19  * The ones marked DEBUG are only valid if CONFIG_SLAB_DEBUG is set.
  20  */
  21 #define SLAB_DEBUG_FREE         0x00000100UL    /* DEBUG: Perform (expensive) checks on free */
  22 #define SLAB_RED_ZONE           0x00000400UL    /* DEBUG: Red zone objs in a cache */
  23 #define SLAB_POISON             0x00000800UL    /* DEBUG: Poison objects */
  24 #define SLAB_HWCACHE_ALIGN      0x00002000UL    /* Align objs on cache lines */
  25 #define SLAB_CACHE_DMA          0x00004000UL    /* Use GFP_DMA memory */
  26 #define SLAB_STORE_USER         0x00010000UL    /* DEBUG: Store the last owner for bug hunting */
  27 #define SLAB_PANIC              0x00040000UL    /* Panic if kmem_cache_create() fails */
  28 /*
  29  * SLAB_DESTROY_BY_RCU - **WARNING** READ THIS!
  30  *
  31  * This delays freeing the SLAB page by a grace period, it does _NOT_
  32  * delay object freeing. This means that if you do kmem_cache_free()
  33  * that memory location is free to be reused at any time. Thus it may
  34  * be possible to see another object there in the same RCU grace period.
  35  *
  36  * This feature only ensures the memory location backing the object
  37  * stays valid, the trick to using this is relying on an independent
  38  * object validation pass. Something like:
  39  *
  40  *  rcu_read_lock()
  41  * again:
  42  *  obj = lockless_lookup(key);
  43  *  if (obj) {
  44  *    if (!try_get_ref(obj)) // might fail for free objects
  45  *      goto again;
  46  *
  47  *    if (obj->key != key) { // not the object we expected
  48  *      put_ref(obj);
  49  *      goto again;
  50  *    }
  51  *  }
  52  *  rcu_read_unlock();
  53  *
  54  * See also the comment on struct slab_rcu in mm/slab.c.
  55  */
  56 #define SLAB_DESTROY_BY_RCU     0x00080000UL    /* Defer freeing slabs to RCU */
  57 #define SLAB_MEM_SPREAD         0x00100000UL    /* Spread some memory over cpuset */
  58 #define SLAB_TRACE              0x00200000UL    /* Trace allocations and frees */
  59
  60 /* Flag to prevent checks on free */
  61 #ifdef CONFIG_DEBUG_OBJECTS
  62 # define SLAB_DEBUG_OBJECTS     0x00400000UL
  63 #else
  64 # define SLAB_DEBUG_OBJECTS     0x00000000UL
  65 #endif
  66
  67 #define SLAB_NOLEAKTRACE        0x00800000UL    /* Avoid kmemleak tracing */
  68
  69 /* Don't track use of uninitialized memory */
  70 #ifdef CONFIG_KMEMCHECK
  71 # define SLAB_NOTRACK           0x01000000UL
  72 #else
  73 # define SLAB_NOTRACK           0x00000000UL
  74 #endif
  75 #ifdef CONFIG_FAILSLAB
  76 # define SLAB_FAILSLAB          0x02000000UL    /* Fault injection mark */
  77 #else
  78 # define SLAB_FAILSLAB          0x00000000UL
  79 #endif
  80
  81 /* The following flags affect the page allocator grouping pages by mobility */
  82 #define SLAB_RECLAIM_ACCOUNT    0x00020000UL            /* Objects are reclaimable */
  83 #define SLAB_TEMPORARY          SLAB_RECLAIM_ACCOUNT    /* Objects are short-lived */
  84 /*
  85  * ZERO_SIZE_PTR will be returned for zero sized kmalloc requests.
  86  *
  87  * Dereferencing ZERO_SIZE_PTR will lead to a distinct access fault.
  88  *
  89  * ZERO_SIZE_PTR can be passed to kfree though in the same way that NULL can.
  90  * Both make kfree a no-op.
