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);
115 static inline struct kmem_cache *
116 __kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
117 slab_flags_t flags, void (*ctor)(void *))
120 static inline slab_flags_t kmem_cache_flags(unsigned int object_size,
121 slab_flags_t flags, const char *name)
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 )
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)
139 #define SLAB_DEBUG_FLAGS (0)
142 #if defined(CONFIG_SLAB)
143 #define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \
144 SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \
146 #elif defined(CONFIG_SLUB)
147 #define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
148 SLAB_TEMPORARY | SLAB_ACCOUNT)
150 #define SLAB_CACHE_FLAGS (0)
153 /* Common flags available with current configuration */
154 #define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
156 /* Common flags permitted for kmem_cache_create */
157 #define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \
162 SLAB_CONSISTENCY_CHECKS | \
165 SLAB_RECLAIM_ACCOUNT | \
169 bool __kmem_cache_empty(struct kmem_cache *);
170 int __kmem_cache_shutdown(struct kmem_cache *);
171 void __kmem_cache_release(struct kmem_cache *);
172 int __kmem_cache_shrink(struct kmem_cache *);
173 void slab_kmem_cache_release(struct kmem_cache *);
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;
185 unsigned int batchcount;
187 unsigned int objects_per_slab;
188 unsigned int cache_order;
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);
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.
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 **);
205 static inline enum node_stat_item cache_vmstat_idx(struct kmem_cache *s)
207 return (s->flags & SLAB_RECLAIM_ACCOUNT) ?
208 NR_SLAB_RECLAIMABLE_B : NR_SLAB_UNRECLAIMABLE_B;
211 #ifdef CONFIG_SLUB_DEBUG
212 #ifdef CONFIG_SLUB_DEBUG_ON
213 DECLARE_STATIC_KEY_TRUE(slub_debug_enabled);
215 DECLARE_STATIC_KEY_FALSE(slub_debug_enabled);
217 extern void print_tracking(struct kmem_cache *s, void *object);
218 long validate_slab_cache(struct kmem_cache *s);
219 static inline bool __slub_debug_enabled(void)
221 return static_branch_unlikely(&slub_debug_enabled);
224 static inline void print_tracking(struct kmem_cache *s, void *object)
227 static inline bool __slub_debug_enabled(void)
234 * Returns true if any of the specified slub_debug flags is enabled for the
235 * cache. Use only for flags parsed by setup_slub_debug() as it also enables
238 static inline bool kmem_cache_debug_flags(struct kmem_cache *s, slab_flags_t flags)
240 if (IS_ENABLED(CONFIG_SLUB_DEBUG))
241 VM_WARN_ON_ONCE(!(flags & SLAB_DEBUG_FLAGS));
242 if (__slub_debug_enabled())
243 return s->flags & flags;
247 #ifdef CONFIG_MEMCG_KMEM
248 int memcg_alloc_page_obj_cgroups(struct page *page, struct kmem_cache *s,
249 gfp_t gfp, bool new_page);
250 void mod_objcg_state(struct obj_cgroup *objcg, struct pglist_data *pgdat,
251 enum node_stat_item idx, int nr);
253 static inline void memcg_free_page_obj_cgroups(struct page *page)
255 kfree(page_objcgs(page));
256 page->memcg_data = 0;
259 static inline size_t obj_full_size(struct kmem_cache *s)
262 * For each accounted object there is an extra space which is used
263 * to store obj_cgroup membership. Charge it too.
265 return s->size + sizeof(struct obj_cgroup *);
269 * Returns false if the allocation should fail.
