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>
51 * State of the slab allocator.
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.
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 */
66 extern enum slab_state slab_state;
68 /* The slab cache mutex protects the management structures during changes */
69 extern struct mutex slab_mutex;
71 /* The list of all slab caches on the system */
72 extern struct list_head slab_caches;
74 /* The slab cache that manages slab cache information */
75 extern struct kmem_cache *kmem_cache;
77 /* A table of kmalloc cache names and sizes */
78 extern const struct kmalloc_info_struct {
79 const char *name[NR_KMALLOC_TYPES];
84 /* Kmalloc array related functions */
85 void setup_kmalloc_cache_index_table(void);
86 void create_kmalloc_caches(slab_flags_t);
88 /* Find the kmalloc slab corresponding for a certain size */
89 struct kmem_cache *kmalloc_slab(size_t, gfp_t);
92 gfp_t kmalloc_fix_flags(gfp_t flags);
94 /* Functions provided by the slab allocators */
95 int __kmem_cache_create(struct kmem_cache *, slab_flags_t flags);
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);
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 *));
109 __kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
110 slab_flags_t flags, void (*ctor)(void *));
112 slab_flags_t kmem_cache_flags(unsigned int object_size,
113 slab_flags_t flags, const char *name,
114 void (*ctor)(void *));
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 *))
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 *))
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 )
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)
141 #define SLAB_DEBUG_FLAGS (0)
144 #if defined(CONFIG_SLAB)
145 #define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \
146 SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \
148 #elif defined(CONFIG_SLUB)
149 #define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
150 SLAB_TEMPORARY | SLAB_ACCOUNT)
152 #define SLAB_CACHE_FLAGS (0)
155 /* Common flags available with current configuration */
156 #define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
158 /* Common flags permitted for kmem_cache_create */
159 #define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \
164 SLAB_CONSISTENCY_CHECKS | \
167 SLAB_RECLAIM_ACCOUNT | \
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 *);
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;
187 unsigned int batchcount;
189 unsigned int objects_per_slab;
190 unsigned int cache_order;
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);
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.
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 **);
207 static inline int cache_vmstat_idx(struct kmem_cache *s)
209 return (s->flags & SLAB_RECLAIM_ACCOUNT) ?
210 NR_SLAB_RECLAIMABLE_B : NR_SLAB_UNRECLAIMABLE_B;
213 #ifdef CONFIG_SLUB_DEBUG
214 #ifdef CONFIG_SLUB_DEBUG_ON
215 DECLARE_STATIC_KEY_TRUE(slub_debug_enabled);
217 DECLARE_STATIC_KEY_FALSE(slub_debug_enabled);
219 extern void print_tracking(struct kmem_cache *s, void *object);
221 static inline void print_tracking(struct kmem_cache *s, void *object)
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
231 static inline bool kmem_cache_debug_flags(struct kmem_cache *s, slab_flags_t flags)
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;
241 #ifdef CONFIG_MEMCG_KMEM
242 static inline struct obj_cgroup **page_obj_cgroups(struct page *page)
245 * page->mem_cgroup and page->obj_cgroups are sharing the same
246 * space. To distinguish between them in case we don't know for sure
247 * that the page is a slab page (e.g. page_cgroup_ino()), let's
248 * always set the lowest bit of obj_cgroups.
250 return (struct obj_cgroup **)
251 ((unsigned long)page->obj_cgroups & ~0x1UL);
254 static inline bool page_has_obj_cgroups(struct page *page)
256 return ((unsigned long)page->obj_cgroups & 0x1UL);
259 int memcg_alloc_page_obj_cgroups(struct page *page, struct kmem_cache *s,
262 static inline void memcg_free_page_obj_cgroups(struct page *page)
264 kfree(page_obj_cgroups(page));
265 page->obj_cgroups = NULL;
268 static inline size_t obj_full_size(struct kmem_cache *s)
271 * For each accounted object there is an extra space which is used
272 * to store obj_cgroup membership. Charge it too.
