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[NR_KMALLOC_TYPES];
  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 gfp_t kmalloc_fix_flags(gfp_t flags);
  93
  94 /* Functions provided by the slab allocators */
  95 int __kmem_cache_create(struct kmem_cache *, slab_flags_t flags);
  96
  97 struct kmem_cache *create_kmalloc_cache(const char *name, unsigned int size,
  98                         slab_flags_t flags, unsigned int useroffset,
  99                         unsigned int usersize);
 100 extern void create_boot_cache(struct kmem_cache *, const char *name,
 101                         unsigned int size, slab_flags_t flags,
 102                         unsigned int useroffset, unsigned int usersize);
 103
 104 int slab_unmergeable(struct kmem_cache *s);
 105 struct kmem_cache *find_mergeable(unsigned size, unsigned align,
 106                 slab_flags_t flags, const char *name, void (*ctor)(void *));
 107 #ifndef CONFIG_SLOB
 108 struct kmem_cache *
 109 __kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
 110                    slab_flags_t flags, void (*ctor)(void *));
 111
 112 slab_flags_t kmem_cache_flags(unsigned int object_size,
 113         slab_flags_t flags, const char *name,
 114         void (*ctor)(void *));
 115 #else
 116 static inline struct kmem_cache *
 117 __kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
 118                    slab_flags_t flags, void (*ctor)(void *))
 119 { return NULL; }
 120
 121 static inline slab_flags_t kmem_cache_flags(unsigned int object_size,
 122         slab_flags_t flags, const char *name,
 123         void (*ctor)(void *))
 124 {
 125         return flags;
 126 }
 127 #endif
 128
 129
 130 /* Legal flag mask for kmem_cache_create(), for various configurations */
 131 #define SLAB_CORE_FLAGS (SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA | \
 132                          SLAB_CACHE_DMA32 | SLAB_PANIC | \
 133                          SLAB_TYPESAFE_BY_RCU | SLAB_DEBUG_OBJECTS )
 134
 135 #if defined(CONFIG_DEBUG_SLAB)
 136 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
 137 #elif defined(CONFIG_SLUB_DEBUG)
 138 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
 139                           SLAB_TRACE | SLAB_CONSISTENCY_CHECKS)
 140 #else
 141 #define SLAB_DEBUG_FLAGS (0)
 142 #endif
 143
 144 #if defined(CONFIG_SLAB)
 145 #define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \
 146                           SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \
 147                           SLAB_ACCOUNT)
 148 #elif defined(CONFIG_SLUB)
 149 #define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
 150                           SLAB_TEMPORARY | SLAB_ACCOUNT)
 151 #else
 152 #define SLAB_CACHE_FLAGS (0)
 153 #endif
 154
 155 /* Common flags available with current configuration */
 156 #define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
 157
 158 /* Common flags permitted for kmem_cache_create */
 159 #define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \
 160                               SLAB_RED_ZONE | \
 161                               SLAB_POISON | \
 162                               SLAB_STORE_USER | \
 163                               SLAB_TRACE | \
 164                               SLAB_CONSISTENCY_CHECKS | \
 165                               SLAB_MEM_SPREAD | \
 166                               SLAB_NOLEAKTRACE | \
 167                               SLAB_RECLAIM_ACCOUNT | \
 168                               SLAB_TEMPORARY | \
 169                               SLAB_ACCOUNT)
 170
 171 bool __kmem_cache_empty(struct kmem_cache *);
 172 int __kmem_cache_shutdown(struct kmem_cache *);
 173 void __kmem_cache_release(struct kmem_cache *);
 174 int __kmem_cache_shrink(struct kmem_cache *);
 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 static inline enum node_stat_item cache_vmstat_idx(struct kmem_cache *s)
 208 {
 209         return (s->flags & SLAB_RECLAIM_ACCOUNT) ?
 210                 NR_SLAB_RECLAIMABLE_B : NR_SLAB_UNRECLAIMABLE_B;
 211 }
 212
 213 #ifdef CONFIG_SLUB_DEBUG
 214 #ifdef CONFIG_SLUB_DEBUG_ON
 215 DECLARE_STATIC_KEY_TRUE(slub_debug_enabled);
 216 #else
 217 DECLARE_STATIC_KEY_FALSE(slub_debug_enabled);
 218 #endif
 219 extern void print_tracking(struct kmem_cache *s, void *object);
 220 #else
 221 static inline void print_tracking(struct kmem_cache *s, void *object)
 222 {
 223 }
 224 #endif
 225
 226 /*
 227  * Returns true if any of the specified slub_debug flags is enabled for the
 228  * cache. Use only for flags parsed by setup_slub_debug() as it also enables
 229  * the static key.
