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         const char *name;       /* Slab name for sysfs */
  26         int refcount;           /* Use counter */
  27         void (*ctor)(void *);   /* Called on object slot creation */
  28         struct list_head list;  /* List of all slab caches on the system */
  29 };
  30
  31 #endif /* CONFIG_SLOB */
  32
  33 #ifdef CONFIG_SLAB
  34 #include <linux/slab_def.h>
  35 #endif
  36
  37 #ifdef CONFIG_SLUB
  38 #include <linux/slub_def.h>
  39 #endif
  40
  41 #include <linux/memcontrol.h>
  42 #include <linux/fault-inject.h>
  43 #include <linux/kasan.h>
  44 #include <linux/kmemleak.h>
  45 #include <linux/random.h>
  46 #include <linux/sched/mm.h>
  47
  48 /*
  49  * State of the slab allocator.
  50  *
  51  * This is used to describe the states of the allocator during bootup.
  52  * Allocators use this to gradually bootstrap themselves. Most allocators
  53  * have the problem that the structures used for managing slab caches are
  54  * allocated from slab caches themselves.
  55  */
  56 enum slab_state {
  57         DOWN,                   /* No slab functionality yet */
  58         PARTIAL,                /* SLUB: kmem_cache_node available */
  59         PARTIAL_NODE,           /* SLAB: kmalloc size for node struct available */
  60         UP,                     /* Slab caches usable but not all extras yet */
  61         FULL                    /* Everything is working */
  62 };
  63
  64 extern enum slab_state slab_state;
  65
  66 /* The slab cache mutex protects the management structures during changes */
  67 extern struct mutex slab_mutex;
  68
  69 /* The list of all slab caches on the system */
  70 extern struct list_head slab_caches;
  71
  72 /* The slab cache that manages slab cache information */
  73 extern struct kmem_cache *kmem_cache;
  74
  75 /* A table of kmalloc cache names and sizes */
  76 extern const struct kmalloc_info_struct {
  77         const char *name;
  78         unsigned long size;
  79 } kmalloc_info[];
  80
  81 unsigned long calculate_alignment(slab_flags_t flags,
  82                 unsigned long align, unsigned long size);
  83
  84 #ifndef CONFIG_SLOB
  85 /* Kmalloc array related functions */
  86 void setup_kmalloc_cache_index_table(void);
  87 void create_kmalloc_caches(slab_flags_t);
  88
  89 /* Find the kmalloc slab corresponding for a certain size */
  90 struct kmem_cache *kmalloc_slab(size_t, gfp_t);
  91 #endif
  92
  93
  94 /* Functions provided by the slab allocators */
  95 int __kmem_cache_create(struct kmem_cache *, slab_flags_t flags);
  96
  97 extern struct kmem_cache *create_kmalloc_cache(const char *name, size_t size,
  98                         slab_flags_t flags);
  99 extern void create_boot_cache(struct kmem_cache *, const char *name,
 100                         size_t size, slab_flags_t flags);
 101
 102 int slab_unmergeable(struct kmem_cache *s);
 103 struct kmem_cache *find_mergeable(size_t size, size_t align,
 104                 slab_flags_t flags, const char *name, void (*ctor)(void *));
 105 #ifndef CONFIG_SLOB
 106 struct kmem_cache *
 107 __kmem_cache_alias(const char *name, size_t size, size_t align,
 108                    slab_flags_t flags, void (*ctor)(void *));
 109
 110 slab_flags_t kmem_cache_flags(unsigned long object_size,
 111         slab_flags_t flags, const char *name,
 112         void (*ctor)(void *));
 113 #else
 114 static inline struct kmem_cache *
 115 __kmem_cache_alias(const char *name, size_t size, size_t align,
 116                    slab_flags_t flags, void (*ctor)(void *))
 117 { return NULL; }
 118
 119 static inline slab_flags_t kmem_cache_flags(unsigned long object_size,
 120         slab_flags_t flags, const char *name,
 121         void (*ctor)(void *))
 122 {
 123         return flags;
 124 }
 125 #endif
 126
 127
 128 /* Legal flag mask for kmem_cache_create(), for various configurations */
 129 #define SLAB_CORE_FLAGS (SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA | SLAB_PANIC | \
 130                          SLAB_TYPESAFE_BY_RCU | SLAB_DEBUG_OBJECTS )
 131
 132 #if defined(CONFIG_DEBUG_SLAB)
 133 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
 134 #elif defined(CONFIG_SLUB_DEBUG)
 135 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
 136                           SLAB_TRACE | SLAB_CONSISTENCY_CHECKS)
 137 #else
 138 #define SLAB_DEBUG_FLAGS (0)
 139 #endif
 140
 141 #if defined(CONFIG_SLAB)
 142 #define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \
 143                           SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \
 144                           SLAB_ACCOUNT)
 145 #elif defined(CONFIG_SLUB)
 146 #define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
 147                           SLAB_TEMPORARY | SLAB_ACCOUNT)
 148 #else
 149 #define SLAB_CACHE_FLAGS (0)
 150 #endif
 151
 152 /* Common flags available with current configuration */
 153 #define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
 154
 155 /* Common flags permitted for kmem_cache_create */
 156 #define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \
 157                               SLAB_RED_ZONE | \
 158                               SLAB_POISON | \
 159                               SLAB_STORE_USER | \
 160                               SLAB_TRACE | \
 161                               SLAB_CONSISTENCY_CHECKS | \
 162                               SLAB_MEM_SPREAD | \
 163                               SLAB_NOLEAKTRACE | \
 164                               SLAB_RECLAIM_ACCOUNT | \
 165                               SLAB_TEMPORARY | \
 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 *);
 171 void __kmemcg_cache_deactivate(struct kmem_cache *s);
 172 void slab_kmem_cache_release(struct kmem_cache *);
 173
 174 struct seq_file;
 175 struct file;
 176
 177 struct slabinfo {
 178         unsigned long active_objs;
 179         unsigned long num_objs;
 180         unsigned long active_slabs;
 181         unsigned long num_slabs;
 182         unsigned long shared_avail;
 183         unsigned int limit;
 184         unsigned int batchcount;
 185         unsigned int shared;
 186         unsigned int objects_per_slab;
 187         unsigned int cache_order;
 188 };
 189
 190 void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo);
 191 void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s);
 192 ssize_t slabinfo_write(struct file *file, const char __user *buffer,
 193                        size_t count, loff_t *ppos);
 194
 195 /*
 196  * Generic implementation of bulk operations
 197  * These are useful for situations in which the allocator cannot
 198  * perform optimizations. In that case segments of the object listed
 199  * may be allocated or freed using these operations.
 200  */
 201 void __kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
 202 int __kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
 203
 204 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
 205
 206 /* List of all root caches. */
 207 extern struct list_head         slab_root_caches;
 208 #define root_caches_node        memcg_params.__root_caches_node
 209
 210 /*
 211  * Iterate over all memcg caches of the given root cache. The caller must hold
 212  * slab_mutex.
 213  */
 214 #define for_each_memcg_cache(iter, root) \
 215         list_for_each_entry(iter, &(root)->memcg_params.children, \
 216                             memcg_params.children_node)
 217
 218 static inline bool is_root_cache(struct kmem_cache *s)
 219 {
 220         return !s->memcg_params.root_cache;
 221 }
 222
 223 static inline bool slab_equal_or_root(struct kmem_cache *s,
 224                                       struct kmem_cache *p)
 225 {
 226         return p == s || p == s->memcg_params.root_cache;
 227 }
 228
 229 /*
 230  * We use suffixes to the name in memcg because we can't have caches
 231  * created in the system with the same name. But when we print them
 232  * locally, better refer to them with the base name
 233  */
 234 static inline const char *cache_name(struct kmem_cache *s)
 235 {
 236         if (!is_root_cache(s))
 237                 s = s->memcg_params.root_cache;
 238         return s->name;
 239 }
 240
 241 /*
 242  * Note, we protect with RCU only the memcg_caches array, not per-memcg caches.
 243  * That said the caller must assure the memcg's cache won't go away by either
 244  * taking a css reference to the owner cgroup, or holding the slab_mutex.
 245  */
 246 static inline struct kmem_cache *
 247 cache_from_memcg_idx(struct kmem_cache *s, int idx)
 248 {
 249         struct kmem_cache *cachep;
 250         struct memcg_cache_array *arr;
 251
 252         rcu_read_lock();
 253         arr = rcu_dereference(s->memcg_params.memcg_caches);
 254
 255         /*
 256          * Make sure we will access the up-to-date value. The code updating
 257          * memcg_caches issues a write barrier to match this (see
 258          * memcg_create_kmem_cache()).