274 return s->size + sizeof(struct obj_cgroup *);
277 static inline struct obj_cgroup *memcg_slab_pre_alloc_hook(struct kmem_cache *s,
281 struct obj_cgroup *objcg;
283 objcg = get_obj_cgroup_from_current();
287 if (obj_cgroup_charge(objcg, flags, objects * obj_full_size(s))) {
288 obj_cgroup_put(objcg);
295 static inline void mod_objcg_state(struct obj_cgroup *objcg,
296 struct pglist_data *pgdat,
299 struct mem_cgroup *memcg;
300 struct lruvec *lruvec;
303 memcg = obj_cgroup_memcg(objcg);
304 lruvec = mem_cgroup_lruvec(memcg, pgdat);
305 mod_memcg_lruvec_state(lruvec, idx, nr);
309 static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
310 struct obj_cgroup *objcg,
311 gfp_t flags, size_t size,
321 flags &= ~__GFP_ACCOUNT;
322 for (i = 0; i < size; i++) {
324 page = virt_to_head_page(p[i]);
326 if (!page_has_obj_cgroups(page) &&
327 memcg_alloc_page_obj_cgroups(page, s, flags)) {
328 obj_cgroup_uncharge(objcg, obj_full_size(s));
332 off = obj_to_index(s, page, p[i]);
333 obj_cgroup_get(objcg);
334 page_obj_cgroups(page)[off] = objcg;
335 mod_objcg_state(objcg, page_pgdat(page),
336 cache_vmstat_idx(s), obj_full_size(s));
338 obj_cgroup_uncharge(objcg, obj_full_size(s));
341 obj_cgroup_put(objcg);
344 static inline void memcg_slab_free_hook(struct kmem_cache *s_orig,
345 void **p, int objects)
347 struct kmem_cache *s;
348 struct obj_cgroup *objcg;
353 if (!memcg_kmem_enabled())
356 for (i = 0; i < objects; i++) {
360 page = virt_to_head_page(p[i]);
361 if (!page_has_obj_cgroups(page))
365 s = page->slab_cache;
369 off = obj_to_index(s, page, p[i]);
370 objcg = page_obj_cgroups(page)[off];
374 page_obj_cgroups(page)[off] = NULL;
375 obj_cgroup_uncharge(objcg, obj_full_size(s));
376 mod_objcg_state(objcg, page_pgdat(page), cache_vmstat_idx(s),
378 obj_cgroup_put(objcg);
382 #else /* CONFIG_MEMCG_KMEM */
383 static inline bool page_has_obj_cgroups(struct page *page)
388 static inline struct mem_cgroup *memcg_from_slab_obj(void *ptr)
393 static inline int memcg_alloc_page_obj_cgroups(struct page *page,
394 struct kmem_cache *s, gfp_t gfp)
399 static inline void memcg_free_page_obj_cgroups(struct page *page)
403 static inline struct obj_cgroup *memcg_slab_pre_alloc_hook(struct kmem_cache *s,
410 static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
411 struct obj_cgroup *objcg,
412 gfp_t flags, size_t size,
417 static inline void memcg_slab_free_hook(struct kmem_cache *s,
418 void **p, int objects)
421 #endif /* CONFIG_MEMCG_KMEM */
423 static inline struct kmem_cache *virt_to_cache(const void *obj)
427 page = virt_to_head_page(obj);
428 if (WARN_ONCE(!PageSlab(page), "%s: Object is not a Slab page!\n",
431 return page->slab_cache;
434 static __always_inline void account_slab_page(struct page *page, int order,
435 struct kmem_cache *s)
437 mod_node_page_state(page_pgdat(page), cache_vmstat_idx(s),
441 static __always_inline void unaccount_slab_page(struct page *page, int order,
442 struct kmem_cache *s)
444 if (memcg_kmem_enabled())
445 memcg_free_page_obj_cgroups(page);
447 mod_node_page_state(page_pgdat(page), cache_vmstat_idx(s),
448 -(PAGE_SIZE << order));
451 static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x)
453 struct kmem_cache *cachep;
455 if (!IS_ENABLED(CONFIG_SLAB_FREELIST_HARDENED) &&
456 !kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS))
459 cachep = virt_to_cache(x);
460 if (WARN(cachep && cachep != s,
461 "%s: Wrong slab cache. %s but object is from %s\n",
462 __func__, s->name, cachep->name))
463 print_tracking(cachep, x);
467 static inline size_t slab_ksize(const struct kmem_cache *s)
470 return s->object_size;
472 #else /* CONFIG_SLUB */
473 # ifdef CONFIG_SLUB_DEBUG
475 * Debugging requires use of the padding between object
476 * and whatever may come after it.
