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 #define KMALLOC_MAX_SIZE (1UL << 30)
 173
 174 #include <linux/slob_def.h>
 175
 176 #else /* CONFIG_SLOB */
 177
 178 /*
 179  * Kmalloc array related definitions
 180  */
 181
 182 #ifdef CONFIG_SLAB
 183 /*
 184  * The largest kmalloc size supported by the SLAB allocators is
 185  * 32 megabyte (2^25) or the maximum allocatable page order if that is
 186  * less than 32 MB.
 187  *
 188  * WARNING: Its not easy to increase this value since the allocators have
 189  * to do various tricks to work around compiler limitations in order to
 190  * ensure proper constant folding.
 191  */
 192 #define KMALLOC_SHIFT_HIGH      ((MAX_ORDER + PAGE_SHIFT - 1) <= 25 ? \
 193                                 (MAX_ORDER + PAGE_SHIFT - 1) : 25)
 194 #define KMALLOC_SHIFT_MAX       KMALLOC_SHIFT_HIGH
 195 #ifndef KMALLOC_SHIFT_LOW
 196 #define KMALLOC_SHIFT_LOW       5
 197 #endif
 198 #else
 199 /*
 200  * SLUB allocates up to order 2 pages directly and otherwise
 201  * passes the request to the page allocator.
 202  */
 203 #define KMALLOC_SHIFT_HIGH      (PAGE_SHIFT + 1)
 204 #define KMALLOC_SHIFT_MAX       (MAX_ORDER + PAGE_SHIFT)
 205 #ifndef KMALLOC_SHIFT_LOW
 206 #define KMALLOC_SHIFT_LOW       3
 207 #endif
 208 #endif
 209
 210 /* Maximum allocatable size */
 211 #define KMALLOC_MAX_SIZE        (1UL << KMALLOC_SHIFT_MAX)
 212 /* Maximum size for which we actually use a slab cache */
 213 #define KMALLOC_MAX_CACHE_SIZE  (1UL << KMALLOC_SHIFT_HIGH)
 214 /* Maximum order allocatable via the slab allocagtor */
 215 #define KMALLOC_MAX_ORDER       (KMALLOC_SHIFT_MAX - PAGE_SHIFT)
 216
 217 /*
 218  * Kmalloc subsystem.
 219  */
 220 #ifndef KMALLOC_MIN_SIZE
 221 #define KMALLOC_MIN_SIZE (1 << KMALLOC_SHIFT_LOW)
 222 #endif
 223
 224 extern struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1];
 225 #ifdef CONFIG_ZONE_DMA
 226 extern struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1];
 227 #endif
 228
 229 /*
 230  * Figure out which kmalloc slab an allocation of a certain size
 231  * belongs to.
 232  * 0 = zero alloc
 233  * 1 =  65 .. 96 bytes
 234  * 2 = 120 .. 192 bytes
 235  * n = 2^(n-1) .. 2^n -1
 236  */
 237 static __always_inline int kmalloc_index(size_t size)
 238 {
 239         if (!size)
 240                 return 0;
 241
 242         if (size <= KMALLOC_MIN_SIZE)
 243                 return KMALLOC_SHIFT_LOW;
 244
 245         if (KMALLOC_MIN_SIZE <= 32 && size > 64 && size <= 96)
 246                 return 1;
 247         if (KMALLOC_MIN_SIZE <= 64 && size > 128 && size <= 192)
 248                 return 2;
 249         if (size <=          8) return 3;
 250         if (size <=         16) return 4;
 251         if (size <=         32) return 5;
 252         if (size <=         64) return 6;
 253         if (size <=        128) return 7;
 254         if (size <=        256) return 8;
 255         if (size <=        512) return 9;
 256         if (size <=       1024) return 10;
 257         if (size <=   2 * 1024) return 11;
 258         if (size <=   4 * 1024) return 12;
 259         if (size <=   8 * 1024) return 13;
 260         if (size <=  16 * 1024) return 14;
 261         if (size <=  32 * 1024) return 15;
 262         if (size <=  64 * 1024) return 16;
 263         if (size <= 128 * 1024) return 17;
 264         if (size <= 256 * 1024) return 18;
 265         if (size <= 512 * 1024) return 19;
 266         if (size <= 1024 * 1024) return 20;
 267         if (size <=  2 * 1024 * 1024) return 21;