 259          */
 260         cachep = READ_ONCE(arr->entries[idx]);
 261         rcu_read_unlock();
 262
 263         return cachep;
 264 }
 265
 266 static inline struct kmem_cache *memcg_root_cache(struct kmem_cache *s)
 267 {
 268         if (is_root_cache(s))
 269                 return s;
 270         return s->memcg_params.root_cache;
 271 }
 272
 273 static __always_inline int memcg_charge_slab(struct page *page,
 274                                              gfp_t gfp, int order,
 275                                              struct kmem_cache *s)
 276 {
 277         if (!memcg_kmem_enabled())
 278                 return 0;
 279         if (is_root_cache(s))
 280                 return 0;
 281         return memcg_kmem_charge_memcg(page, gfp, order, s->memcg_params.memcg);
 282 }
 283
 284 static __always_inline void memcg_uncharge_slab(struct page *page, int order,
 285                                                 struct kmem_cache *s)
 286 {
 287         if (!memcg_kmem_enabled())
 288                 return;
 289         memcg_kmem_uncharge(page, order);
 290 }
 291
 292 extern void slab_init_memcg_params(struct kmem_cache *);
 293 extern void memcg_link_cache(struct kmem_cache *s);
 294 extern void slab_deactivate_memcg_cache_rcu_sched(struct kmem_cache *s,
 295                                 void (*deact_fn)(struct kmem_cache *));
 296
 297 #else /* CONFIG_MEMCG && !CONFIG_SLOB */
 298
 299 /* If !memcg, all caches are root. */
 300 #define slab_root_caches        slab_caches
 301 #define root_caches_node        list
 302
 303 #define for_each_memcg_cache(iter, root) \
 304         for ((void)(iter), (void)(root); 0; )
 305
 306 static inline bool is_root_cache(struct kmem_cache *s)
 307 {
 308         return true;
 309 }
 310
 311 static inline bool slab_equal_or_root(struct kmem_cache *s,
 312                                       struct kmem_cache *p)
 313 {
 314         return true;
 315 }
 316
 317 static inline const char *cache_name(struct kmem_cache *s)
 318 {
 319         return s->name;
 320 }
 321
 322 static inline struct kmem_cache *
 323 cache_from_memcg_idx(struct kmem_cache *s, int idx)
 324 {
 325         return NULL;
 326 }
 327
 328 static inline struct kmem_cache *memcg_root_cache(struct kmem_cache *s)
 329 {
 330         return s;
 331 }
 332
 333 static inline int memcg_charge_slab(struct page *page, gfp_t gfp, int order,
 334                                     struct kmem_cache *s)
 335 {
 336         return 0;
 337 }
 338
 339 static inline void memcg_uncharge_slab(struct page *page, int order,
 340                                        struct kmem_cache *s)
 341 {
 342 }
 343
 344 static inline void slab_init_memcg_params(struct kmem_cache *s)
 345 {
 346 }
 347
 348 static inline void memcg_link_cache(struct kmem_cache *s)
 349 {
 350 }
 351
 352 #endif /* CONFIG_MEMCG && !CONFIG_SLOB */
 353
 354 static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x)
 355 {
 356         struct kmem_cache *cachep;
 357         struct page *page;
 358
 359         /*
 360          * When kmemcg is not being used, both assignments should return the
 361          * same value. but we don't want to pay the assignment price in that
 362          * case. If it is not compiled in, the compiler should be smart enough
 363          * to not do even the assignment. In that case, slab_equal_or_root
 364          * will also be a constant.
 365          */
 366         if (!memcg_kmem_enabled() &&
 367             !unlikely(s->flags & SLAB_CONSISTENCY_CHECKS))
 368                 return s;
 369
 370         page = virt_to_head_page(x);
 371         cachep = page->slab_cache;
 372         if (slab_equal_or_root(cachep, s))
 373                 return cachep;
 374
 375         pr_err("%s: Wrong slab cache. %s but object is from %s\n",
 376                __func__, s->name, cachep->name);
 377         WARN_ON_ONCE(1);
 378         return s;
 379 }
 380
 381 static inline size_t slab_ksize(const struct kmem_cache *s)
 382 {
 383 #ifndef CONFIG_SLUB
 384         return s->object_size;
 385
 386 #else /* CONFIG_SLUB */
 387 # ifdef CONFIG_SLUB_DEBUG
 388         /*
 389          * Debugging requires use of the padding between object
 390          * and whatever may come after it.
 391          */
 392         if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
 393                 return s->object_size;
 394 # endif
 395         if (s->flags & SLAB_KASAN)
 396                 return s->object_size;
 397         /*
 398          * If we have the need to store the freelist pointer
 399          * back there or track user information then we can
 400          * only use the space before that information.
