mm/slab.h

   1 #ifndef MM_SLAB_H
   2 #define MM_SLAB_H
   3 /*
   4  * Internal slab definitions
   5  */
   6
   7 #ifdef CONFIG_SLOB
   8 /*
   9  * Common fields provided in kmem_cache by all slab allocators
  10  * This struct is either used directly by the allocator (SLOB)
  11  * or the allocator must include definitions for all fields
  12  * provided in kmem_cache_common in their definition of kmem_cache.
  13  *
  14  * Once we can do anonymous structs (C11 standard) we could put a
  15  * anonymous struct definition in these allocators so that the
  16  * separate allocations in the kmem_cache structure of SLAB and
  17  * SLUB is no longer needed.
  18  */
  19 struct kmem_cache {
  20         unsigned int object_size;/* The original size of the object */
  21         unsigned int size;      /* The aligned/padded/added on size  */
  22         unsigned int align;     /* Alignment as calculated */
  23         unsigned long flags;    /* Active flags on the slab */
  24         const char *name;       /* Slab name for sysfs */
  25         int refcount;           /* Use counter */
  26         void (*ctor)(void *);   /* Called on object slot creation */
  27         struct list_head list;  /* List of all slab caches on the system */
  28 };
  29
  30 #endif /* CONFIG_SLOB */
  31
  32 #ifdef CONFIG_SLAB
  33 #include <linux/slab_def.h>
  34 #endif
  35
  36 #ifdef CONFIG_SLUB
  37 #include <linux/slub_def.h>
  38 #endif
  39
  40 #include <linux/memcontrol.h>
  41 #include <linux/fault-inject.h>
  42 #include <linux/kmemcheck.h>
  43 #include <linux/kasan.h>
  44 #include <linux/kmemleak.h>
  45 #include <linux/random.h>
  46
  47 /*
  48  * State of the slab allocator.
  49  *
  50  * This is used to describe the states of the allocator during bootup.
  51  * Allocators use this to gradually bootstrap themselves. Most allocators
  52  * have the problem that the structures used for managing slab caches are
  53  * allocated from slab caches themselves.
  54  */
  55 enum slab_state {
  56         DOWN,                   /* No slab functionality yet */
  57         PARTIAL,                /* SLUB: kmem_cache_node available */
  58         PARTIAL_NODE,           /* SLAB: kmalloc size for node struct available */
  59         UP,                     /* Slab caches usable but not all extras yet */
  60         FULL                    /* Everything is working */
  61 };
  62
  63 extern enum slab_state slab_state;
  64
  65 /* The slab cache mutex protects the management structures during changes */
  66 extern struct mutex slab_mutex;
  67
  68 /* The list of all slab caches on the system */
  69 extern struct list_head slab_caches;
  70
  71 /* The slab cache that manages slab cache information */
  72 extern struct kmem_cache *kmem_cache;
  73
  74 /* A table of kmalloc cache names and sizes */
  75 extern const struct kmalloc_info_struct {
  76         const char *name;
  77         unsigned long size;
  78 } kmalloc_info[];
  79
  80 unsigned long calculate_alignment(unsigned long flags,
  81                 unsigned long align, unsigned long size);
  82
  83 #ifndef CONFIG_SLOB
  84 /* Kmalloc array related functions */
  85 void setup_kmalloc_cache_index_table(void);
  86 void create_kmalloc_caches(unsigned long);
  87
  88 /* Find the kmalloc slab corresponding for a certain size */
  89 struct kmem_cache *kmalloc_slab(size_t, gfp_t);
  90 #endif
  91
  92
  93 /* Functions provided by the slab allocators */
  94 extern int __kmem_cache_create(struct kmem_cache *, unsigned long flags);
  95
  96 extern struct kmem_cache *create_kmalloc_cache(const char *name, size_t size,
  97                         unsigned long flags);
  98 extern void create_boot_cache(struct kmem_cache *, const char *name,
  99                         size_t size, unsigned long flags);
 100
 101 int slab_unmergeable(struct kmem_cache *s);
 102 struct kmem_cache *find_mergeable(size_t size, size_t align,
 103                 unsigned long flags, const char *name, void (*ctor)(void *));
 104 #ifndef CONFIG_SLOB
 105 struct kmem_cache *
 106 __kmem_cache_alias(const char *name, size_t size, size_t align,
 107                    unsigned long flags, void (*ctor)(void *));
 108
 109 unsigned long kmem_cache_flags(unsigned long object_size,
 110         unsigned long flags, const char *name,
 111         void (*ctor)(void *));
 112 #else
 113 static inline struct kmem_cache *
 114 __kmem_cache_alias(const char *name, size_t size, size_t align,
 115                    unsigned long flags, void (*ctor)(void *))
 116 { return NULL; }
 117
 118 static inline unsigned long kmem_cache_flags(unsigned long object_size,
 119         unsigned long flags, const char *name,
 120         void (*ctor)(void *))
