1 /* SPDX-License-Identifier: GPL-2.0 */
5 * Internal slab definitions
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.
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.
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 */
33 #endif /* CONFIG_SLOB */
36 #include <linux/slab_def.h>
40 #include <linux/slub_def.h>
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>
52 * State of the slab allocator.
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.
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 */
67 extern enum slab_state slab_state;
69 /* The slab cache mutex protects the management structures during changes */
70 extern struct mutex slab_mutex;
72 /* The list of all slab caches on the system */
73 extern struct list_head slab_caches;
75 /* The slab cache that manages slab cache information */
76 extern struct kmem_cache *kmem_cache;
78 /* A table of kmalloc cache names and sizes */
79 extern const struct kmalloc_info_struct {
80 const char *name[NR_KMALLOC_TYPES];
85 /* Kmalloc array related functions */
86 void setup_kmalloc_cache_index_table(void);
87 void create_kmalloc_caches(slab_flags_t);
89 /* Find the kmalloc slab corresponding for a certain size */
90 struct kmem_cache *kmalloc_slab(size_t, gfp_t);
93 gfp_t kmalloc_fix_flags(gfp_t flags);
95 /* Functions provided by the slab allocators */
96 int __kmem_cache_create(struct kmem_cache *, slab_flags_t flags);
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);
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 *));
110 __kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
111 slab_flags_t flags, void (*ctor)(void *));
113 slab_flags_t kmem_cache_flags(unsigned int object_size,
114 slab_flags_t flags, const char *name,
115 void (*ctor)(void *));
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 *))
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 *))
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 )
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)
142 #define SLAB_DEBUG_FLAGS (0)
145 #if defined(CONFIG_SLAB)
146 #define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \
147 SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \
149 #elif defined(CONFIG_SLUB)
150 #define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
151 SLAB_TEMPORARY | SLAB_ACCOUNT)
153 #define SLAB_CACHE_FLAGS (0)
156 /* Common flags available with current configuration */
157 #define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
159 /* Common flags permitted for kmem_cache_create */
160 #define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \
165 SLAB_CONSISTENCY_CHECKS | \
168 SLAB_RECLAIM_ACCOUNT | \
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 *);
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;
188 unsigned int batchcount;
190 unsigned int objects_per_slab;
191 unsigned int cache_order;
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);
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.
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 **);
208 static inline int cache_vmstat_idx(struct kmem_cache *s)
210 return (s->flags & SLAB_RECLAIM_ACCOUNT) ?
211 NR_SLAB_RECLAIMABLE_B : NR_SLAB_UNRECLAIMABLE_B;
214 #ifdef CONFIG_SLUB_DEBUG
215 #ifdef CONFIG_SLUB_DEBUG_ON
216 DECLARE_STATIC_KEY_TRUE(slub_debug_enabled);
218 DECLARE_STATIC_KEY_FALSE(slub_debug_enabled);
220 extern void print_tracking(struct kmem_cache *s, void *object);
222 static inline void print_tracking(struct kmem_cache *s, void *object)
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
232 static inline bool kmem_cache_debug_flags(struct kmem_cache *s, slab_flags_t flags)
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;
242 #ifdef CONFIG_MEMCG_KMEM
243 static inline struct obj_cgroup **page_obj_cgroups(struct page *page)
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.
251 return (struct obj_cgroup **)
252 ((unsigned long)page->obj_cgroups & ~0x1UL);
255 static inline bool page_has_obj_cgroups(struct page *page)
257 return ((unsigned long)page->obj_cgroups & 0x1UL);
260 int memcg_alloc_page_obj_cgroups(struct page *page, struct kmem_cache *s,
263 static inline void memcg_free_page_obj_cgroups(struct page *page)
265 kfree(page_obj_cgroups(page));
266 page->obj_cgroups = NULL;
269 static inline size_t obj_full_size(struct kmem_cache *s)
272 * For each accounted object there is an extra space which is used
273 * to store obj_cgroup membership. Charge it too.
