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 | \
 130                          SLAB_CACHE_DMA32 | SLAB_PANIC | \
 131                          SLAB_TYPESAFE_BY_RCU | SLAB_DEBUG_OBJECTS )
 132
 133 #if defined(CONFIG_DEBUG_SLAB)
 134 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
 135 #elif defined(CONFIG_SLUB_DEBUG)
 136 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
 137                           SLAB_TRACE | SLAB_CONSISTENCY_CHECKS)
 138 #else
 139 #define SLAB_DEBUG_FLAGS (0)
 140 #endif
 141
 142 #if defined(CONFIG_SLAB)
 143 #define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \
 144                           SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \
 145                           SLAB_ACCOUNT)
 146 #elif defined(CONFIG_SLUB)
 147 #define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
 148                           SLAB_TEMPORARY | SLAB_ACCOUNT)
 149 #else
 150 #define SLAB_CACHE_FLAGS (0)
 151 #endif
 152
 153 /* Common flags available with current configuration */
 154 #define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
 155
 156 /* Common flags permitted for kmem_cache_create */
 157 #define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \
 158                               SLAB_RED_ZONE | \
 159                               SLAB_POISON | \
 160                               SLAB_STORE_USER | \
 161                               SLAB_TRACE | \
 162                               SLAB_CONSISTENCY_CHECKS | \
 163                               SLAB_MEM_SPREAD | \
 164                               SLAB_NOLEAKTRACE | \
 165                               SLAB_RECLAIM_ACCOUNT | \
 166                               SLAB_TEMPORARY | \
 167                               SLAB_ACCOUNT)
 168
 169 int __kmem_cache_shutdown(struct kmem_cache *);
 170 void __kmem_cache_release(struct kmem_cache *);
 171 int __kmem_cache_shrink(struct kmem_cache *);
 172 void __kmemcg_cache_deactivate(struct kmem_cache *s);
 173 void slab_kmem_cache_release(struct kmem_cache *);
 174
 175 struct seq_file;
 176 struct file;
 177
 178 struct slabinfo {
 179         unsigned long active_objs;
 180         unsigned long num_objs;
 181         unsigned long active_slabs;
 182         unsigned long num_slabs;
 183         unsigned long shared_avail;
 184         unsigned int limit;
 185         unsigned int batchcount;
 186         unsigned int shared;
 187         unsigned int objects_per_slab;
 188         unsigned int cache_order;
 189 };
 190
 191 void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo);
 192 void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s);
 193 ssize_t slabinfo_write(struct file *file, const char __user *buffer,
 194                        size_t count, loff_t *ppos);
 195
 196 /*
 197  * Generic implementation of bulk operations
 198  * These are useful for situations in which the allocator cannot
 199  * perform optimizations. In that case segments of the object listed
 200  * may be allocated or freed using these operations.
 201  */
 202 void __kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
 203 int __kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
 204
 205 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
 206
 207 /* List of all root caches. */
 208 extern struct list_head         slab_root_caches;
 209 #define root_caches_node        memcg_params.__root_caches_node
 210
 211 /*
 212  * Iterate over all memcg caches of the given root cache. The caller must hold
 213  * slab_mutex.
 214  */
 215 #define for_each_memcg_cache(iter, root) \
 216         list_for_each_entry(iter, &(root)->memcg_params.children, \
 217                             memcg_params.children_node)
 218
 219 static inline bool is_root_cache(struct kmem_cache *s)
 220 {
 221         return !s->memcg_params.root_cache;
 222 }
 223
 224 static inline bool slab_equal_or_root(struct kmem_cache *s,
 225                                       struct kmem_cache *p)
 226 {
 227         return p == s || p == s->memcg_params.root_cache;
 228 }
 229
 230 /*
 231  * We use suffixes to the name in memcg because we can't have caches
 232  * created in the system with the same name. But when we print them
 233  * locally, better refer to them with the base name
 234  */
 235 static inline const char *cache_name(struct kmem_cache *s)
 236 {
 237         if (!is_root_cache(s))
 238                 s = s->memcg_params.root_cache;
 239         return s->name;
 240 }
 241
 242 /*
 243  * Note, we protect with RCU only the memcg_caches array, not per-memcg caches.
 244  * That said the caller must assure the memcg's cache won't go away by either
 245  * taking a css reference to the owner cgroup, or holding the slab_mutex.