 121 {
 122         return flags;
 123 }
 124 #endif
 125
 126
 127 /* Legal flag mask for kmem_cache_create(), for various configurations */
 128 #define SLAB_CORE_FLAGS (SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA | SLAB_PANIC | \
 129                          SLAB_DESTROY_BY_RCU | SLAB_DEBUG_OBJECTS )
 130
 131 #if defined(CONFIG_DEBUG_SLAB)
 132 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
 133 #elif defined(CONFIG_SLUB_DEBUG)
 134 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
 135                           SLAB_TRACE | SLAB_CONSISTENCY_CHECKS)
 136 #else
 137 #define SLAB_DEBUG_FLAGS (0)
 138 #endif
 139
 140 #if defined(CONFIG_SLAB)
 141 #define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \
 142                           SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \
 143                           SLAB_NOTRACK | SLAB_ACCOUNT)
 144 #elif defined(CONFIG_SLUB)
 145 #define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
 146                           SLAB_TEMPORARY | SLAB_NOTRACK | SLAB_ACCOUNT)
 147 #else
 148 #define SLAB_CACHE_FLAGS (0)
 149 #endif
 150
 151 /* Common flags available with current configuration */
 152 #define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
 153
 154 /* Common flags permitted for kmem_cache_create */
 155 #define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \
 156                               SLAB_RED_ZONE | \
 157                               SLAB_POISON | \
 158                               SLAB_STORE_USER | \
 159                               SLAB_TRACE | \
 160                               SLAB_CONSISTENCY_CHECKS | \
 161                               SLAB_MEM_SPREAD | \
 162                               SLAB_NOLEAKTRACE | \
 163                               SLAB_RECLAIM_ACCOUNT | \
 164                               SLAB_TEMPORARY | \
 165                               SLAB_NOTRACK | \
 166                               SLAB_ACCOUNT)
 167
 168 int __kmem_cache_shutdown(struct kmem_cache *);
 169 void __kmem_cache_release(struct kmem_cache *);
 170 int __kmem_cache_shrink(struct kmem_cache *, bool);
 171 void slab_kmem_cache_release(struct kmem_cache *);
 172
 173 struct seq_file;
 174 struct file;
 175
 176 struct slabinfo {
 177         unsigned long active_objs;
 178         unsigned long num_objs;
 179         unsigned long active_slabs;
 180         unsigned long num_slabs;
 181         unsigned long shared_avail;
 182         unsigned int limit;
 183         unsigned int batchcount;
 184         unsigned int shared;
 185         unsigned int objects_per_slab;
 186         unsigned int cache_order;
 187 };
 188
 189 void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo);
 190 void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s);
 191 ssize_t slabinfo_write(struct file *file, const char __user *buffer,
 192                        size_t count, loff_t *ppos);
 193
 194 /*
 195  * Generic implementation of bulk operations
 196  * These are useful for situations in which the allocator cannot
 197  * perform optimizations. In that case segments of the object listed
 198  * may be allocated or freed using these operations.
 199  */
 200 void __kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
 201 int __kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
 202
 203 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
 204 /*
 205  * Iterate over all memcg caches of the given root cache. The caller must hold
 206  * slab_mutex.
 207  */
 208 #define for_each_memcg_cache(iter, root) \
 209         list_for_each_entry(iter, &(root)->memcg_params.children, \
 210                             memcg_params.children_node)
 211
 212 static inline bool is_root_cache(struct kmem_cache *s)
 213 {
 214         return !s->memcg_params.root_cache;
 215 }
 216
 217 static inline bool slab_equal_or_root(struct kmem_cache *s,
 218                                       struct kmem_cache *p)
 219 {
 220         return p == s || p == s->memcg_params.root_cache;
 221 }
 222
 223 /*
 224  * We use suffixes to the name in memcg because we can't have caches
 225  * created in the system with the same name. But when we print them
 226  * locally, better refer to them with the base name
 227  */
 228 static inline const char *cache_name(struct kmem_cache *s)
 229 {
 230         if (!is_root_cache(s))
 231                 s = s->memcg_params.root_cache;
 232         return s->name;
 233 }
 234
 235 /*
 236  * Note, we protect with RCU only the memcg_caches array, not per-memcg caches.
 237  * That said the caller must assure the memcg's cache won't go away by either
 238  * taking a css reference to the owner cgroup, or holding the slab_mutex.