 230  */
 231 static inline bool kmem_cache_debug_flags(struct kmem_cache *s, slab_flags_t flags)
 232 {
 233 #ifdef CONFIG_SLUB_DEBUG
 234         VM_WARN_ON_ONCE(!(flags & SLAB_DEBUG_FLAGS));
 235         if (static_branch_unlikely(&slub_debug_enabled))
 236                 return s->flags & flags;
 237 #endif
 238         return false;
 239 }
 240
 241 #ifdef CONFIG_MEMCG_KMEM
 242 int memcg_alloc_page_obj_cgroups(struct page *page, struct kmem_cache *s,
 243                                  gfp_t gfp);
 244
 245 static inline void memcg_free_page_obj_cgroups(struct page *page)
 246 {
 247         kfree(page_objcgs(page));
 248         page->memcg_data = 0;
 249 }
 250
 251 static inline size_t obj_full_size(struct kmem_cache *s)
 252 {
 253         /*
 254          * For each accounted object there is an extra space which is used
 255          * to store obj_cgroup membership. Charge it too.
 256          */
 257         return s->size + sizeof(struct obj_cgroup *);
 258 }
 259
 260 /*
 261  * Returns false if the allocation should fail.
 262  */
 263 static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s,
 264                                              struct obj_cgroup **objcgp,
 265                                              size_t objects, gfp_t flags)
 266 {
 267         struct obj_cgroup *objcg;
 268
 269         if (!memcg_kmem_enabled())
 270                 return true;
 271
 272         if (!(flags & __GFP_ACCOUNT) && !(s->flags & SLAB_ACCOUNT))
 273                 return true;
 274
 275         objcg = get_obj_cgroup_from_current();
 276         if (!objcg)
 277                 return true;
 278
 279         if (obj_cgroup_charge(objcg, flags, objects * obj_full_size(s))) {
 280                 obj_cgroup_put(objcg);
 281                 return false;
 282         }
 283
 284         *objcgp = objcg;
 285         return true;
 286 }
 287
 288 static inline void mod_objcg_state(struct obj_cgroup *objcg,
 289                                    struct pglist_data *pgdat,
 290                                    enum node_stat_item idx, int nr)
 291 {
 292         struct mem_cgroup *memcg;
 293         struct lruvec *lruvec;
 294
 295         rcu_read_lock();
 296         memcg = obj_cgroup_memcg(objcg);
 297         lruvec = mem_cgroup_lruvec(memcg, pgdat);
 298         mod_memcg_lruvec_state(lruvec, idx, nr);
 299         rcu_read_unlock();
 300 }
 301
 302 static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
 303                                               struct obj_cgroup *objcg,
 304                                               gfp_t flags, size_t size,
 305                                               void **p)
 306 {
 307         struct page *page;
 308         unsigned long off;
 309         size_t i;
 310
 311         if (!memcg_kmem_enabled() || !objcg)
 312                 return;
 313
 314         flags &= ~__GFP_ACCOUNT;
 315         for (i = 0; i < size; i++) {
 316                 if (likely(p[i])) {
 317                         page = virt_to_head_page(p[i]);
 318
 319                         if (!page_objcgs(page) &&
 320                             memcg_alloc_page_obj_cgroups(page, s, flags)) {
 321                                 obj_cgroup_uncharge(objcg, obj_full_size(s));
 322                                 continue;
 323                         }
 324
 325                         off = obj_to_index(s, page, p[i]);
 326                         obj_cgroup_get(objcg);
 327                         page_objcgs(page)[off] = objcg;
 328                         mod_objcg_state(objcg, page_pgdat(page),
 329                                         cache_vmstat_idx(s), obj_full_size(s));
 330                 } else {
 331                         obj_cgroup_uncharge(objcg, obj_full_size(s));
 332                 }
 333         }
 334         obj_cgroup_put(objcg);
 335 }
 336
 337 static inline void memcg_slab_free_hook(struct kmem_cache *s_orig,
 338                                         void **p, int objects)
 339 {
 340         struct kmem_cache *s;
 341         struct obj_cgroup **objcgs;
 342         struct obj_cgroup *objcg;
 343         struct page *page;
 344         unsigned int off;
 345         int i;
 346
 347         if (!memcg_kmem_enabled())
 348                 return;
 349
 350         for (i = 0; i < objects; i++) {
 351                 if (unlikely(!p[i]))
 352                         continue;
 353
