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 #include <linux/kmemleak.h>
  50
  51 /*
  52  * State of the slab allocator.
  53  *
  54  * This is used to describe the states of the allocator during bootup.
  55  * Allocators use this to gradually bootstrap themselves. Most allocators
  56  * have the problem that the structures used for managing slab caches are
  57  * allocated from slab caches themselves.
  58  */
  59 enum slab_state {
  60         DOWN,                   /* No slab functionality yet */
  61         PARTIAL,                /* SLUB: kmem_cache_node available */
  62         PARTIAL_NODE,           /* SLAB: kmalloc size for node struct available */
  63         UP,                     /* Slab caches usable but not all extras yet */
  64         FULL                    /* Everything is working */
  65 };
  66
  67 extern enum slab_state slab_state;
  68
  69 /* The slab cache mutex protects the management structures during changes */
  70 extern struct mutex slab_mutex;
  71
  72 /* The list of all slab caches on the system */
  73 extern struct list_head slab_caches;
  74
  75 /* The slab cache that manages slab cache information */
  76 extern struct kmem_cache *kmem_cache;
  77
  78 /* A table of kmalloc cache names and sizes */
  79 extern const struct kmalloc_info_struct {
  80         const char *name[NR_KMALLOC_TYPES];
  81         unsigned int size;
  82 } kmalloc_info[];
  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 gfp_t kmalloc_fix_flags(gfp_t flags);
  94
  95 /* Functions provided by the slab allocators */
  96 int __kmem_cache_create(struct kmem_cache *, slab_flags_t flags);
  97
  98 struct kmem_cache *create_kmalloc_cache(const char *name, unsigned int size,
  99                         slab_flags_t flags, unsigned int useroffset,
 100                         unsigned int usersize);
 101 extern void create_boot_cache(struct kmem_cache *, const char *name,
 102                         unsigned int size, slab_flags_t flags,
 103                         unsigned int useroffset, unsigned int usersize);
 104
 105 int slab_unmergeable(struct kmem_cache *s);
 106 struct kmem_cache *find_mergeable(unsigned size, unsigned align,
 107                 slab_flags_t flags, const char *name, void (*ctor)(void *));
 108 #ifndef CONFIG_SLOB
 109 struct kmem_cache *
 110 __kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
 111                    slab_flags_t flags, void (*ctor)(void *));
 112
 113 slab_flags_t kmem_cache_flags(unsigned int object_size,
 114         slab_flags_t flags, const char *name,
 115         void (*ctor)(void *));
 116 #else
 117 static inline struct kmem_cache *
 118 __kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
 119                    slab_flags_t flags, void (*ctor)(void *))
 120 { return NULL; }
 121
 122 static inline slab_flags_t kmem_cache_flags(unsigned int object_size,
 123         slab_flags_t flags, const char *name,
 124         void (*ctor)(void *))
 125 {
 126         return flags;
 127 }
 128 #endif
 129
 130
 131 /* Legal flag mask for kmem_cache_create(), for various configurations */
 132 #define SLAB_CORE_FLAGS (SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA | \
 133                          SLAB_CACHE_DMA32 | SLAB_PANIC | \
 134                          SLAB_TYPESAFE_BY_RCU | SLAB_DEBUG_OBJECTS )
 135
 136 #if defined(CONFIG_DEBUG_SLAB)
 137 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
 138 #elif defined(CONFIG_SLUB_DEBUG)
 139 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
 140                           SLAB_TRACE | SLAB_CONSISTENCY_CHECKS)
 141 #else
 142 #define SLAB_DEBUG_FLAGS (0)
 143 #endif
 144
 145 #if defined(CONFIG_SLAB)
 146 #define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \
 147                           SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \
 148                           SLAB_ACCOUNT)
 149 #elif defined(CONFIG_SLUB)
 150 #define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
 151                           SLAB_TEMPORARY | SLAB_ACCOUNT)
 152 #else
 153 #define SLAB_CACHE_FLAGS (0)
 154 #endif
 155
 156 /* Common flags available with current configuration */
 157 #define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
 158
 159 /* Common flags permitted for kmem_cache_create */
 160 #define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \
 161                               SLAB_RED_ZONE | \
 162                               SLAB_POISON | \
 163                               SLAB_STORE_USER | \
 164                               SLAB_TRACE | \
 165                               SLAB_CONSISTENCY_CHECKS | \
 166                               SLAB_MEM_SPREAD | \
 167                               SLAB_NOLEAKTRACE | \
 168                               SLAB_RECLAIM_ACCOUNT | \
 169                               SLAB_TEMPORARY | \
 170                               SLAB_ACCOUNT)
 171
 172 bool __kmem_cache_empty(struct kmem_cache *);
 173 int __kmem_cache_shutdown(struct kmem_cache *);
 174 void __kmem_cache_release(struct kmem_cache *);
 175 int __kmem_cache_shrink(struct kmem_cache *);
 176 void slab_kmem_cache_release(struct kmem_cache *);
 177
 178 struct seq_file;
 179 struct file;
 180
 181 struct slabinfo {
 182         unsigned long active_objs;
 183         unsigned long num_objs;
 184         unsigned long active_slabs;
 185         unsigned long num_slabs;
 186         unsigned long shared_avail;
 187         unsigned int limit;
 188         unsigned int batchcount;
 189         unsigned int shared;
 190         unsigned int objects_per_slab;
 191         unsigned int cache_order;
 192 };
 193
 194 void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo);
 195 void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s);
 196 ssize_t slabinfo_write(struct file *file, const char __user *buffer,
 197                        size_t count, loff_t *ppos);
 198
 199 /*
 200  * Generic implementation of bulk operations
 201  * These are useful for situations in which the allocator cannot
 202  * perform optimizations. In that case segments of the object listed
 203  * may be allocated or freed using these operations.
 204  */
 205 void __kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
 206 int __kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
 207
 208 static inline int cache_vmstat_idx(struct kmem_cache *s)
 209 {
 210         return (s->flags & SLAB_RECLAIM_ACCOUNT) ?
 211                 NR_SLAB_RECLAIMABLE_B : NR_SLAB_UNRECLAIMABLE_B;
 212 }
 213
 214 #ifdef CONFIG_SLUB_DEBUG
 215 #ifdef CONFIG_SLUB_DEBUG_ON
 216 DECLARE_STATIC_KEY_TRUE(slub_debug_enabled);
 217 #else
 218 DECLARE_STATIC_KEY_FALSE(slub_debug_enabled);
 219 #endif
 220 extern void print_tracking(struct kmem_cache *s, void *object);
 221 #else
 222 static inline void print_tracking(struct kmem_cache *s, void *object)
 223 {
 224 }
 225 #endif
 226
 227 /*
 228  * Returns true if any of the specified slub_debug flags is enabled for the
 229  * cache. Use only for flags parsed by setup_slub_debug() as it also enables
 230  * the static key.
 231  */
 232 static inline bool kmem_cache_debug_flags(struct kmem_cache *s, slab_flags_t flags)
 233 {
 234 #ifdef CONFIG_SLUB_DEBUG
 235         VM_WARN_ON_ONCE(!(flags & SLAB_DEBUG_FLAGS));
 236         if (static_branch_unlikely(&slub_debug_enabled))
 237                 return s->flags & flags;
 238 #endif
 239         return false;
 240 }
 241
 242 #ifdef CONFIG_MEMCG_KMEM
 243 static inline struct obj_cgroup **page_obj_cgroups(struct page *page)
 244 {
 245         /*
 246          * page->mem_cgroup and page->obj_cgroups are sharing the same
 247          * space. To distinguish between them in case we don't know for sure
 248          * that the page is a slab page (e.g. page_cgroup_ino()), let's
 249          * always set the lowest bit of obj_cgroups.
 250          */
 251         return (struct obj_cgroup **)
 252                 ((unsigned long)page->obj_cgroups & ~0x1UL);
 253 }
 254
 255 static inline bool page_has_obj_cgroups(struct page *page)
 256 {
 257         return ((unsigned long)page->obj_cgroups & 0x1UL);
 258 }
 259
 260 int memcg_alloc_page_obj_cgroups(struct page *page, struct kmem_cache *s,
 261                                  gfp_t gfp);
 262
 263 static inline void memcg_free_page_obj_cgroups(struct page *page)
 264 {
 265         kfree(page_obj_cgroups(page));
 266         page->obj_cgroups = NULL;
 267 }
 268
 269 static inline size_t obj_full_size(struct kmem_cache *s)
 270 {
 271         /*
 272          * For each accounted object there is an extra space which is used
 273          * to store obj_cgroup membership. Charge it too.
