mm/slab.h

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 #ifndef MM_SLAB_H
   3 #define MM_SLAB_H
   4 /*
   5  * Internal slab definitions
   6  */
   7
   8 #ifdef CONFIG_SLOB
   9 /*
  10  * Common fields provided in kmem_cache by all slab allocators
  11  * This struct is either used directly by the allocator (SLOB)
  12  * or the allocator must include definitions for all fields
  13  * provided in kmem_cache_common in their definition of kmem_cache.
  14  *
  15  * Once we can do anonymous structs (C11 standard) we could put a
  16  * anonymous struct definition in these allocators so that the
  17  * separate allocations in the kmem_cache structure of SLAB and
  18  * SLUB is no longer needed.
  19  */
  20 struct kmem_cache {
  21         unsigned int object_size;/* The original size of the object */
  22         unsigned int size;      /* The aligned/padded/added on size  */
  23         unsigned int align;     /* Alignment as calculated */
  24         slab_flags_t flags;     /* Active flags on the slab */
  25         unsigned int useroffset;/* Usercopy region offset */
  26         unsigned int usersize;  /* Usercopy region size */
  27         const char *name;       /* Slab name for sysfs */
  28         int refcount;           /* Use counter */
  29         void (*ctor)(void *);   /* Called on object slot creation */
  30         struct list_head list;  /* List of all slab caches on the system */
  31 };
  32
  33 #endif /* CONFIG_SLOB */
  34
  35 #ifdef CONFIG_SLAB
  36 #include <linux/slab_def.h>
  37 #endif
  38
  39 #ifdef CONFIG_SLUB
  40 #include <linux/slub_def.h>
  41 #endif
  42
  43 #include <linux/memcontrol.h>
  44 #include <linux/fault-inject.h>
  45 #include <linux/kasan.h>
  46 #include <linux/kmemleak.h>
  47 #include <linux/random.h>
  48 #include <linux/sched/mm.h>
  49
  50 /*
  51  * State of the slab allocator.
  52  *
  53  * This is used to describe the states of the allocator during bootup.
  54  * Allocators use this to gradually bootstrap themselves. Most allocators
  55  * have the problem that the structures used for managing slab caches are
  56  * allocated from slab caches themselves.
  57  */
  58 enum slab_state {
  59         DOWN,                   /* No slab functionality yet */
  60         PARTIAL,                /* SLUB: kmem_cache_node available */
  61         PARTIAL_NODE,           /* SLAB: kmalloc size for node struct available */
  62         UP,                     /* Slab caches usable but not all extras yet */
  63         FULL                    /* Everything is working */
  64 };
  65
  66 extern enum slab_state slab_state;
  67
  68 /* The slab cache mutex protects the management structures during changes */
  69 extern struct mutex slab_mutex;
  70
  71 /* The list of all slab caches on the system */
  72 extern struct list_head slab_caches;
  73
  74 /* The slab cache that manages slab cache information */
  75 extern struct kmem_cache *kmem_cache;
  76
  77 /* A table of kmalloc cache names and sizes */
  78 extern const struct kmalloc_info_struct {
  79         const char *name;
  80         unsigned int size;
  81 } kmalloc_info[];
  82
  83 #ifndef CONFIG_SLOB
  84 /* Kmalloc array related functions */
  85 void setup_kmalloc_cache_index_table(void);
  86 void create_kmalloc_caches(slab_flags_t);
  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 int __kmem_cache_create(struct kmem_cache *, slab_flags_t flags);
  95
  96 struct kmem_cache *create_kmalloc_cache(const char *name, unsigned int size,
  97                         slab_flags_t flags, unsigned int useroffset,
  98                         unsigned int usersize);
  99 extern void create_boot_cache(struct kmem_cache *, const char *name,
 100                         unsigned int size, slab_flags_t flags,
 101                         unsigned int useroffset, unsigned int usersize);
 102
 103 int slab_unmergeable(struct kmem_cache *s);
 104 struct kmem_cache *find_mergeable(unsigned size, unsigned align,