 354                 page = virt_to_head_page(p[i]);
 355                 objcgs = page_objcgs(page);
 356                 if (!objcgs)
 357                         continue;
 358
 359                 if (!s_orig)
 360                         s = page->slab_cache;
 361                 else
 362                         s = s_orig;
 363
 364                 off = obj_to_index(s, page, p[i]);
 365                 objcg = objcgs[off];
 366                 if (!objcg)
 367                         continue;
 368
 369                 objcgs[off] = NULL;
 370                 obj_cgroup_uncharge(objcg, obj_full_size(s));
 371                 mod_objcg_state(objcg, page_pgdat(page), cache_vmstat_idx(s),
 372                                 -obj_full_size(s));
 373                 obj_cgroup_put(objcg);
 374         }
 375 }
 376
 377 #else /* CONFIG_MEMCG_KMEM */
 378 static inline struct mem_cgroup *memcg_from_slab_obj(void *ptr)
 379 {
 380         return NULL;
 381 }
 382
 383 static inline int memcg_alloc_page_obj_cgroups(struct page *page,
 384                                                struct kmem_cache *s, gfp_t gfp)
 385 {
 386         return 0;
 387 }
 388
 389 static inline void memcg_free_page_obj_cgroups(struct page *page)
 390 {
 391 }
 392
 393 static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s,
 394                                              struct obj_cgroup **objcgp,
 395                                              size_t objects, gfp_t flags)
 396 {
 397         return true;
 398 }
 399
 400 static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
 401                                               struct obj_cgroup *objcg,
 402                                               gfp_t flags, size_t size,
 403                                               void **p)
 404 {
 405 }
 406
 407 static inline void memcg_slab_free_hook(struct kmem_cache *s,
 408                                         void **p, int objects)
 409 {
 410 }
 411 #endif /* CONFIG_MEMCG_KMEM */
 412
 413 static inline struct kmem_cache *virt_to_cache(const void *obj)
 414 {
 415         struct page *page;
 416
 417         page = virt_to_head_page(obj);
 418         if (WARN_ONCE(!PageSlab(page), "%s: Object is not a Slab page!\n",
 419                                         __func__))
 420                 return NULL;
 421         return page->slab_cache;
 422 }
 423
 424 static __always_inline void account_slab_page(struct page *page, int order,
 425                                               struct kmem_cache *s)
 426 {
 427         mod_node_page_state(page_pgdat(page), cache_vmstat_idx(s),
 428                             PAGE_SIZE << order);
 429 }
 430
 431 static __always_inline void unaccount_slab_page(struct page *page, int order,
 432                                                 struct kmem_cache *s)
 433 {
 434         if (memcg_kmem_enabled())
 435                 memcg_free_page_obj_cgroups(page);
 436
 437         mod_node_page_state(page_pgdat(page), cache_vmstat_idx(s),
 438                             -(PAGE_SIZE << order));
 439 }
 440
 441 static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x)
 442 {
 443         struct kmem_cache *cachep;
 444
 445         if (!IS_ENABLED(CONFIG_SLAB_FREELIST_HARDENED) &&
 446             !kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS))
 447                 return s;
 448
 449         cachep = virt_to_cache(x);
 450         if (WARN(cachep && cachep != s,
 451                   "%s: Wrong slab cache. %s but object is from %s\n",
 452                   __func__, s->name, cachep->name))
 453                 print_tracking(cachep, x);
 454         return cachep;
 455 }
 456
 457 static inline size_t slab_ksize(const struct kmem_cache *s)
 458 {
 459 #ifndef CONFIG_SLUB
 460         return s->object_size;
 461
 462 #else /* CONFIG_SLUB */
 463 # ifdef CONFIG_SLUB_DEBUG
 464         /*
 465          * Debugging requires use of the padding between object
 466          * and whatever may come after it.
 467          */
 468         if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
 469                 return s->object_size;
 470 # endif
 471         if (s->flags & SLAB_KASAN)
 472                 return s->object_size;
 473         /*
 474          * If we have the need to store the freelist pointer
 475          * back there or track user information then we can
 476          * only use the space before that information.