 268         if (size <=  4 * 1024 * 1024) return 22;
 269         if (size <=  8 * 1024 * 1024) return 23;
 270         if (size <=  16 * 1024 * 1024) return 24;
 271         if (size <=  32 * 1024 * 1024) return 25;
 272         if (size <=  64 * 1024 * 1024) return 26;
 273         BUG();
 274
 275         /* Will never be reached. Needed because the compiler may complain */
 276         return -1;
 277 }
 278
 279 #ifdef CONFIG_SLAB
 280 #include <linux/slab_def.h>
 281 #elif defined(CONFIG_SLUB)
 282 #include <linux/slub_def.h>
 283 #else
 284 #error "Unknown slab allocator"
 285 #endif
 286
 287 /*
 288  * Determine size used for the nth kmalloc cache.
 289  * return size or 0 if a kmalloc cache for that
 290  * size does not exist
 291  */
 292 static __always_inline int kmalloc_size(int n)
 293 {
 294         if (n > 2)
 295                 return 1 << n;
 296
 297         if (n == 1 && KMALLOC_MIN_SIZE <= 32)
 298                 return 96;
 299
 300         if (n == 2 && KMALLOC_MIN_SIZE <= 64)
 301                 return 192;
 302
 303         return 0;
 304 }
 305 #endif /* !CONFIG_SLOB */
 306
 307 /*
 308  * Setting ARCH_SLAB_MINALIGN in arch headers allows a different alignment.
 309  * Intended for arches that get misalignment faults even for 64 bit integer
 310  * aligned buffers.
 311  */
 312 #ifndef ARCH_SLAB_MINALIGN
 313 #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
 314 #endif
 315 /*
 316  * This is the main placeholder for memcg-related information in kmem caches.
 317  * struct kmem_cache will hold a pointer to it, so the memory cost while
 318  * disabled is 1 pointer. The runtime cost while enabled, gets bigger than it
 319  * would otherwise be if that would be bundled in kmem_cache: we'll need an
 320  * extra pointer chase. But the trade off clearly lays in favor of not
 321  * penalizing non-users.
 322  *
 323  * Both the root cache and the child caches will have it. For the root cache,
 324  * this will hold a dynamically allocated array large enough to hold
 325  * information about the currently limited memcgs in the system.
 326  *
 327  * Child caches will hold extra metadata needed for its operation. Fields are:
 328  *
 329  * @memcg: pointer to the memcg this cache belongs to
 330  * @list: list_head for the list of all caches in this memcg
 331  * @root_cache: pointer to the global, root cache, this cache was derived from
 332  * @dead: set to true after the memcg dies; the cache may still be around.
 333  * @nr_pages: number of pages that belongs to this cache.
 334  * @destroy: worker to be called whenever we are ready, or believe we may be
 335  *           ready, to destroy this cache.
 336  */
 337 struct memcg_cache_params {
 338         bool is_root_cache;
 339         union {
 340                 struct kmem_cache *memcg_caches[0];
 341                 struct {
 342                         struct mem_cgroup *memcg;
 343                         struct list_head list;
 344                         struct kmem_cache *root_cache;
 345                         bool dead;
 346                         atomic_t nr_pages;
 347                         struct work_struct destroy;
 348                 };
 349         };
 350 };
 351
 352 int memcg_update_all_caches(int num_memcgs);
 353
 354 struct seq_file;
 355 int cache_show(struct kmem_cache *s, struct seq_file *m);
 356 void print_slabinfo_header(struct seq_file *m);
 357
 358 /**
 359  * kmalloc_array - allocate memory for an array.