 274          */
 275         return s->size + sizeof(struct obj_cgroup *);
 276 }
 277
 278 static inline struct obj_cgroup *memcg_slab_pre_alloc_hook(struct kmem_cache *s,
 279                                                            size_t objects,
 280                                                            gfp_t flags)
 281 {
 282         struct obj_cgroup *objcg;
 283
 284         if (memcg_kmem_bypass())
 285                 return NULL;
 286
 287         objcg = get_obj_cgroup_from_current();
 288         if (!objcg)
 289                 return NULL;
 290
 291         if (obj_cgroup_charge(objcg, flags, objects * obj_full_size(s))) {
 292                 obj_cgroup_put(objcg);
 293                 return NULL;
 294         }
 295
 296         return objcg;
 297 }
 298
 299 static inline void mod_objcg_state(struct obj_cgroup *objcg,
 300                                    struct pglist_data *pgdat,
 301                                    int idx, int nr)
 302 {
 303         struct mem_cgroup *memcg;
 304         struct lruvec *lruvec;
 305
 306         rcu_read_lock();
 307         memcg = obj_cgroup_memcg(objcg);
 308         lruvec = mem_cgroup_lruvec(memcg, pgdat);
 309         mod_memcg_lruvec_state(lruvec, idx, nr);
 310         rcu_read_unlock();
 311 }
 312
 313 static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
 314                                               struct obj_cgroup *objcg,
 315                                               gfp_t flags, size_t size,
 316                                               void **p)
 317 {
 318         struct page *page;
 319         unsigned long off;
 320         size_t i;
 321
 322         if (!objcg)
 323                 return;
 324
 325         flags &= ~__GFP_ACCOUNT;
 326         for (i = 0; i < size; i++) {
 327                 if (likely(p[i])) {
 328                         page = virt_to_head_page(p[i]);
 329
 330                         if (!page_has_obj_cgroups(page) &&
 331                             memcg_alloc_page_obj_cgroups(page, s, flags)) {
 332                                 obj_cgroup_uncharge(objcg, obj_full_size(s));
 333                                 continue;
 334                         }
 335
 336                         off = obj_to_index(s, page, p[i]);
 337                         obj_cgroup_get(objcg);
 338                         page_obj_cgroups(page)[off] = objcg;
 339                         mod_objcg_state(objcg, page_pgdat(page),
 340                                         cache_vmstat_idx(s), obj_full_size(s));
 341                 } else {
 342                         obj_cgroup_uncharge(objcg, obj_full_size(s));
 343                 }
 344         }
 345         obj_cgroup_put(objcg);
 346 }
 347
 348 static inline void memcg_slab_free_hook(struct kmem_cache *s, struct page *page,
 349                                         void *p)
 350 {
 351         struct obj_cgroup *objcg;
 352         unsigned int off;
 353
 354         if (!memcg_kmem_enabled())
 355                 return;
 356
 357         if (!page_has_obj_cgroups(page))
 358                 return;
 359
 360         off = obj_to_index(s, page, p);
 361         objcg = page_obj_cgroups(page)[off];
 362         page_obj_cgroups(page)[off] = NULL;
 363
 364         if (!objcg)
 365                 return;
 366
 367         obj_cgroup_uncharge(objcg, obj_full_size(s));
 368         mod_objcg_state(objcg, page_pgdat(page), cache_vmstat_idx(s),
 369                         -obj_full_size(s));
 370
 371         obj_cgroup_put(objcg);
 372 }
 373
 374 #else /* CONFIG_MEMCG_KMEM */
 375 static inline bool page_has_obj_cgroups(struct page *page)
 376 {
 377         return false;
 378 }
 379
 380 static inline struct mem_cgroup *memcg_from_slab_obj(void *ptr)
 381 {
 382         return NULL;
 383 }
 384
 385 static inline int memcg_alloc_page_obj_cgroups(struct page *page,
 386                                                struct kmem_cache *s, gfp_t gfp)
 387 {
 388         return 0;
 389 }
 390
 391 static inline void memcg_free_page_obj_cgroups(struct page *page)
 392 {
 393 }
 394
 395 static inline struct obj_cgroup *memcg_slab_pre_alloc_hook(struct kmem_cache *s,
 396                                                            size_t objects,
 397                                                            gfp_t flags)
 398 {
 399         return NULL;
 400 }
 401
 402 static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
 403                                               struct obj_cgroup *objcg,
 404                                               gfp_t flags, size_t size,