 477          */
 478         if (s->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_STORE_USER))
 479                 return s->inuse;
 480         /*
 481          * Else we can use all the padding etc for the allocation
 482          */
 483         return s->size;
 484 #endif
 485 }
 486
 487 static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s,
 488                                                      struct obj_cgroup **objcgp,
 489                                                      size_t size, gfp_t flags)
 490 {
 491         flags &= gfp_allowed_mask;
 492
 493         might_alloc(flags);
 494
 495         if (should_failslab(s, flags))
 496                 return NULL;
 497
 498         if (!memcg_slab_pre_alloc_hook(s, objcgp, size, flags))
 499                 return NULL;
 500
 501         return s;
 502 }
 503
 504 static inline void slab_post_alloc_hook(struct kmem_cache *s,
 505                                         struct obj_cgroup *objcg,
 506                                         gfp_t flags, size_t size, void **p)
 507 {
 508         size_t i;
 509
 510         flags &= gfp_allowed_mask;
 511         for (i = 0; i < size; i++) {
 512                 p[i] = kasan_slab_alloc(s, p[i], flags);
 513                 /* As p[i] might get tagged, call kmemleak hook after KASAN. */
 514                 kmemleak_alloc_recursive(p[i], s->object_size, 1,
 515                                          s->flags, flags);
 516         }
 517
 518         memcg_slab_post_alloc_hook(s, objcg, flags, size, p);
 519 }
 520
 521 #ifndef CONFIG_SLOB
 522 /*
 523  * The slab lists for all objects.
 524  */
 525 struct kmem_cache_node {
 526         spinlock_t list_lock;
 527
 528 #ifdef CONFIG_SLAB
 529         struct list_head slabs_partial; /* partial list first, better asm code */
 530         struct list_head slabs_full;
 531         struct list_head slabs_free;
 532         unsigned long total_slabs;      /* length of all slab lists */
 533         unsigned long free_slabs;       /* length of free slab list only */
 534         unsigned long free_objects;
 535         unsigned int free_limit;
 536         unsigned int colour_next;       /* Per-node cache coloring */
 537         struct array_cache *shared;     /* shared per node */
 538         struct alien_cache **alien;     /* on other nodes */
 539         unsigned long next_reap;        /* updated without locking */
 540         int free_touched;               /* updated without locking */
 541 #endif
 542
 543 #ifdef CONFIG_SLUB
 544         unsigned long nr_partial;
 545         struct list_head partial;
 546 #ifdef CONFIG_SLUB_DEBUG
 547         atomic_long_t nr_slabs;
 548         atomic_long_t total_objects;
 549         struct list_head full;
 550 #endif
 551 #endif
 552
 553 };
 554
 555 static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
 556 {
 557         return s->node[node];
 558 }
 559
 560 /*
 561  * Iterator over all nodes. The body will be executed for each node that has
 562  * a kmem_cache_node structure allocated (which is true for all online nodes)
 563  */
 564 #define for_each_kmem_cache_node(__s, __node, __n) \
 565         for (__node = 0; __node < nr_node_ids; __node++) \
 566                  if ((__n = get_node(__s, __node)))
 567
 568 #endif
 569
 570 void *slab_start(struct seq_file *m, loff_t *pos);
 571 void *slab_next(struct seq_file *m, void *p, loff_t *pos);
 572 void slab_stop(struct seq_file *m, void *p);
 573 int memcg_slab_show(struct seq_file *m, void *p);
 574
 575 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
 576 void dump_unreclaimable_slab(void);
 577 #else
 578 static inline void dump_unreclaimable_slab(void)
 579 {
 580 }
 581 #endif
 582
 583 void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr);
 584
 585 #ifdef CONFIG_SLAB_FREELIST_RANDOM
 586 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
 587                         gfp_t gfp);
 588 void cache_random_seq_destroy(struct kmem_cache *cachep);
 589 #else
 590 static inline int cache_random_seq_create(struct kmem_cache *cachep,
 591                                         unsigned int count, gfp_t gfp)
 592 {
 593         return 0;
 594 }
 595 static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { }
 596 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
 597
 598 static inline bool slab_want_init_on_alloc(gfp_t flags, struct kmem_cache *c)
 599 {
 600         if (static_branch_unlikely(&init_on_alloc)) {
 601                 if (c->ctor)
 602                         return false;
 603                 if (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON))
 604                         return flags & __GFP_ZERO;
 605                 return true;
 606         }
 607         return flags & __GFP_ZERO;
 608 }
 609
 610 static inline bool slab_want_init_on_free(struct kmem_cache *c)
 611 {
 612         if (static_branch_unlikely(&init_on_free))
 613                 return !(c->ctor ||
 614                          (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)));
 615         return false;
 616 }
 617
 618 #endif /* MM_SLAB_H */