  91  */
  92 #define ZERO_SIZE_PTR ((void *)16)
  93
  94 #define ZERO_OR_NULL_PTR(x) ((unsigned long)(x) <= \
  95                                 (unsigned long)ZERO_SIZE_PTR)
  96
  97
  98 struct mem_cgroup;
  99 /*
 100  * struct kmem_cache related prototypes
 101  */
 102 void __init kmem_cache_init(void);
 103 int slab_is_available(void);
 104
 105 struct kmem_cache *kmem_cache_create(const char *, size_t, size_t,
 106                         unsigned long,
 107                         void (*)(void *));
 108 struct kmem_cache *
 109 kmem_cache_create_memcg(struct mem_cgroup *, const char *, size_t, size_t,
 110                         unsigned long, void (*)(void *), struct kmem_cache *);
 111 void kmem_cache_destroy(struct kmem_cache *);
 112 int kmem_cache_shrink(struct kmem_cache *);
 113 void kmem_cache_free(struct kmem_cache *, void *);
 114
 115 /*
 116  * Please use this macro to create slab caches. Simply specify the
 117  * name of the structure and maybe some flags that are listed above.
 118  *
 119  * The alignment of the struct determines object alignment. If you
 120  * f.e. add ____cacheline_aligned_in_smp to the struct declaration
 121  * then the objects will be properly aligned in SMP configurations.
 122  */
 123 #define KMEM_CACHE(__struct, __flags) kmem_cache_create(#__struct,\
 124                 sizeof(struct __struct), __alignof__(struct __struct),\
 125                 (__flags), NULL)
 126
 127 /*
 128  * Common kmalloc functions provided by all allocators
 129  */
 130 void * __must_check __krealloc(const void *, size_t, gfp_t);
 131 void * __must_check krealloc(const void *, size_t, gfp_t);
 132 void kfree(const void *);
 133 void kzfree(const void *);
 134 size_t ksize(const void *);
 135
 136 /*
 137  * Some archs want to perform DMA into kmalloc caches and need a guaranteed
 138  * alignment larger than the alignment of a 64-bit integer.
 139  * Setting ARCH_KMALLOC_MINALIGN in arch headers allows that.
 140  */
 141 #if defined(ARCH_DMA_MINALIGN) && ARCH_DMA_MINALIGN > 8
 142 #define ARCH_KMALLOC_MINALIGN ARCH_DMA_MINALIGN
 143 #define KMALLOC_MIN_SIZE ARCH_DMA_MINALIGN
 144 #define KMALLOC_SHIFT_LOW ilog2(ARCH_DMA_MINALIGN)
 145 #else
 146 #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
 147 #endif
 148
 149 #ifdef CONFIG_SLOB
 150 /*
 151  * Common fields provided in kmem_cache by all slab allocators
 152  * This struct is either used directly by the allocator (SLOB)
 153  * or the allocator must include definitions for all fields
 154  * provided in kmem_cache_common in their definition of kmem_cache.
 155  *
 156  * Once we can do anonymous structs (C11 standard) we could put a
 157  * anonymous struct definition in these allocators so that the
 158  * separate allocations in the kmem_cache structure of SLAB and
 159  * SLUB is no longer needed.
 160  */
 161 struct kmem_cache {
 162         unsigned int object_size;/* The original size of the object */
 163         unsigned int size;      /* The aligned/padded/added on size  */
 164         unsigned int align;     /* Alignment as calculated */
 165         unsigned long flags;    /* Active flags on the slab */
 166         const char *name;       /* Slab name for sysfs */
 167         int refcount;           /* Use counter */
 168         void (*ctor)(void *);   /* Called on object slot creation */
 169         struct list_head list;  /* List of all slab caches on the system */
 170 };
 171
 172 #endif /* CONFIG_SLOB */
 173
 174 /*
 175  * Kmalloc array related definitions
 176  */
 177
 178 #ifdef CONFIG_SLAB
 179 /*
 180  * The largest kmalloc size supported by the SLAB allocators is
 181  * 32 megabyte (2^25) or the maximum allocatable page order if that is
 182  * less than 32 MB.