 105                 slab_flags_t flags, const char *name, void (*ctor)(void *));
 106 #ifndef CONFIG_SLOB
 107 struct kmem_cache *
 108 __kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
 109                    slab_flags_t flags, void (*ctor)(void *));
 110
 111 slab_flags_t kmem_cache_flags(unsigned int object_size,
 112         slab_flags_t flags, const char *name,
 113         void (*ctor)(void *));
 114 #else
 115 static inline struct kmem_cache *
 116 __kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
 117                    slab_flags_t flags, void (*ctor)(void *))
 118 { return NULL; }
 119
 120 static inline slab_flags_t kmem_cache_flags(unsigned int object_size,
 121         slab_flags_t flags, const char *name,
 122         void (*ctor)(void *))
 123 {
 124         return flags;
 125 }
 126 #endif
 127
 128
 129 /* Legal flag mask for kmem_cache_create(), for various configurations */
 130 #define SLAB_CORE_FLAGS (SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA | \
 131                          SLAB_CACHE_DMA32 | SLAB_PANIC | \
 132                          SLAB_TYPESAFE_BY_RCU | SLAB_DEBUG_OBJECTS )
 133
 134 #if defined(CONFIG_DEBUG_SLAB)
 135 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
 136 #elif defined(CONFIG_SLUB_DEBUG)
 137 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
 138                           SLAB_TRACE | SLAB_CONSISTENCY_CHECKS)
 139 #else
 140 #define SLAB_DEBUG_FLAGS (0)
 141 #endif
 142
 143 #if defined(CONFIG_SLAB)
 144 #define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \
 145                           SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \
 146                           SLAB_ACCOUNT)
 147 #elif defined(CONFIG_SLUB)
 148 #define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
 149                           SLAB_TEMPORARY | SLAB_ACCOUNT)
 150 #else
 151 #define SLAB_CACHE_FLAGS (0)
 152 #endif
 153
 154 /* Common flags available with current configuration */
 155 #define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
 156
 157 /* Common flags permitted for kmem_cache_create */
 158 #define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \
 159                               SLAB_RED_ZONE | \
 160                               SLAB_POISON | \
 161                               SLAB_STORE_USER | \
 162                               SLAB_TRACE | \
 163                               SLAB_CONSISTENCY_CHECKS | \
 164                               SLAB_MEM_SPREAD | \
 165                               SLAB_NOLEAKTRACE | \
 166                               SLAB_RECLAIM_ACCOUNT | \
 167                               SLAB_TEMPORARY | \
 168                               SLAB_ACCOUNT)
 169
 170 bool __kmem_cache_empty(struct kmem_cache *);
 171 int __kmem_cache_shutdown(struct kmem_cache *);
 172 void __kmem_cache_release(struct kmem_cache *);
 173 int __kmem_cache_shrink(struct kmem_cache *);
 174 void __kmemcg_cache_deactivate(struct kmem_cache *s);
 175 void slab_kmem_cache_release(struct kmem_cache *);
 176
 177 struct seq_file;
 178 struct file;
 179
 180 struct slabinfo {
 181         unsigned long active_objs;
 182         unsigned long num_objs;
 183         unsigned long active_slabs;
 184         unsigned long num_slabs;
 185         unsigned long shared_avail;
 186         unsigned int limit;
 187         unsigned int batchcount;
 188         unsigned int shared;
 189         unsigned int objects_per_slab;
 190         unsigned int cache_order;
 191 };
 192
 193 void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo);
 194 void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s);
 195 ssize_t slabinfo_write(struct file *file, const char __user *buffer,
 196                        size_t count, loff_t *ppos);
 197
 198 /*
 199  * Generic implementation of bulk operations
 200  * These are useful for situations in which the allocator cannot
 201  * perform optimizations. In that case segments of the object listed
 202  * may be allocated or freed using these operations.