 246  */
 247 static inline struct kmem_cache *
 248 cache_from_memcg_idx(struct kmem_cache *s, int idx)
 249 {
 250         struct kmem_cache *cachep;
 251         struct memcg_cache_array *arr;
 252
 253         rcu_read_lock();
 254         arr = rcu_dereference(s->memcg_params.memcg_caches);
 255
 256         /*
 257          * Make sure we will access the up-to-date value. The code updating
 258          * memcg_caches issues a write barrier to match this (see
 259          * memcg_create_kmem_cache()).
 260          */
 261         cachep = READ_ONCE(arr->entries[idx]);
 262         rcu_read_unlock();
 263
 264         return cachep;
 265 }
 266
 267 static inline struct kmem_cache *memcg_root_cache(struct kmem_cache *s)
 268 {
 269         if (is_root_cache(s))
 270                 return s;
 271         return s->memcg_params.root_cache;
 272 }
 273
 274 static __always_inline int memcg_charge_slab(struct page *page,
 275                                              gfp_t gfp, int order,
 276                                              struct kmem_cache *s)
 277 {
 278         if (!memcg_kmem_enabled())
 279                 return 0;
 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         if (!memcg_kmem_enabled())
 289                 return;
 290         memcg_kmem_uncharge(page, order);
 291 }
 292
 293 extern void slab_init_memcg_params(struct kmem_cache *);
 294 extern void memcg_link_cache(struct kmem_cache *s);
 295 extern void slab_deactivate_memcg_cache_rcu_sched(struct kmem_cache *s,
 296                                 void (*deact_fn)(struct kmem_cache *));
 297
 298 #else /* CONFIG_MEMCG && !CONFIG_SLOB */
 299
 300 /* If !memcg, all caches are root. */
 301 #define slab_root_caches        slab_caches
 302 #define root_caches_node        list
 303
 304 #define for_each_memcg_cache(iter, root) \
 305         for ((void)(iter), (void)(root); 0; )
 306
 307 static inline bool is_root_cache(struct kmem_cache *s)
 308 {
 309         return true;
 310 }
 311
 312 static inline bool slab_equal_or_root(struct kmem_cache *s,
 313                                       struct kmem_cache *p)
 314 {
 315         return true;
 316 }
 317
 318 static inline const char *cache_name(struct kmem_cache *s)
 319 {
 320         return s->name;
 321 }
 322
 323 static inline struct kmem_cache *
 324 cache_from_memcg_idx(struct kmem_cache *s, int idx)
 325 {
 326         return NULL;
 327 }
 328
 329 static inline struct kmem_cache *memcg_root_cache(struct kmem_cache *s)
 330 {
 331         return s;
 332 }
 333
 334 static inline int memcg_charge_slab(struct page *page, gfp_t gfp, int order,
 335                                     struct kmem_cache *s)
 336 {
 337         return 0;
 338 }
 339
 340 static inline void memcg_uncharge_slab(struct page *page, int order,
 341                                        struct kmem_cache *s)
 342 {
 343 }
 344
 345 static inline void slab_init_memcg_params(struct kmem_cache *s)
 346 {
 347 }
 348
 349 static inline void memcg_link_cache(struct kmem_cache *s)
 350 {
 351 }
 352
 353 #endif /* CONFIG_MEMCG && !CONFIG_SLOB */
 354
 355 static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x)
 356 {
 357         struct kmem_cache *cachep;
 358         struct page *page;
 359
 360         /*
 361          * When kmemcg is not being used, both assignments should return the
 362          * same value. but we don't want to pay the assignment price in that
 363          * case. If it is not compiled in, the compiler should be smart enough
 364          * to not do even the assignment. In that case, slab_equal_or_root
 365          * will also be a constant.
 366          */
 367         if (!memcg_kmem_enabled() &&
 368             !unlikely(s->flags & SLAB_CONSISTENCY_CHECKS))
 369                 return s;
 370
 371         page = virt_to_head_page(x);
 372         cachep = page->slab_cache;
 373         if (slab_equal_or_root(cachep, s))
 374                 return cachep;
 375
 376         pr_err("%s: Wrong slab cache. %s but object is from %s\n",
 377                __func__, s->name, cachep->name);
 378         WARN_ON_ONCE(1);
 379         return s;
 380 }
 381
 382 static inline size_t slab_ksize(const struct kmem_cache *s)
 383 {
 384 #ifndef CONFIG_SLUB
 385         return s->object_size;
 386
 387 #else /* CONFIG_SLUB */
 388 # ifdef CONFIG_SLUB_DEBUG
 389         /*
 390          * Debugging requires use of the padding between object
 391          * and whatever may come after it.