 183  *
 184  * WARNING: Its not easy to increase this value since the allocators have
 185  * to do various tricks to work around compiler limitations in order to
 186  * ensure proper constant folding.
 187  */
 188 #define KMALLOC_SHIFT_HIGH      ((MAX_ORDER + PAGE_SHIFT - 1) <= 25 ? \
 189                                 (MAX_ORDER + PAGE_SHIFT - 1) : 25)
 190 #define KMALLOC_SHIFT_MAX       KMALLOC_SHIFT_HIGH
 191 #ifndef KMALLOC_SHIFT_LOW
 192 #define KMALLOC_SHIFT_LOW       5
 193 #endif
 194 #endif
 195
 196 #ifdef CONFIG_SLUB
 197 /*
 198  * SLUB allocates up to order 2 pages directly and otherwise
 199  * passes the request to the page allocator.
 200  */
 201 #define KMALLOC_SHIFT_HIGH      (PAGE_SHIFT + 1)
 202 #define KMALLOC_SHIFT_MAX       (MAX_ORDER + PAGE_SHIFT)
 203 #ifndef KMALLOC_SHIFT_LOW
 204 #define KMALLOC_SHIFT_LOW       3
 205 #endif
 206 #endif
 207
 208 #ifdef CONFIG_SLOB
 209 /*
 210  * SLOB passes all page size and larger requests to the page allocator.
 211  * No kmalloc array is necessary since objects of different sizes can
 212  * be allocated from the same page.
 213  */
 214 #define KMALLOC_SHIFT_MAX       30
 215 #define KMALLOC_SHIFT_HIGH      PAGE_SHIFT
 216 #ifndef KMALLOC_SHIFT_LOW
 217 #define KMALLOC_SHIFT_LOW       3
 218 #endif
 219 #endif
 220
 221 /* Maximum allocatable size */
 222 #define KMALLOC_MAX_SIZE        (1UL << KMALLOC_SHIFT_MAX)
 223 /* Maximum size for which we actually use a slab cache */
 224 #define KMALLOC_MAX_CACHE_SIZE  (1UL << KMALLOC_SHIFT_HIGH)
 225 /* Maximum order allocatable via the slab allocagtor */
 226 #define KMALLOC_MAX_ORDER       (KMALLOC_SHIFT_MAX - PAGE_SHIFT)
 227
 228 /*
 229  * Kmalloc subsystem.
 230  */
 231 #ifndef KMALLOC_MIN_SIZE
 232 #define KMALLOC_MIN_SIZE (1 << KMALLOC_SHIFT_LOW)
 233 #endif
 234
 235 #ifndef CONFIG_SLOB
 236 extern struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1];
 237 #ifdef CONFIG_ZONE_DMA
 238 extern struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1];
 239 #endif
 240
 241 /*
 242  * Figure out which kmalloc slab an allocation of a certain size
 243  * belongs to.
 244  * 0 = zero alloc
 245  * 1 =  65 .. 96 bytes
 246  * 2 = 120 .. 192 bytes
 247  * n = 2^(n-1) .. 2^n -1
 248  */
 249 static __always_inline int kmalloc_index(size_t size)
 250 {
 251         if (!size)
 252                 return 0;
 253
 254         if (size <= KMALLOC_MIN_SIZE)
 255                 return KMALLOC_SHIFT_LOW;
 256
 257         if (KMALLOC_MIN_SIZE <= 32 && size > 64 && size <= 96)
 258                 return 1;
 259         if (KMALLOC_MIN_SIZE <= 64 && size > 128 && size <= 192)
 260                 return 2;
 261         if (size <=          8) return 3;
 262         if (size <=         16) return 4;
 263         if (size <=         32) return 5;
 264         if (size <=         64) return 6;
 265         if (size <=        128) return 7;
 266         if (size <=        256) return 8;
 267         if (size <=        512) return 9;
 268         if (size <=       1024) return 10;
 269         if (size <=   2 * 1024) return 11;
 270         if (size <=   4 * 1024) return 12;
 271         if (size <=   8 * 1024) return 13;
 272         if (size <=  16 * 1024) return 14;
 273         if (size <=  32 * 1024) return 15;
 274         if (size <=  64 * 1024) return 16;
 275         if (size <= 128 * 1024) return 17;
 276         if (size <= 256 * 1024) return 18;
 277         if (size <= 512 * 1024) return 19;
 278         if (size <= 1024 * 1024) return 20;
 279         if (size <=  2 * 1024 * 1024) return 21;