 405                                               void **p)
 406 {
 407 }
 408
 409 static inline void memcg_slab_free_hook(struct kmem_cache *s, struct page *page,
 410                                         void *p)
 411 {
 412 }
 413 #endif /* CONFIG_MEMCG_KMEM */
 414
 415 static inline struct kmem_cache *virt_to_cache(const void *obj)
 416 {
 417         struct page *page;
 418
 419         page = virt_to_head_page(obj);
 420         if (WARN_ONCE(!PageSlab(page), "%s: Object is not a Slab page!\n",
 421                                         __func__))
 422                 return NULL;
 423         return page->slab_cache;
 424 }
 425
 426 static __always_inline void account_slab_page(struct page *page, int order,
 427                                               struct kmem_cache *s)
 428 {
 429         mod_node_page_state(page_pgdat(page), cache_vmstat_idx(s),
 430                             PAGE_SIZE << order);
 431 }
 432
 433 static __always_inline void unaccount_slab_page(struct page *page, int order,
 434                                                 struct kmem_cache *s)
 435 {
 436         if (memcg_kmem_enabled())
 437                 memcg_free_page_obj_cgroups(page);
 438
 439         mod_node_page_state(page_pgdat(page), cache_vmstat_idx(s),
 440                             -(PAGE_SIZE << order));
 441 }
 442
 443 static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x)
 444 {
 445         struct kmem_cache *cachep;
 446
 447         if (!IS_ENABLED(CONFIG_SLAB_FREELIST_HARDENED) &&
 448             !kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS))
 449                 return s;
 450
 451         cachep = virt_to_cache(x);
 452         if (WARN(cachep && cachep != s,
 453                   "%s: Wrong slab cache. %s but object is from %s\n",
 454                   __func__, s->name, cachep->name))
 455                 print_tracking(cachep, x);
 456         return cachep;
 457 }
 458
 459 static inline size_t slab_ksize(const struct kmem_cache *s)
 460 {
 461 #ifndef CONFIG_SLUB
 462         return s->object_size;
 463
 464 #else /* CONFIG_SLUB */
 465 # ifdef CONFIG_SLUB_DEBUG
 466         /*
 467          * Debugging requires use of the padding between object
 468          * and whatever may come after it.
 469          */
 470         if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
 471                 return s->object_size;
 472 # endif
 473         if (s->flags & SLAB_KASAN)
 474                 return s->object_size;
 475         /*
 476          * If we have the need to store the freelist pointer
 477          * back there or track user information then we can
 478          * only use the space before that information.
 479          */
 480         if (s->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_STORE_USER))
 481                 return s->inuse;
 482         /*
 483          * Else we can use all the padding etc for the allocation
 484          */
 485         return s->size;
 486 #endif
 487 }
 488
 489 static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s,
 490                                                      struct obj_cgroup **objcgp,
 491                                                      size_t size, gfp_t flags)
 492 {
 493         flags &= gfp_allowed_mask;
 494
 495         fs_reclaim_acquire(flags);
 496         fs_reclaim_release(flags);
 497
 498         might_sleep_if(gfpflags_allow_blocking(flags));
 499
 500         if (should_failslab(s, flags))
 501                 return NULL;
 502
 503         if (memcg_kmem_enabled() &&
 504             ((flags & __GFP_ACCOUNT) || (s->flags & SLAB_ACCOUNT)))
 505                 *objcgp = memcg_slab_pre_alloc_hook(s, size, flags);
 506
 507         return s;
 508 }
 509
 510 static inline void slab_post_alloc_hook(struct kmem_cache *s,
 511                                         struct obj_cgroup *objcg,
 512                                         gfp_t flags, size_t size, void **p)
 513 {
 514         size_t i;
 515
 516         flags &= gfp_allowed_mask;
 517         for (i = 0; i < size; i++) {
 518                 p[i] = kasan_slab_alloc(s, p[i], flags);
 519                 /* As p[i] might get tagged, call kmemleak hook after KASAN. */
 520                 kmemleak_alloc_recursive(p[i], s->object_size, 1,
 521                                          s->flags, flags);
 522         }
 523
 524         if (memcg_kmem_enabled())
 525                 memcg_slab_post_alloc_hook(s, objcg, flags, size, p);
 526 }
 527
 528 #ifndef CONFIG_SLOB
 529 /*
 530  * The slab lists for all objects.