 392          */
 393         if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
 394                 return s->object_size;
 395 # endif
 396         if (s->flags & SLAB_KASAN)
 397                 return s->object_size;
 398         /*
 399          * If we have the need to store the freelist pointer
 400          * back there or track user information then we can
 401          * only use the space before that information.
 402          */
 403         if (s->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_STORE_USER))
 404                 return s->inuse;
 405         /*
 406          * Else we can use all the padding etc for the allocation
 407          */
 408         return s->size;
 409 #endif
 410 }
 411
 412 static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s,
 413                                                      gfp_t flags)
 414 {
 415         flags &= gfp_allowed_mask;
 416
 417         fs_reclaim_acquire(flags);
 418         fs_reclaim_release(flags);
 419
 420         might_sleep_if(gfpflags_allow_blocking(flags));
 421
 422         if (should_failslab(s, flags))
 423                 return NULL;
 424
 425         if (memcg_kmem_enabled() &&
 426             ((flags & __GFP_ACCOUNT) || (s->flags & SLAB_ACCOUNT)))
 427                 return memcg_kmem_get_cache(s);
 428
 429         return s;
 430 }
 431
 432 static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags,
 433                                         size_t size, void **p)
 434 {
 435         size_t i;
 436
 437         flags &= gfp_allowed_mask;
 438         for (i = 0; i < size; i++) {
 439                 void *object = p[i];
 440
 441                 kmemleak_alloc_recursive(object, s->object_size, 1,
 442                                          s->flags, flags);
 443                 kasan_slab_alloc(s, object, flags);
 444         }
 445
 446         if (memcg_kmem_enabled())
 447                 memcg_kmem_put_cache(s);
 448 }
 449
 450 #ifndef CONFIG_SLOB
 451 /*
 452  * The slab lists for all objects.
 453  */
 454 struct kmem_cache_node {
 455         spinlock_t list_lock;
 456
 457 #ifdef CONFIG_SLAB
 458         struct list_head slabs_partial; /* partial list first, better asm code */
 459         struct list_head slabs_full;
 460         struct list_head slabs_free;
 461         unsigned long total_slabs;      /* length of all slab lists */
 462         unsigned long free_slabs;       /* length of free slab list only */
 463         unsigned long free_objects;
 464         unsigned int free_limit;
 465         unsigned int colour_next;       /* Per-node cache coloring */
 466         struct array_cache *shared;     /* shared per node */
 467         struct alien_cache **alien;     /* on other nodes */
 468         unsigned long next_reap;        /* updated without locking */
 469         int free_touched;               /* updated without locking */
 470 #endif
 471
 472 #ifdef CONFIG_SLUB
 473         unsigned long nr_partial;
 474         struct list_head partial;
 475 #ifdef CONFIG_SLUB_DEBUG
 476         atomic_long_t nr_slabs;
 477         atomic_long_t total_objects;
 478         struct list_head full;
 479 #endif
 480 #endif
 481
 482 };
 483
 484 static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
 485 {
 486         return s->node[node];
 487 }
 488
 489 /*
 490  * Iterator over all nodes. The body will be executed for each node that has
 491  * a kmem_cache_node structure allocated (which is true for all online nodes)
 492  */
 493 #define for_each_kmem_cache_node(__s, __node, __n) \
 494         for (__node = 0; __node < nr_node_ids; __node++) \
 495                  if ((__n = get_node(__s, __node)))
 496
 497 #endif
 498
 499 void *slab_start(struct seq_file *m, loff_t *pos);
 500 void *slab_next(struct seq_file *m, void *p, loff_t *pos);
 501 void slab_stop(struct seq_file *m, void *p);
 502 void *memcg_slab_start(struct seq_file *m, loff_t *pos);
 503 void *memcg_slab_next(struct seq_file *m, void *p, loff_t *pos);
 504 void memcg_slab_stop(struct seq_file *m, void *p);
 505 int memcg_slab_show(struct seq_file *m, void *p);
 506
 507 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
 508 void dump_unreclaimable_slab(void);
 509 #else
 510 static inline void dump_unreclaimable_slab(void)
 511 {
 512 }
 513 #endif
 514
 515 void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr);
 516
 517 #ifdef CONFIG_SLAB_FREELIST_RANDOM
 518 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
 519                         gfp_t gfp);
 520 void cache_random_seq_destroy(struct kmem_cache *cachep);
 521 #else
 522 static inline int cache_random_seq_create(struct kmem_cache *cachep,
 523                                         unsigned int count, gfp_t gfp)
 524 {
 525         return 0;
 526 }
 527 static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { }
 528 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
 529
 530 #endif /* MM_SLAB_H */