 280         if (size <=  4 * 1024 * 1024) return 22;
 281         if (size <=  8 * 1024 * 1024) return 23;
 282         if (size <=  16 * 1024 * 1024) return 24;
 283         if (size <=  32 * 1024 * 1024) return 25;
 284         if (size <=  64 * 1024 * 1024) return 26;
 285         BUG();
 286
 287         /* Will never be reached. Needed because the compiler may complain */
 288         return -1;
 289 }
 290 #endif /* !CONFIG_SLOB */
 291
 292 #ifdef CONFIG_SLAB
 293 #include <linux/slab_def.h>
 294 #endif
 295
 296 #ifdef CONFIG_SLUB
 297 #include <linux/slub_def.h>
 298 #endif
 299
 300 #ifdef CONFIG_SLOB
 301 #include <linux/slob_def.h>
 302 #endif
 303
 304 /*
 305  * Determine size used for the nth kmalloc cache.
 306  * return size or 0 if a kmalloc cache for that
 307  * size does not exist
 308  */
 309 static __always_inline int kmalloc_size(int n)
 310 {
 311 #ifndef CONFIG_SLOB
 312         if (n > 2)
 313                 return 1 << n;
 314
 315         if (n == 1 && KMALLOC_MIN_SIZE <= 32)
 316                 return 96;
 317
 318         if (n == 2 && KMALLOC_MIN_SIZE <= 64)
 319                 return 192;
 320 #endif
 321         return 0;
 322 }
 323
 324 /*
 325  * Setting ARCH_SLAB_MINALIGN in arch headers allows a different alignment.
 326  * Intended for arches that get misalignment faults even for 64 bit integer
 327  * aligned buffers.
 328  */
 329 #ifndef ARCH_SLAB_MINALIGN
 330 #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
 331 #endif
 332 /*
 333  * This is the main placeholder for memcg-related information in kmem caches.
 334  * struct kmem_cache will hold a pointer to it, so the memory cost while
 335  * disabled is 1 pointer. The runtime cost while enabled, gets bigger than it
 336  * would otherwise be if that would be bundled in kmem_cache: we'll need an
 337  * extra pointer chase. But the trade off clearly lays in favor of not
 338  * penalizing non-users.
 339  *
 340  * Both the root cache and the child caches will have it. For the root cache,
 341  * this will hold a dynamically allocated array large enough to hold
 342  * information about the currently limited memcgs in the system.
 343  *
 344  * Child caches will hold extra metadata needed for its operation. Fields are:
 345  *
 346  * @memcg: pointer to the memcg this cache belongs to
 347  * @list: list_head for the list of all caches in this memcg
 348  * @root_cache: pointer to the global, root cache, this cache was derived from
 349  * @dead: set to true after the memcg dies; the cache may still be around.
 350  * @nr_pages: number of pages that belongs to this cache.
 351  * @destroy: worker to be called whenever we are ready, or believe we may be
 352  *           ready, to destroy this cache.
 353  */
 354 struct memcg_cache_params {
 355         bool is_root_cache;
 356         union {
 357                 struct kmem_cache *memcg_caches[0];
 358                 struct {
 359                         struct mem_cgroup *memcg;
 360                         struct list_head list;
 361                         struct kmem_cache *root_cache;
 362                         bool dead;
 363                         atomic_t nr_pages;
 364                         struct work_struct destroy;
 365                 };
 366         };
 367 };
 368
 369 int memcg_update_all_caches(int num_memcgs);
 370
 371 struct seq_file;
 372 int cache_show(struct kmem_cache *s, struct seq_file *m);
 373 void print_slabinfo_header(struct seq_file *m);
 374
 375 /**
 376  * kmalloc - allocate memory
 377  * @size: how many bytes of memory are required.