 531  */
 532 struct kmem_cache_node {
 533         spinlock_t list_lock;
 534
 535 #ifdef CONFIG_SLAB
 536         struct list_head slabs_partial; /* partial list first, better asm code */
 537         struct list_head slabs_full;
 538         struct list_head slabs_free;
 539         unsigned long total_slabs;      /* length of all slab lists */
 540         unsigned long free_slabs;       /* length of free slab list only */
 541         unsigned long free_objects;
 542         unsigned int free_limit;
 543         unsigned int colour_next;       /* Per-node cache coloring */
 544         struct array_cache *shared;     /* shared per node */
 545         struct alien_cache **alien;     /* on other nodes */
 546         unsigned long next_reap;        /* updated without locking */
 547         int free_touched;               /* updated without locking */
 548 #endif
 549
 550 #ifdef CONFIG_SLUB
 551         unsigned long nr_partial;
 552         struct list_head partial;
 553 #ifdef CONFIG_SLUB_DEBUG
 554         atomic_long_t nr_slabs;
 555         atomic_long_t total_objects;
 556         struct list_head full;
 557 #endif
 558 #endif
 559
 560 };
 561
 562 static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
 563 {
 564         return s->node[node];
 565 }
 566
 567 /*
 568  * Iterator over all nodes. The body will be executed for each node that has
 569  * a kmem_cache_node structure allocated (which is true for all online nodes)
 570  */
 571 #define for_each_kmem_cache_node(__s, __node, __n) \
 572         for (__node = 0; __node < nr_node_ids; __node++) \
 573                  if ((__n = get_node(__s, __node)))
 574
 575 #endif
 576
 577 void *slab_start(struct seq_file *m, loff_t *pos);
 578 void *slab_next(struct seq_file *m, void *p, loff_t *pos);
 579 void slab_stop(struct seq_file *m, void *p);
 580 int memcg_slab_show(struct seq_file *m, void *p);
 581
 582 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
 583 void dump_unreclaimable_slab(void);
 584 #else
 585 static inline void dump_unreclaimable_slab(void)
 586 {
 587 }
 588 #endif
 589
 590 void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr);
 591
 592 #ifdef CONFIG_SLAB_FREELIST_RANDOM
 593 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
 594                         gfp_t gfp);
 595 void cache_random_seq_destroy(struct kmem_cache *cachep);
 596 #else
 597 static inline int cache_random_seq_create(struct kmem_cache *cachep,
 598                                         unsigned int count, gfp_t gfp)
 599 {
 600         return 0;
 601 }
 602 static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { }
 603 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
 604
 605 static inline bool slab_want_init_on_alloc(gfp_t flags, struct kmem_cache *c)
 606 {
 607         if (static_branch_unlikely(&init_on_alloc)) {
 608                 if (c->ctor)
 609                         return false;
 610                 if (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON))
 611                         return flags & __GFP_ZERO;
 612                 return true;
 613         }
 614         return flags & __GFP_ZERO;
 615 }
 616
 617 static inline bool slab_want_init_on_free(struct kmem_cache *c)
 618 {
 619         if (static_branch_unlikely(&init_on_free))
 620                 return !(c->ctor ||
 621                          (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)));
 622         return false;
 623 }
 624
 625 #endif /* MM_SLAB_H */