275 return s->size + sizeof(struct obj_cgroup *);
278 static inline struct obj_cgroup *memcg_slab_pre_alloc_hook(struct kmem_cache *s,
282 struct obj_cgroup *objcg;
284 if (memcg_kmem_bypass())
287 objcg = get_obj_cgroup_from_current();
291 if (obj_cgroup_charge(objcg, flags, objects * obj_full_size(s))) {
292 obj_cgroup_put(objcg);
299 static inline void mod_objcg_state(struct obj_cgroup *objcg,
300 struct pglist_data *pgdat,
303 struct mem_cgroup *memcg;
304 struct lruvec *lruvec;
307 memcg = obj_cgroup_memcg(objcg);
308 lruvec = mem_cgroup_lruvec(memcg, pgdat);
309 mod_memcg_lruvec_state(lruvec, idx, nr);
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,
325 flags &= ~__GFP_ACCOUNT;
326 for (i = 0; i < size; i++) {
328 page = virt_to_head_page(p[i]);
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));
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));
342 obj_cgroup_uncharge(objcg, obj_full_size(s));
345 obj_cgroup_put(objcg);
348 static inline void memcg_slab_free_hook(struct kmem_cache *s, struct page *page,
351 struct obj_cgroup *objcg;
354 if (!memcg_kmem_enabled())
357 if (!page_has_obj_cgroups(page))
360 off = obj_to_index(s, page, p);
361 objcg = page_obj_cgroups(page)[off];
362 page_obj_cgroups(page)[off] = NULL;
367 obj_cgroup_uncharge(objcg, obj_full_size(s));
368 mod_objcg_state(objcg, page_pgdat(page), cache_vmstat_idx(s),
371 obj_cgroup_put(objcg);
374 #else /* CONFIG_MEMCG_KMEM */
375 static inline bool page_has_obj_cgroups(struct page *page)
380 static inline struct mem_cgroup *memcg_from_slab_obj(void *ptr)
385 static inline int memcg_alloc_page_obj_cgroups(struct page *page,
386 struct kmem_cache *s, gfp_t gfp)
391 static inline void memcg_free_page_obj_cgroups(struct page *page)
395 static inline struct obj_cgroup *memcg_slab_pre_alloc_hook(struct kmem_cache *s,
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,
409 static inline void memcg_slab_free_hook(struct kmem_cache *s, struct page *page,
413 #endif /* CONFIG_MEMCG_KMEM */
415 static inline struct kmem_cache *virt_to_cache(const void *obj)
419 page = virt_to_head_page(obj);
420 if (WARN_ONCE(!PageSlab(page), "%s: Object is not a Slab page!\n",
423 return page->slab_cache;
426 static __always_inline void account_slab_page(struct page *page, int order,
427 struct kmem_cache *s)
429 mod_node_page_state(page_pgdat(page), cache_vmstat_idx(s),
433 static __always_inline void unaccount_slab_page(struct page *page, int order,
434 struct kmem_cache *s)
436 if (memcg_kmem_enabled())
437 memcg_free_page_obj_cgroups(page);
439 mod_node_page_state(page_pgdat(page), cache_vmstat_idx(s),
440 -(PAGE_SIZE << order));
443 static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x)
445 struct kmem_cache *cachep;
447 if (!IS_ENABLED(CONFIG_SLAB_FREELIST_HARDENED) &&
448 !kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS))
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);
459 static inline size_t slab_ksize(const struct kmem_cache *s)
462 return s->object_size;
464 #else /* CONFIG_SLUB */
465 # ifdef CONFIG_SLUB_DEBUG
467 * Debugging requires use of the padding between object
468 * and whatever may come after it.
470 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
471 return s->object_size;
473 if (s->flags & SLAB_KASAN)
474 return s->object_size;
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.
480 if (s->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_STORE_USER))
483 * Else we can use all the padding etc for the allocation
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)
493 flags &= gfp_allowed_mask;
495 fs_reclaim_acquire(flags);
496 fs_reclaim_release(flags);
498 might_sleep_if(gfpflags_allow_blocking(flags));
500 if (should_failslab(s, flags))
503 if (memcg_kmem_enabled() &&
504 ((flags & __GFP_ACCOUNT) || (s->flags & SLAB_ACCOUNT)))
505 *objcgp = memcg_slab_pre_alloc_hook(s, size, flags);
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)
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,
524 if (memcg_kmem_enabled())
525 memcg_slab_post_alloc_hook(s, objcg, flags, size, p);
530 * The slab lists for all objects.
532 struct kmem_cache_node {
533 spinlock_t list_lock;
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 */
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;
562 static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
564 return s->node[node];
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)
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)))
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);
582 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
583 void dump_unreclaimable_slab(void);
585 static inline void dump_unreclaimable_slab(void)
590 void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr);
592 #ifdef CONFIG_SLAB_FREELIST_RANDOM
593 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
595 void cache_random_seq_destroy(struct kmem_cache *cachep);
597 static inline int cache_random_seq_create(struct kmem_cache *cachep,
598 unsigned int count, gfp_t gfp)
602 static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { }
603 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
605 static inline bool slab_want_init_on_alloc(gfp_t flags, struct kmem_cache *c)
607 if (static_branch_unlikely(&init_on_alloc)) {
610 if (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON))
611 return flags & __GFP_ZERO;
614 return flags & __GFP_ZERO;
617 static inline bool slab_want_init_on_free(struct kmem_cache *c)
619 if (static_branch_unlikely(&init_on_free))
621 (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)));
625 #endif /* MM_SLAB_H */