 401          */
 402         if (s->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_STORE_USER))
 403                 return s->inuse;
 404         /*
 405          * Else we can use all the padding etc for the allocation
 406          */
 407         return s->size;
 408 #endif
 409 }
 410
 411 static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s,
 412                                                      gfp_t flags)
 413 {
 414         flags &= gfp_allowed_mask;
 415
 416         fs_reclaim_acquire(flags);
 417         fs_reclaim_release(flags);
 418
 419         might_sleep_if(gfpflags_allow_blocking(flags));
 420
 421         if (should_failslab(s, flags))
 422                 return NULL;
 423
 424         if (memcg_kmem_enabled() &&
 425             ((flags & __GFP_ACCOUNT) || (s->flags & SLAB_ACCOUNT)))
 426                 return memcg_kmem_get_cache(s);
 427
 428         return s;
 429 }
 430
 431 static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags,
 432                                         size_t size, void **p)
 433 {
 434         size_t i;
 435
 436         flags &= gfp_allowed_mask;
 437         for (i = 0; i < size; i++) {
 438                 void *object = p[i];
 439
 440                 kmemleak_alloc_recursive(object, s->object_size, 1,
 441                                          s->flags, flags);
 442                 kasan_slab_alloc(s, object, flags);
 443         }
 444
 445         if (memcg_kmem_enabled())
 446                 memcg_kmem_put_cache(s);
 447 }
 448
 449 #ifndef CONFIG_SLOB
 450 /*
 451  * The slab lists for all objects.
 452  */
 453 struct kmem_cache_node {
 454         spinlock_t list_lock;
 455
 456 #ifdef CONFIG_SLAB
 457         struct list_head slabs_partial; /* partial list first, better asm code */
 458         struct list_head slabs_full;
 459         struct list_head slabs_free;
 460         unsigned long total_slabs;      /* length of all slab lists */
 461         unsigned long free_slabs;       /* length of free slab list only */
 462         unsigned long free_objects;
 463         unsigned int free_limit;
 464         unsigned int colour_next;       /* Per-node cache coloring */
 465         struct array_cache *shared;     /* shared per node */
 466         struct alien_cache **alien;     /* on other nodes */
 467         unsigned long next_reap;        /* updated without locking */
 468         int free_touched;               /* updated without locking */
 469 #endif
 470
 471 #ifdef CONFIG_SLUB
 472         unsigned long nr_partial;
 473         struct list_head partial;
 474 #ifdef CONFIG_SLUB_DEBUG
 475         atomic_long_t nr_slabs;
 476         atomic_long_t total_objects;
 477         struct list_head full;
 478 #endif
 479 #endif
 480
 481 };
 482
 483 static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
 484 {
 485         return s->node[node];
 486 }
 487
 488 /*
 489  * Iterator over all nodes. The body will be executed for each node that has
 490  * a kmem_cache_node structure allocated (which is true for all online nodes)
 491  */
 492 #define for_each_kmem_cache_node(__s, __node, __n) \
 493         for (__node = 0; __node < nr_node_ids; __node++) \
 494                  if ((__n = get_node(__s, __node)))
 495
 496 #endif
 497
 498 void *slab_start(struct seq_file *m, loff_t *pos);
 499 void *slab_next(struct seq_file *m, void *p, loff_t *pos);
 500 void slab_stop(struct seq_file *m, void *p);
 501 void *memcg_slab_start(struct seq_file *m, loff_t *pos);
 502 void *memcg_slab_next(struct seq_file *m, void *p, loff_t *pos);
 503 void memcg_slab_stop(struct seq_file *m, void *p);
 504 int memcg_slab_show(struct seq_file *m, void *p);
 505
 506 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
 507 void dump_unreclaimable_slab(void);
 508 #else
 509 static inline void dump_unreclaimable_slab(void)
 510 {
 511 }
 512 #endif
 513
 514 void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr);
 515
 516 #ifdef CONFIG_SLAB_FREELIST_RANDOM
 517 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
 518                         gfp_t gfp);
 519 void cache_random_seq_destroy(struct kmem_cache *cachep);
 520 #else
 521 static inline int cache_random_seq_create(struct kmem_cache *cachep,
 522                                         unsigned int count, gfp_t gfp)
 523 {
 524         return 0;
 525 }
 526 static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { }
 527 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
 528
 529 #endif /* MM_SLAB_H */