271 static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s,
272 struct obj_cgroup **objcgp,
273 size_t objects, gfp_t flags)
275 struct obj_cgroup *objcg;
277 if (!memcg_kmem_enabled())
280 if (!(flags & __GFP_ACCOUNT) && !(s->flags & SLAB_ACCOUNT))
283 objcg = get_obj_cgroup_from_current();
287 if (obj_cgroup_charge(objcg, flags, objects * obj_full_size(s))) {
288 obj_cgroup_put(objcg);
296 static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
297 struct obj_cgroup *objcg,
298 gfp_t flags, size_t size,
305 if (!memcg_kmem_enabled() || !objcg)
308 for (i = 0; i < size; i++) {
310 page = virt_to_head_page(p[i]);
312 if (!page_objcgs(page) &&
313 memcg_alloc_page_obj_cgroups(page, s, flags,
315 obj_cgroup_uncharge(objcg, obj_full_size(s));
319 off = obj_to_index(s, page, p[i]);
320 obj_cgroup_get(objcg);
321 page_objcgs(page)[off] = objcg;
322 mod_objcg_state(objcg, page_pgdat(page),
323 cache_vmstat_idx(s), obj_full_size(s));
325 obj_cgroup_uncharge(objcg, obj_full_size(s));
328 obj_cgroup_put(objcg);
331 static inline void memcg_slab_free_hook(struct kmem_cache *s_orig,
332 void **p, int objects)
334 struct kmem_cache *s;
335 struct obj_cgroup **objcgs;
336 struct obj_cgroup *objcg;
341 if (!memcg_kmem_enabled())
344 for (i = 0; i < objects; i++) {
348 page = virt_to_head_page(p[i]);
349 objcgs = page_objcgs_check(page);
354 s = page->slab_cache;
358 off = obj_to_index(s, page, p[i]);
364 obj_cgroup_uncharge(objcg, obj_full_size(s));
365 mod_objcg_state(objcg, page_pgdat(page), cache_vmstat_idx(s),
367 obj_cgroup_put(objcg);
371 #else /* CONFIG_MEMCG_KMEM */
372 static inline struct mem_cgroup *memcg_from_slab_obj(void *ptr)
377 static inline int memcg_alloc_page_obj_cgroups(struct page *page,
378 struct kmem_cache *s, gfp_t gfp,
384 static inline void memcg_free_page_obj_cgroups(struct page *page)
388 static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s,
389 struct obj_cgroup **objcgp,
390 size_t objects, gfp_t flags)
395 static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
396 struct obj_cgroup *objcg,
397 gfp_t flags, size_t size,
402 static inline void memcg_slab_free_hook(struct kmem_cache *s,
403 void **p, int objects)
406 #endif /* CONFIG_MEMCG_KMEM */
408 static inline struct kmem_cache *virt_to_cache(const void *obj)
412 page = virt_to_head_page(obj);
413 if (WARN_ONCE(!PageSlab(page), "%s: Object is not a Slab page!\n",
416 return page->slab_cache;
419 static __always_inline void account_slab_page(struct page *page, int order,
420 struct kmem_cache *s,
423 if (memcg_kmem_enabled() && (s->flags & SLAB_ACCOUNT))
424 memcg_alloc_page_obj_cgroups(page, s, gfp, true);
426 mod_node_page_state(page_pgdat(page), cache_vmstat_idx(s),
430 static __always_inline void unaccount_slab_page(struct page *page, int order,
431 struct kmem_cache *s)
433 if (memcg_kmem_enabled())
434 memcg_free_page_obj_cgroups(page);
436 mod_node_page_state(page_pgdat(page), cache_vmstat_idx(s),
437 -(PAGE_SIZE << order));
440 static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x)
442 struct kmem_cache *cachep;
444 if (!IS_ENABLED(CONFIG_SLAB_FREELIST_HARDENED) &&
445 !kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS))
448 cachep = virt_to_cache(x);
449 if (WARN(cachep && cachep != s,
450 "%s: Wrong slab cache. %s but object is from %s\n",
451 __func__, s->name, cachep->name))
452 print_tracking(cachep, x);
456 static inline size_t slab_ksize(const struct kmem_cache *s)
459 return s->object_size;
461 #else /* CONFIG_SLUB */
462 # ifdef CONFIG_SLUB_DEBUG
464 * Debugging requires use of the padding between object
465 * and whatever may come after it.
467 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
468 return s->object_size;
470 if (s->flags & SLAB_KASAN)
471 return s->object_size;
473 * If we have the need to store the freelist pointer
474 * back there or track user information then we can
475 * only use the space before that information.