 360  * @n: number of elements.
 361  * @size: element size.
 362  * @flags: the type of memory to allocate.
 363  *
 364  * The @flags argument may be one of:
 365  *
 366  * %GFP_USER - Allocate memory on behalf of user.  May sleep.
 367  *
 368  * %GFP_KERNEL - Allocate normal kernel ram.  May sleep.
 369  *
 370  * %GFP_ATOMIC - Allocation will not sleep.  May use emergency pools.
 371  *   For example, use this inside interrupt handlers.
 372  *
 373  * %GFP_HIGHUSER - Allocate pages from high memory.
 374  *
 375  * %GFP_NOIO - Do not do any I/O at all while trying to get memory.
 376  *
 377  * %GFP_NOFS - Do not make any fs calls while trying to get memory.
 378  *
 379  * %GFP_NOWAIT - Allocation will not sleep.
 380  *
 381  * %GFP_THISNODE - Allocate node-local memory only.
 382  *
 383  * %GFP_DMA - Allocation suitable for DMA.
 384  *   Should only be used for kmalloc() caches. Otherwise, use a
 385  *   slab created with SLAB_DMA.
 386  *
 387  * Also it is possible to set different flags by OR'ing
 388  * in one or more of the following additional @flags:
 389  *
 390  * %__GFP_COLD - Request cache-cold pages instead of
 391  *   trying to return cache-warm pages.
 392  *
 393  * %__GFP_HIGH - This allocation has high priority and may use emergency pools.
 394  *
 395  * %__GFP_NOFAIL - Indicate that this allocation is in no way allowed to fail
 396  *   (think twice before using).
 397  *
 398  * %__GFP_NORETRY - If memory is not immediately available,
 399  *   then give up at once.
 400  *
 401  * %__GFP_NOWARN - If allocation fails, don't issue any warnings.
 402  *
 403  * %__GFP_REPEAT - If allocation fails initially, try once more before failing.
 404  *
 405  * There are other flags available as well, but these are not intended
 406  * for general use, and so are not documented here. For a full list of
 407  * potential flags, always refer to linux/gfp.h.
 408  */
 409 static inline void *kmalloc_array(size_t n, size_t size, gfp_t flags)
 410 {
 411         if (size != 0 && n > SIZE_MAX / size)
 412                 return NULL;
 413         return __kmalloc(n * size, flags);
 414 }
 415
 416 /**
 417  * kcalloc - allocate memory for an array. The memory is set to zero.
 418  * @n: number of elements.
 419  * @size: element size.
 420  * @flags: the type of memory to allocate (see kmalloc).
 421  */
 422 static inline void *kcalloc(size_t n, size_t size, gfp_t flags)
 423 {
 424         return kmalloc_array(n, size, flags | __GFP_ZERO);
 425 }
 426
 427 #if !defined(CONFIG_NUMA) && !defined(CONFIG_SLOB)
 428 /**
 429  * kmalloc_node - allocate memory from a specific node
 430  * @size: how many bytes of memory are required.
 431  * @flags: the type of memory to allocate (see kcalloc).
 432  * @node: node to allocate from.
 433  *
 434  * kmalloc() for non-local nodes, used to allocate from a specific node
 435  * if available. Equivalent to kmalloc() in the non-NUMA single-node
 436  * case.
 437  */
 438 static inline void *kmalloc_node(size_t size, gfp_t flags, int node)
 439 {
 440         return kmalloc(size, flags);
 441 }
 442
 443 static inline void *__kmalloc_node(size_t size, gfp_t flags, int node)
 444 {
 445         return __kmalloc(size, flags);
 446 }
 447
 448 void *kmem_cache_alloc(struct kmem_cache *, gfp_t);
 449
 450 static inline void *kmem_cache_alloc_node(struct kmem_cache *cachep,
 451                                         gfp_t flags, int node)
 452 {
 453         return kmem_cache_alloc(cachep, flags);
 454 }
 455 #endif /* !CONFIG_NUMA && !CONFIG_SLOB */
 456
 457 /*
 458  * kmalloc_track_caller is a special version of kmalloc that records the
 459  * calling function of the routine calling it for slab leak tracking instead
 460  * of just the calling function (confusing, eh?).