 203  */
 204 void __kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
 205 int __kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
 206
 207 #ifdef CONFIG_MEMCG_KMEM
 208
 209 /* List of all root caches. */
 210 extern struct list_head         slab_root_caches;
 211 #define root_caches_node        memcg_params.__root_caches_node
 212
 213 /*
 214  * Iterate over all memcg caches of the given root cache. The caller must hold
 215  * slab_mutex.
 216  */
 217 #define for_each_memcg_cache(iter, root) \
 218         list_for_each_entry(iter, &(root)->memcg_params.children, \
 219                             memcg_params.children_node)
 220
 221 static inline bool is_root_cache(struct kmem_cache *s)
 222 {
 223         return !s->memcg_params.root_cache;
 224 }
 225
 226 static inline bool slab_equal_or_root(struct kmem_cache *s,
 227                                       struct kmem_cache *p)
 228 {
 229         return p == s || p == s->memcg_params.root_cache;
 230 }
 231
 232 /*
 233  * We use suffixes to the name in memcg because we can't have caches
 234  * created in the system with the same name. But when we print them
 235  * locally, better refer to them with the base name
 236  */
 237 static inline const char *cache_name(struct kmem_cache *s)
 238 {
 239         if (!is_root_cache(s))
 240                 s = s->memcg_params.root_cache;
 241         return s->name;
 242 }
 243
 244 /*
 245  * Note, we protect with RCU only the memcg_caches array, not per-memcg caches.
 246  * That said the caller must assure the memcg's cache won't go away by either
 247  * taking a css reference to the owner cgroup, or holding the slab_mutex.
 248  */
 249 static inline struct kmem_cache *
 250 cache_from_memcg_idx(struct kmem_cache *s, int idx)
 251 {
 252         struct kmem_cache *cachep;
 253         struct memcg_cache_array *arr;
 254
 255         rcu_read_lock();
 256         arr = rcu_dereference(s->memcg_params.memcg_caches);
 257
 258         /*
 259          * Make sure we will access the up-to-date value. The code updating
 260          * memcg_caches issues a write barrier to match this (see
 261          * memcg_create_kmem_cache()).
 262          */
 263         cachep = READ_ONCE(arr->entries[idx]);
 264         rcu_read_unlock();
 265
 266         return cachep;
 267 }
 268
 269 static inline struct kmem_cache *memcg_root_cache(struct kmem_cache *s)
 270 {
 271         if (is_root_cache(s))
 272                 return s;
 273         return s->memcg_params.root_cache;
 274 }
 275
 276 static __always_inline int memcg_charge_slab(struct page *page,
 277                                              gfp_t gfp, int order,
 278                                              struct kmem_cache *s)
 279 {
 280         if (is_root_cache(s))
 281                 return 0;
 282         return memcg_kmem_charge_memcg(page, gfp, order, s->memcg_params.memcg);
 283 }
 284
 285 static __always_inline void memcg_uncharge_slab(struct page *page, int order,
 286                                                 struct kmem_cache *s)
 287 {
 288         memcg_kmem_uncharge(page, order);
 289 }
 290
 291 extern void slab_init_memcg_params(struct kmem_cache *);
 292 extern void memcg_link_cache(struct kmem_cache *s, struct mem_cgroup *memcg);
 293 extern void slab_deactivate_memcg_cache_rcu_sched(struct kmem_cache *s,
 294                                 void (*work_fn)(struct kmem_cache *));
 295
 296 #else /* CONFIG_MEMCG_KMEM */
 297
 298 /* If !memcg, all caches are root. */
 299 #define slab_root_caches        slab_caches