477 if (s->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_STORE_USER))
480 * Else we can use all the padding etc for the allocation
486 static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s,
487 struct obj_cgroup **objcgp,
488 size_t size, gfp_t flags)
490 flags &= gfp_allowed_mask;
494 if (should_failslab(s, flags))
497 if (!memcg_slab_pre_alloc_hook(s, objcgp, size, flags))
503 static inline void slab_post_alloc_hook(struct kmem_cache *s,
504 struct obj_cgroup *objcg, gfp_t flags,
505 size_t size, void **p, bool init)
509 flags &= gfp_allowed_mask;
512 * As memory initialization might be integrated into KASAN,
513 * kasan_slab_alloc and initialization memset must be
514 * kept together to avoid discrepancies in behavior.
516 * As p[i] might get tagged, memset and kmemleak hook come after KASAN.
518 for (i = 0; i < size; i++) {
519 p[i] = kasan_slab_alloc(s, p[i], flags, init);
520 if (p[i] && init && !kasan_has_integrated_init())
521 memset(p[i], 0, s->object_size);
522 kmemleak_alloc_recursive(p[i], s->object_size, 1,
526 memcg_slab_post_alloc_hook(s, objcg, flags, size, p);
531 * The slab lists for all objects.
533 struct kmem_cache_node {
534 spinlock_t list_lock;
537 struct list_head slabs_partial; /* partial list first, better asm code */
538 struct list_head slabs_full;
539 struct list_head slabs_free;
540 unsigned long total_slabs; /* length of all slab lists */
541 unsigned long free_slabs; /* length of free slab list only */
542 unsigned long free_objects;
543 unsigned int free_limit;
544 unsigned int colour_next; /* Per-node cache coloring */
545 struct array_cache *shared; /* shared per node */
546 struct alien_cache **alien; /* on other nodes */
547 unsigned long next_reap; /* updated without locking */
548 int free_touched; /* updated without locking */
552 unsigned long nr_partial;
553 struct list_head partial;
554 #ifdef CONFIG_SLUB_DEBUG
555 atomic_long_t nr_slabs;
556 atomic_long_t total_objects;
557 struct list_head full;
563 static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
565 return s->node[node];
569 * Iterator over all nodes. The body will be executed for each node that has
570 * a kmem_cache_node structure allocated (which is true for all online nodes)
572 #define for_each_kmem_cache_node(__s, __node, __n) \
573 for (__node = 0; __node < nr_node_ids; __node++) \
574 if ((__n = get_node(__s, __node)))
578 void *slab_start(struct seq_file *m, loff_t *pos);
579 void *slab_next(struct seq_file *m, void *p, loff_t *pos);
580 void slab_stop(struct seq_file *m, void *p);
581 int memcg_slab_show(struct seq_file *m, void *p);
583 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
584 void dump_unreclaimable_slab(void);
586 static inline void dump_unreclaimable_slab(void)
591 void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr);
593 #ifdef CONFIG_SLAB_FREELIST_RANDOM
594 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
596 void cache_random_seq_destroy(struct kmem_cache *cachep);
598 static inline int cache_random_seq_create(struct kmem_cache *cachep,
599 unsigned int count, gfp_t gfp)
603 static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { }
604 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
606 static inline bool slab_want_init_on_alloc(gfp_t flags, struct kmem_cache *c)
608 if (static_branch_maybe(CONFIG_INIT_ON_ALLOC_DEFAULT_ON,
612 if (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON))
613 return flags & __GFP_ZERO;
616 return flags & __GFP_ZERO;
619 static inline bool slab_want_init_on_free(struct kmem_cache *c)
621 if (static_branch_maybe(CONFIG_INIT_ON_FREE_DEFAULT_ON,
624 (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)));
628 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG)
629 void debugfs_slab_release(struct kmem_cache *);
631 static inline void debugfs_slab_release(struct kmem_cache *s) { }
635 #define KS_ADDRS_COUNT 16
636 struct kmem_obj_info {
638 struct page *kp_page;
640 unsigned long kp_data_offset;
641 struct kmem_cache *kp_slab_cache;
643 void *kp_stack[KS_ADDRS_COUNT];
644 void *kp_free_stack[KS_ADDRS_COUNT];
646 void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct page *page);
649 #endif /* MM_SLAB_H */