 239  */
 240 static inline struct kmem_cache *
 241 cache_from_memcg_idx(struct kmem_cache *s, int idx)
 242 {
 243         struct kmem_cache *cachep;
 244         struct memcg_cache_array *arr;
 245
 246         rcu_read_lock();
 247         arr = rcu_dereference(s->memcg_params.memcg_caches);
 248
 249         /*
 250          * Make sure we will access the up-to-date value. The code updating
 251          * memcg_caches issues a write barrier to match this (see
 252          * memcg_create_kmem_cache()).
 253          */
 254         cachep = lockless_dereference(arr->entries[idx]);
 255         rcu_read_unlock();
 256
 257         return cachep;
 258 }
 259
 260 static inline struct kmem_cache *memcg_root_cache(struct kmem_cache *s)
 261 {
 262         if (is_root_cache(s))
 263                 return s;
 264         return s->memcg_params.root_cache;
 265 }
 266
 267 static __always_inline int memcg_charge_slab(struct page *page,
 268                                              gfp_t gfp, int order,
 269                                              struct kmem_cache *s)
 270 {
 271         int ret;
 272
 273         if (!memcg_kmem_enabled())
 274                 return 0;
 275         if (is_root_cache(s))
 276                 return 0;
 277
 278         ret = memcg_kmem_charge_memcg(page, gfp, order, s->memcg_params.memcg);
 279         if (ret)
 280                 return ret;
 281
 282         memcg_kmem_update_page_stat(page,
 283                         (s->flags & SLAB_RECLAIM_ACCOUNT) ?
 284                         MEMCG_SLAB_RECLAIMABLE : MEMCG_SLAB_UNRECLAIMABLE,
 285                         1 << order);
 286         return 0;
 287 }
 288
 289 static __always_inline void memcg_uncharge_slab(struct page *page, int order,
 290                                                 struct kmem_cache *s)
 291 {
 292         if (!memcg_kmem_enabled())
 293                 return;
 294
 295         memcg_kmem_update_page_stat(page,
 296                         (s->flags & SLAB_RECLAIM_ACCOUNT) ?
 297                         MEMCG_SLAB_RECLAIMABLE : MEMCG_SLAB_UNRECLAIMABLE,
 298                         -(1 << order));
 299         memcg_kmem_uncharge(page, order);
 300 }
 301
 302 extern void slab_init_memcg_params(struct kmem_cache *);
 303
 304 #else /* CONFIG_MEMCG && !CONFIG_SLOB */
 305
 306 #define for_each_memcg_cache(iter, root) \
 307         for ((void)(iter), (void)(root); 0; )
 308
 309 static inline bool is_root_cache(struct kmem_cache *s)
 310 {
 311         return true;
 312 }
 313
 314 static inline bool slab_equal_or_root(struct kmem_cache *s,
 315                                       struct kmem_cache *p)
 316 {
 317         return true;
 318 }
 319
 320 static inline const char *cache_name(struct kmem_cache *s)
 321 {
 322         return s->name;
 323 }
 324
 325 static inline struct kmem_cache *
 326 cache_from_memcg_idx(struct kmem_cache *s, int idx)
 327 {
 328         return NULL;
 329 }
 330
 331 static inline struct kmem_cache *memcg_root_cache(struct kmem_cache *s)
 332 {
 333         return s;
 334 }
 335
 336 static inline int memcg_charge_slab(struct page *page, gfp_t gfp, int order,
 337                                     struct kmem_cache *s)
 338 {
 339         return 0;
 340 }
 341
 342 static inline void memcg_uncharge_slab(struct page *page, int order,
 343                                        struct kmem_cache *s)
 344 {
 345 }
 346
 347 static inline void slab_init_memcg_params(struct kmem_cache *s)
 348 {
 349 }
 350 #endif /* CONFIG_MEMCG && !CONFIG_SLOB */
 351
 352 static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x)
 353 {
 354         struct kmem_cache *cachep;
 355         struct page *page;
 356
 357         /*
 358          * When kmemcg is not being used, both assignments should return the
 359          * same value. but we don't want to pay the assignment price in that
 360          * case. If it is not compiled in, the compiler should be smart enough
 361          * to not do even the assignment. In that case, slab_equal_or_root
 362          * will also be a constant.
 363          */
 364         if (!memcg_kmem_enabled() &&
 365             !unlikely(s->flags & SLAB_CONSISTENCY_CHECKS))
 366                 return s;
 367
 368         page = virt_to_head_page(x);
 369         cachep = page->slab_cache;
 370         if (slab_equal_or_root(cachep, s))
 371                 return cachep;
 372
 373         pr_err("%s: Wrong slab cache. %s but object is from %s\n",
 374                __func__, s->name, cachep->name);
 375         WARN_ON_ONCE(1);
 376         return s;
 377 }
 378
 379 static inline size_t slab_ksize(const struct kmem_cache *s)
 380 {
 381 #ifndef CONFIG_SLUB
 382         return s->object_size;
 383
 384 #else /* CONFIG_SLUB */
 385 # ifdef CONFIG_SLUB_DEBUG
 386         /*
 387          * Debugging requires use of the padding between object
 388          * and whatever may come after it.