 461  * It's useful when the call to kmalloc comes from a widely-used standard
 462  * allocator where we care about the real place the memory allocation
 463  * request comes from.
 464  */
 465 #if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_SLUB) || \
 466         (defined(CONFIG_SLAB) && defined(CONFIG_TRACING)) || \
 467         (defined(CONFIG_SLOB) && defined(CONFIG_TRACING))
 468 extern void *__kmalloc_track_caller(size_t, gfp_t, unsigned long);
 469 #define kmalloc_track_caller(size, flags) \
 470         __kmalloc_track_caller(size, flags, _RET_IP_)
 471 #else
 472 #define kmalloc_track_caller(size, flags) \
 473         __kmalloc(size, flags)
 474 #endif /* DEBUG_SLAB */
 475
 476 #ifdef CONFIG_NUMA
 477 /*
 478  * kmalloc_node_track_caller is a special version of kmalloc_node that
 479  * records the calling function of the routine calling it for slab leak
 480  * tracking instead of just the calling function (confusing, eh?).
 481  * It's useful when the call to kmalloc_node comes from a widely-used
 482  * standard allocator where we care about the real place the memory
 483  * allocation request comes from.
 484  */
 485 #if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_SLUB) || \
 486         (defined(CONFIG_SLAB) && defined(CONFIG_TRACING)) || \
 487         (defined(CONFIG_SLOB) && defined(CONFIG_TRACING))
 488 extern void *__kmalloc_node_track_caller(size_t, gfp_t, int, unsigned long);
 489 #define kmalloc_node_track_caller(size, flags, node) \
 490         __kmalloc_node_track_caller(size, flags, node, \
 491                         _RET_IP_)
 492 #else
 493 #define kmalloc_node_track_caller(size, flags, node) \
 494         __kmalloc_node(size, flags, node)
 495 #endif
 496
 497 #else /* CONFIG_NUMA */
 498
 499 #define kmalloc_node_track_caller(size, flags, node) \
 500         kmalloc_track_caller(size, flags)
 501
 502 #endif /* CONFIG_NUMA */
 503
 504 /*
 505  * Shortcuts
 506  */
 507 static inline void *kmem_cache_zalloc(struct kmem_cache *k, gfp_t flags)
 508 {
 509         return kmem_cache_alloc(k, flags | __GFP_ZERO);
 510 }
 511
 512 /**
 513  * kzalloc - allocate memory. The memory is set to zero.
 514  * @size: how many bytes of memory are required.
 515  * @flags: the type of memory to allocate (see kmalloc).
 516  */
 517 static inline void *kzalloc(size_t size, gfp_t flags)
 518 {
 519         return kmalloc(size, flags | __GFP_ZERO);
 520 }
 521
 522 /**
 523  * kzalloc_node - allocate zeroed memory from a particular memory node.
 524  * @size: how many bytes of memory are required.
 525  * @flags: the type of memory to allocate (see kmalloc).
 526  * @node: memory node from which to allocate
 527  */
 528 static inline void *kzalloc_node(size_t size, gfp_t flags, int node)
 529 {
 530         return kmalloc_node(size, flags | __GFP_ZERO, node);
 531 }
 532
 533 /*
 534  * Determine the size of a slab object
 535  */
 536 static inline unsigned int kmem_cache_size(struct kmem_cache *s)
 537 {
 538         return s->object_size;
 539 }
 540
 541 void __init kmem_cache_init_late(void);
 542
 543 #endif  /* _LINUX_SLAB_H */