478 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
479 return s->object_size;
481 if (s->flags & SLAB_KASAN)
482 return s->object_size;
484 * If we have the need to store the freelist pointer
485 * back there or track user information then we can
486 * only use the space before that information.
488 if (s->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_STORE_USER))
491 * Else we can use all the padding etc for the allocation
497 static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s,
498 struct obj_cgroup **objcgp,
499 size_t size, gfp_t flags)
501 flags &= gfp_allowed_mask;
503 fs_reclaim_acquire(flags);
504 fs_reclaim_release(flags);
506 might_sleep_if(gfpflags_allow_blocking(flags));
508 if (should_failslab(s, flags))
511 if (memcg_kmem_enabled() &&
512 ((flags & __GFP_ACCOUNT) || (s->flags & SLAB_ACCOUNT)))
513 *objcgp = memcg_slab_pre_alloc_hook(s, size, flags);
518 static inline void slab_post_alloc_hook(struct kmem_cache *s,
519 struct obj_cgroup *objcg,
520 gfp_t flags, size_t size, void **p)
524 flags &= gfp_allowed_mask;
525 for (i = 0; i < size; i++) {
526 p[i] = kasan_slab_alloc(s, p[i], flags);
527 /* As p[i] might get tagged, call kmemleak hook after KASAN. */
528 kmemleak_alloc_recursive(p[i], s->object_size, 1,
532 if (memcg_kmem_enabled())
533 memcg_slab_post_alloc_hook(s, objcg, flags, size, p);
538 * The slab lists for all objects.
540 struct kmem_cache_node {
541 spinlock_t list_lock;
544 struct list_head slabs_partial; /* partial list first, better asm code */
545 struct list_head slabs_full;
546 struct list_head slabs_free;
547 unsigned long total_slabs; /* length of all slab lists */
548 unsigned long free_slabs; /* length of free slab list only */
549 unsigned long free_objects;
550 unsigned int free_limit;
551 unsigned int colour_next; /* Per-node cache coloring */
552 struct array_cache *shared; /* shared per node */
553 struct alien_cache **alien; /* on other nodes */
554 unsigned long next_reap; /* updated without locking */
555 int free_touched; /* updated without locking */
559 unsigned long nr_partial;
560 struct list_head partial;
561 #ifdef CONFIG_SLUB_DEBUG
562 atomic_long_t nr_slabs;
563 atomic_long_t total_objects;
564 struct list_head full;
570 static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
572 return s->node[node];
576 * Iterator over all nodes. The body will be executed for each node that has
577 * a kmem_cache_node structure allocated (which is true for all online nodes)
579 #define for_each_kmem_cache_node(__s, __node, __n) \
580 for (__node = 0; __node < nr_node_ids; __node++) \
581 if ((__n = get_node(__s, __node)))
585 void *slab_start(struct seq_file *m, loff_t *pos);
586 void *slab_next(struct seq_file *m, void *p, loff_t *pos);
587 void slab_stop(struct seq_file *m, void *p);
588 int memcg_slab_show(struct seq_file *m, void *p);
590 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
591 void dump_unreclaimable_slab(void);
593 static inline void dump_unreclaimable_slab(void)
598 void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr);
600 #ifdef CONFIG_SLAB_FREELIST_RANDOM
601 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
603 void cache_random_seq_destroy(struct kmem_cache *cachep);
605 static inline int cache_random_seq_create(struct kmem_cache *cachep,
606 unsigned int count, gfp_t gfp)
610 static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { }
611 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
613 static inline bool slab_want_init_on_alloc(gfp_t flags, struct kmem_cache *c)
615 if (static_branch_unlikely(&init_on_alloc)) {
618 if (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON))
619 return flags & __GFP_ZERO;
622 return flags & __GFP_ZERO;
625 static inline bool slab_want_init_on_free(struct kmem_cache *c)
627 if (static_branch_unlikely(&init_on_free))
629 (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)));
633 #endif /* MM_SLAB_H */