 389          */
 390         if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
 391                 return s->object_size;
 392 # endif
 393         if (s->flags & SLAB_KASAN)
 394                 return s->object_size;
 395         /*
 396          * If we have the need to store the freelist pointer
 397          * back there or track user information then we can
 398          * only use the space before that information.
 399          */
 400         if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER))
 401                 return s->inuse;
 402         /*
 403          * Else we can use all the padding etc for the allocation
 404          */
 405         return s->size;
 406 #endif
 407 }
 408
 409 static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s,
 410                                                      gfp_t flags)
 411 {
 412         flags &= gfp_allowed_mask;
 413         lockdep_trace_alloc(flags);
 414         might_sleep_if(gfpflags_allow_blocking(flags));
 415
 416         if (should_failslab(s, flags))
 417                 return NULL;
 418
 419         if (memcg_kmem_enabled() &&
 420             ((flags & __GFP_ACCOUNT) || (s->flags & SLAB_ACCOUNT)))
 421                 return memcg_kmem_get_cache(s);
 422
 423         return s;
 424 }
 425
 426 static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags,
 427                                         size_t size, void **p)
 428 {
 429         size_t i;
 430
 431         flags &= gfp_allowed_mask;
 432         for (i = 0; i < size; i++) {
 433                 void *object = p[i];
 434
 435                 kmemcheck_slab_alloc(s, flags, object, slab_ksize(s));
 436                 kmemleak_alloc_recursive(object, s->object_size, 1,
 437                                          s->flags, flags);
 438                 kasan_slab_alloc(s, object, flags);
 439         }
 440
 441         if (memcg_kmem_enabled())
 442                 memcg_kmem_put_cache(s);
 443 }
 444
 445 #ifndef CONFIG_SLOB
 446 /*
 447  * The slab lists for all objects.
 448  */
 449 struct kmem_cache_node {
 450         spinlock_t list_lock;
 451
 452 #ifdef CONFIG_SLAB
 453         struct list_head slabs_partial; /* partial list first, better asm code */
 454         struct list_head slabs_full;
 455         struct list_head slabs_free;
 456         unsigned long total_slabs;      /* length of all slab lists */
 457         unsigned long free_slabs;       /* length of free slab list only */
 458         unsigned long free_objects;
 459         unsigned int free_limit;
 460         unsigned int colour_next;       /* Per-node cache coloring */
 461         struct array_cache *shared;     /* shared per node */
 462         struct alien_cache **alien;     /* on other nodes */
 463         unsigned long next_reap;        /* updated without locking */
 464         int free_touched;               /* updated without locking */
 465 #endif
 466
 467 #ifdef CONFIG_SLUB
 468         unsigned long nr_partial;
 469         struct list_head partial;
 470 #ifdef CONFIG_SLUB_DEBUG
 471         atomic_long_t nr_slabs;
 472         atomic_long_t total_objects;
 473         struct list_head full;
 474 #endif
 475 #endif
 476
 477 };
 478
 479 static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
 480 {
 481         return s->node[node];
 482 }
 483
 484 /*
 485  * Iterator over all nodes. The body will be executed for each node that has
 486  * a kmem_cache_node structure allocated (which is true for all online nodes)
 487  */
 488 #define for_each_kmem_cache_node(__s, __node, __n) \
 489         for (__node = 0; __node < nr_node_ids; __node++) \
 490                  if ((__n = get_node(__s, __node)))
 491
 492 #endif
 493
 494 void *slab_start(struct seq_file *m, loff_t *pos);
 495 void *slab_next(struct seq_file *m, void *p, loff_t *pos);
 496 void slab_stop(struct seq_file *m, void *p);
 497 void *memcg_slab_start(struct seq_file *m, loff_t *pos);
 498 void *memcg_slab_next(struct seq_file *m, void *p, loff_t *pos);
 499 void memcg_slab_stop(struct seq_file *m, void *p);
 500 int memcg_slab_show(struct seq_file *m, void *p);
 501
 502 void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr);
 503
 504 #ifdef CONFIG_SLAB_FREELIST_RANDOM
 505 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
 506                         gfp_t gfp);
 507 void cache_random_seq_destroy(struct kmem_cache *cachep);
 508 #else
 509 static inline int cache_random_seq_create(struct kmem_cache *cachep,
 510                                         unsigned int count, gfp_t gfp)
 511 {
 512         return 0;
 513 }
 514 static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { }
 515 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
 516
 517 #endif /* MM_SLAB_H */