 378  * @flags: the type of memory to allocate.
 379  *
 380  * The @flags argument may be one of:
 381  *
 382  * %GFP_USER - Allocate memory on behalf of user.  May sleep.
 383  *
 384  * %GFP_KERNEL - Allocate normal kernel ram.  May sleep.
 385  *
 386  * %GFP_ATOMIC - Allocation will not sleep.  May use emergency pools.
 387  *   For example, use this inside interrupt handlers.
 388  *
 389  * %GFP_HIGHUSER - Allocate pages from high memory.
 390  *
 391  * %GFP_NOIO - Do not do any I/O at all while trying to get memory.
 392  *
 393  * %GFP_NOFS - Do not make any fs calls while trying to get memory.
 394  *
 395  * %GFP_NOWAIT - Allocation will not sleep.
 396  *
 397  * %GFP_THISNODE - Allocate node-local memory only.
 398  *
 399  * %GFP_DMA - Allocation suitable for DMA.
 400  *   Should only be used for kmalloc() caches. Otherwise, use a
 401  *   slab created with SLAB_DMA.
 402  *
 403  * Also it is possible to set different flags by OR'ing
 404  * in one or more of the following additional @flags:
 405  *
 406  * %__GFP_COLD - Request cache-cold pages instead of
 407  *   trying to return cache-warm pages.
 408  *
 409  * %__GFP_HIGH - This allocation has high priority and may use emergency pools.
 410  *
 411  * %__GFP_NOFAIL - Indicate that this allocation is in no way allowed to fail
 412  *   (think twice before using).
 413  *
 414  * %__GFP_NORETRY - If memory is not immediately available,
 415  *   then give up at once.
 416  *
 417  * %__GFP_NOWARN - If allocation fails, don't issue any warnings.
 418  *
 419  * %__GFP_REPEAT - If allocation fails initially, try once more before failing.
 420  *
 421  * There are other flags available as well, but these are not intended
 422  * for general use, and so are not documented here. For a full list of
 423  * potential flags, always refer to linux/gfp.h.
 424  *
 425  * kmalloc is the normal method of allocating memory
 426  * in the kernel.
 427  */
 428 static __always_inline void *kmalloc(size_t size, gfp_t flags);
 429
 430 /**
 431  * kmalloc_array - allocate memory for an array.
 432  * @n: number of elements.
 433  * @size: element size.
 434  * @flags: the type of memory to allocate (see kmalloc).
 435  */
 436 static inline void *kmalloc_array(size_t n, size_t size, gfp_t flags)
 437 {
 438         if (size != 0 && n > SIZE_MAX / size)
 439                 return NULL;
 440         return __kmalloc(n * size, flags);
 441 }
 442
 443 /**
 444  * kcalloc - allocate memory for an array. The memory is set to zero.
 445  * @n: number of elements.
 446  * @size: element size.
 447  * @flags: the type of memory to allocate (see kmalloc).
 448  */
 449 static inline void *kcalloc(size_t n, size_t size, gfp_t flags)
 450 {
 451         return kmalloc_array(n, size, flags | __GFP_ZERO);
 452 }
 453
 454 #if !defined(CONFIG_NUMA) && !defined(CONFIG_SLOB)
 455 /**
 456  * kmalloc_node - allocate memory from a specific node
 457  * @size: how many bytes of memory are required.
 458  * @flags: the type of memory to allocate (see kmalloc).
 459  * @node: node to allocate from.
 460  *
 461  * kmalloc() for non-local nodes, used to allocate from a specific node
 462  * if available. Equivalent to kmalloc() in the non-NUMA single-node
 463  * case.