 300 #define root_caches_node        list
 301
 302 #define for_each_memcg_cache(iter, root) \
 303         for ((void)(iter), (void)(root); 0; )
 304
 305 static inline bool is_root_cache(struct kmem_cache *s)
 306 {
 307         return true;
 308 }
 309
 310 static inline bool slab_equal_or_root(struct kmem_cache *s,
 311                                       struct kmem_cache *p)
 312 {
 313         return s == p;
 314 }
 315
 316 static inline const char *cache_name(struct kmem_cache *s)
 317 {
 318         return s->name;
 319 }
 320
 321 static inline struct kmem_cache *
 322 cache_from_memcg_idx(struct kmem_cache *s, int idx)
 323 {
 324         return NULL;
 325 }
 326
 327 static inline struct kmem_cache *memcg_root_cache(struct kmem_cache *s)
 328 {
 329         return s;
 330 }
 331
 332 static inline int memcg_charge_slab(struct page *page, gfp_t gfp, int order,
 333                                     struct kmem_cache *s)
 334 {
 335         return 0;
 336 }
 337
 338 static inline void memcg_uncharge_slab(struct page *page, int order,
 339                                        struct kmem_cache *s)
 340 {
 341 }
 342
 343 static inline void slab_init_memcg_params(struct kmem_cache *s)
 344 {
 345 }
 346
 347 static inline void memcg_link_cache(struct kmem_cache *s,
 348                                     struct mem_cgroup *memcg)
 349 {
 350 }
 351
 352 #endif /* CONFIG_MEMCG_KMEM */
 353
 354 static inline struct kmem_cache *virt_to_cache(const void *obj)
 355 {
 356         struct page *page;
 357
 358         page = virt_to_head_page(obj);
 359         if (WARN_ONCE(!PageSlab(page), "%s: Object is not a Slab page!\n",
 360                                         __func__))
 361                 return NULL;
 362         return page->slab_cache;
 363 }
 364
 365 static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x)
 366 {
 367         struct kmem_cache *cachep;
 368
 369         /*
 370          * When kmemcg is not being used, both assignments should return the
 371          * same value. but we don't want to pay the assignment price in that
 372          * case. If it is not compiled in, the compiler should be smart enough
 373          * to not do even the assignment. In that case, slab_equal_or_root
 374          * will also be a constant.
 375          */
 376         if (!memcg_kmem_enabled() &&
 377             !IS_ENABLED(CONFIG_SLAB_FREELIST_HARDENED) &&
 378             !unlikely(s->flags & SLAB_CONSISTENCY_CHECKS))
 379                 return s;
 380
 381         cachep = virt_to_cache(x);
 382         WARN_ONCE(cachep && !slab_equal_or_root(cachep, s),
 383                   "%s: Wrong slab cache. %s but object is from %s\n",
 384                   __func__, s->name, cachep->name);
 385         return cachep;
 386 }
 387
 388 static inline size_t slab_ksize(const struct kmem_cache *s)
 389 {
 390 #ifndef CONFIG_SLUB
 391         return s->object_size;
 392
 393 #else /* CONFIG_SLUB */
 394 # ifdef CONFIG_SLUB_DEBUG
 395         /*
 396          * Debugging requires use of the padding between object
 397          * and whatever may come after it.
 398          */
 399         if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
 400                 return s->object_size;
 401 # endif
 402         if (s->flags & SLAB_KASAN)
 403                 return s->object_size;
 404         /*
 405          * If we have the need to store the freelist pointer
 406          * back there or track user information then we can
 407          * only use the space before that information.