 464  */
 465 static inline void *kmalloc_node(size_t size, gfp_t flags, int node)
 466 {
 467         return kmalloc(size, flags);
 468 }
 469
 470 static inline void *__kmalloc_node(size_t size, gfp_t flags, int node)
 471 {
 472         return __kmalloc(size, flags);
 473 }
 474
 475 void *kmem_cache_alloc(struct kmem_cache *, gfp_t);
 476
 477 static inline void *kmem_cache_alloc_node(struct kmem_cache *cachep,
 478                                         gfp_t flags, int node)
 479 {
 480         return kmem_cache_alloc(cachep, flags);
 481 }
 482 #endif /* !CONFIG_NUMA && !CONFIG_SLOB */
 483
 484 /*
 485  * kmalloc_track_caller is a special version of kmalloc that records the
 486  * calling function of the routine calling it for slab leak tracking instead
 487  * of just the calling function (confusing, eh?).
 488  * It's useful when the call to kmalloc comes from a widely-used standard
 489  * allocator where we care about the real place the memory allocation
 490  * request comes from.
 491  */
 492 #if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_SLUB) || \
 493         (defined(CONFIG_SLAB) && defined(CONFIG_TRACING)) || \
 494         (defined(CONFIG_SLOB) && defined(CONFIG_TRACING))
 495 extern void *__kmalloc_track_caller(size_t, gfp_t, unsigned long);
 496 #define kmalloc_track_caller(size, flags) \
 497         __kmalloc_track_caller(size, flags, _RET_IP_)
 498 #else
 499 #define kmalloc_track_caller(size, flags) \
 500         __kmalloc(size, flags)
 501 #endif /* DEBUG_SLAB */
 502
 503 #ifdef CONFIG_NUMA
 504 /*
 505  * kmalloc_node_track_caller is a special version of kmalloc_node that
 506  * records the calling function of the routine calling it for slab leak
 507  * tracking instead of just the calling function (confusing, eh?).
 508  * It's useful when the call to kmalloc_node comes from a widely-used
 509  * standard allocator where we care about the real place the memory
 510  * allocation request comes from.
 511  */
 512 #if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_SLUB) || \
 513         (defined(CONFIG_SLAB) && defined(CONFIG_TRACING)) || \
 514         (defined(CONFIG_SLOB) && defined(CONFIG_TRACING))
 515 extern void *__kmalloc_node_track_caller(size_t, gfp_t, int, unsigned long);
 516 #define kmalloc_node_track_caller(size, flags, node) \
 517         __kmalloc_node_track_caller(size, flags, node, \
 518                         _RET_IP_)
 519 #else
 520 #define kmalloc_node_track_caller(size, flags, node) \
 521         __kmalloc_node(size, flags, node)
 522 #endif
 523
 524 #else /* CONFIG_NUMA */
 525
 526 #define kmalloc_node_track_caller(size, flags, node) \
 527         kmalloc_track_caller(size, flags)
 528
 529 #endif /* CONFIG_NUMA */
 530
 531 /*
 532  * Shortcuts
 533  */
 534 static inline void *kmem_cache_zalloc(struct kmem_cache *k, gfp_t flags)
 535 {
 536         return kmem_cache_alloc(k, flags | __GFP_ZERO);
 537 }
 538
 539 /**
 540  * kzalloc - allocate memory. The memory is set to zero.
 541  * @size: how many bytes of memory are required.
 542  * @flags: the type of memory to allocate (see kmalloc).
 543  */
 544 static inline void *kzalloc(size_t size, gfp_t flags)
 545 {
 546         return kmalloc(size, flags | __GFP_ZERO);
 547 }
 548
 549 /**
 550  * kzalloc_node - allocate zeroed memory from a particular memory node.
 551  * @size: how many bytes of memory are required.
 552  * @flags: the type of memory to allocate (see kmalloc).
 553  * @node: memory node from which to allocate
 554  */
 555 static inline void *kzalloc_node(size_t size, gfp_t flags, int node)
 556 {
 557         return kmalloc_node(size, flags | __GFP_ZERO, node);
 558 }
 559
 560 /*
 561  * Determine the size of a slab object
 562  */
 563 static inline unsigned int kmem_cache_size(struct kmem_cache *s)
 564 {
 565         return s->object_size;
 566 }
 567
 568 void __init kmem_cache_init_late(void);
 569
 570 #endif  /* _LINUX_SLAB_H */