 408          */
 409         if (s->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_STORE_USER))
 410                 return s->inuse;
 411         /*
 412          * Else we can use all the padding etc for the allocation
 413          */
 414         return s->size;
 415 #endif
 416 }
 417
 418 static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s,
 419                                                      gfp_t flags)
 420 {
 421         flags &= gfp_allowed_mask;
 422
 423         fs_reclaim_acquire(flags);
 424         fs_reclaim_release(flags);
 425
 426         might_sleep_if(gfpflags_allow_blocking(flags));
 427
 428         if (should_failslab(s, flags))
 429                 return NULL;
 430
 431         if (memcg_kmem_enabled() &&
 432             ((flags & __GFP_ACCOUNT) || (s->flags & SLAB_ACCOUNT)))
 433                 return memcg_kmem_get_cache(s);
 434
 435         return s;
 436 }
 437
 438 static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags,
 439                                         size_t size, void **p)
 440 {
 441         size_t i;
 442
 443         flags &= gfp_allowed_mask;
 444         for (i = 0; i < size; i++) {
 445                 p[i] = kasan_slab_alloc(s, p[i], flags);
 446                 /* As p[i] might get tagged, call kmemleak hook after KASAN. */
 447                 kmemleak_alloc_recursive(p[i], s->object_size, 1,
 448                                          s->flags, flags);
 449         }
 450
 451         if (memcg_kmem_enabled())
 452                 memcg_kmem_put_cache(s);
 453 }
 454
 455 #ifndef CONFIG_SLOB
 456 /*
 457  * The slab lists for all objects.
 458  */
 459 struct kmem_cache_node {
 460         spinlock_t list_lock;
 461
 462 #ifdef CONFIG_SLAB
 463         struct list_head slabs_partial; /* partial list first, better asm code */
 464         struct list_head slabs_full;
 465         struct list_head slabs_free;
 466         unsigned long total_slabs;      /* length of all slab lists */
 467         unsigned long free_slabs;       /* length of free slab list only */
 468         unsigned long free_objects;
 469         unsigned int free_limit;
 470         unsigned int colour_next;       /* Per-node cache coloring */
 471         struct array_cache *shared;     /* shared per node */
 472         struct alien_cache **alien;     /* on other nodes */
 473         unsigned long next_reap;        /* updated without locking */
 474         int free_touched;               /* updated without locking */
 475 #endif
 476
 477 #ifdef CONFIG_SLUB
 478         unsigned long nr_partial;
 479         struct list_head partial;
 480 #ifdef CONFIG_SLUB_DEBUG
 481         atomic_long_t nr_slabs;
 482         atomic_long_t total_objects;
 483         struct list_head full;
 484 #endif
 485 #endif
 486
 487 };
 488
 489 static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
 490 {
 491         return s->node[node];
 492 }
 493
 494 /*
 495  * Iterator over all nodes. The body will be executed for each node that has
 496  * a kmem_cache_node structure allocated (which is true for all online nodes)
 497  */
 498 #define for_each_kmem_cache_node(__s, __node, __n) \
 499         for (__node = 0; __node < nr_node_ids; __node++) \
 500                  if ((__n = get_node(__s, __node)))
 501
 502 #endif
 503
 504 void *slab_start(struct seq_file *m, loff_t *pos);
 505 void *slab_next(struct seq_file *m, void *p, loff_t *pos);
 506 void slab_stop(struct seq_file *m, void *p);
 507 void *memcg_slab_start(struct seq_file *m, loff_t *pos);
 508 void *memcg_slab_next(struct seq_file *m, void *p, loff_t *pos);
 509 void memcg_slab_stop(struct seq_file *m, void *p);
 510 int memcg_slab_show(struct seq_file *m, void *p);
 511
 512 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
 513 void dump_unreclaimable_slab(void);
 514 #else
 515 static inline void dump_unreclaimable_slab(void)
 516 {
 517 }
 518 #endif
 519
 520 void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr);
 521
 522 #ifdef CONFIG_SLAB_FREELIST_RANDOM
 523 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
 524                         gfp_t gfp);
 525 void cache_random_seq_destroy(struct kmem_cache *cachep);
 526 #else
 527 static inline int cache_random_seq_create(struct kmem_cache *cachep,
 528                                         unsigned int count, gfp_t gfp)
 529 {
 530         return 0;
 531 }
 532 static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { }
 533 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
 534
 535 #endif /* MM_SLAB_H */