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 (SLAB_NOLEAKTRACE)
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 int 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);
219 static inline void print_tracking(struct kmem_cache *s, void *object)
225 * Returns true if any of the specified slub_debug flags is enabled for the
226 * cache. Use only for flags parsed by setup_slub_debug() as it also enables
229 static inline bool kmem_cache_debug_flags(struct kmem_cache *s, slab_flags_t flags)
231 #ifdef CONFIG_SLUB_DEBUG
232 VM_WARN_ON_ONCE(!(flags & SLAB_DEBUG_FLAGS));
233 if (static_branch_unlikely(&slub_debug_enabled))
234 return s->flags & flags;
239 #ifdef CONFIG_MEMCG_KMEM
240 static inline struct obj_cgroup **page_obj_cgroups(struct page *page)
243 * page->mem_cgroup and page->obj_cgroups are sharing the same
244 * space. To distinguish between them in case we don't know for sure
245 * that the page is a slab page (e.g. page_cgroup_ino()), let's
246 * always set the lowest bit of obj_cgroups.
248 return (struct obj_cgroup **)
249 ((unsigned long)page->obj_cgroups & ~0x1UL);
252 static inline bool page_has_obj_cgroups(struct page *page)
254 return ((unsigned long)page->obj_cgroups & 0x1UL);
257 int memcg_alloc_page_obj_cgroups(struct page *page, struct kmem_cache *s,
260 static inline void memcg_free_page_obj_cgroups(struct page *page)
262 kfree(page_obj_cgroups(page));
263 page->obj_cgroups = NULL;
266 static inline size_t obj_full_size(struct kmem_cache *s)
269 * For each accounted object there is an extra space which is used
270 * to store obj_cgroup membership. Charge it too.
272 return s->size + sizeof(struct obj_cgroup *);
276 * Returns false if the allocation should fail.
278 static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s,
279 struct obj_cgroup **objcgp,
280 size_t objects, gfp_t flags)
282 struct obj_cgroup *objcg;
284 if (!memcg_kmem_enabled())
287 if (!(flags & __GFP_ACCOUNT) && !(s->flags & SLAB_ACCOUNT))
290 objcg = get_obj_cgroup_from_current();
294 if (obj_cgroup_charge(objcg, flags, objects * obj_full_size(s))) {
295 obj_cgroup_put(objcg);
303 static inline void mod_objcg_state(struct obj_cgroup *objcg,
304 struct pglist_data *pgdat,
307 struct mem_cgroup *memcg;
308 struct lruvec *lruvec;
311 memcg = obj_cgroup_memcg(objcg);
312 lruvec = mem_cgroup_lruvec(memcg, pgdat);
313 mod_memcg_lruvec_state(lruvec, idx, nr);
317 static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
318 struct obj_cgroup *objcg,
319 gfp_t flags, size_t size,
326 if (!memcg_kmem_enabled() || !objcg)
329 for (i = 0; i < size; i++) {
331 page = virt_to_head_page(p[i]);
333 if (!page_has_obj_cgroups(page) &&
334 memcg_alloc_page_obj_cgroups(page, s, flags)) {
335 obj_cgroup_uncharge(objcg, obj_full_size(s));
339 off = obj_to_index(s, page, p[i]);
340 obj_cgroup_get(objcg);
341 page_obj_cgroups(page)[off] = objcg;
342 mod_objcg_state(objcg, page_pgdat(page),
343 cache_vmstat_idx(s), obj_full_size(s));
345 obj_cgroup_uncharge(objcg, obj_full_size(s));
348 obj_cgroup_put(objcg);
351 static inline void memcg_slab_free_hook(struct kmem_cache *s_orig,
352 void **p, int objects)
354 struct kmem_cache *s;
355 struct obj_cgroup *objcg;
360 if (!memcg_kmem_enabled())
363 for (i = 0; i < objects; i++) {
367 page = virt_to_head_page(p[i]);
368 if (!page_has_obj_cgroups(page))
372 s = page->slab_cache;
376 off = obj_to_index(s, page, p[i]);
377 objcg = page_obj_cgroups(page)[off];
381 page_obj_cgroups(page)[off] = NULL;
382 obj_cgroup_uncharge(objcg, obj_full_size(s));
383 mod_objcg_state(objcg, page_pgdat(page), cache_vmstat_idx(s),
385 obj_cgroup_put(objcg);
389 #else /* CONFIG_MEMCG_KMEM */
390 static inline bool page_has_obj_cgroups(struct page *page)
395 static inline struct mem_cgroup *memcg_from_slab_obj(void *ptr)
400 static inline int memcg_alloc_page_obj_cgroups(struct page *page,
401 struct kmem_cache *s, gfp_t gfp)
406 static inline void memcg_free_page_obj_cgroups(struct page *page)
410 static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s,
411 struct obj_cgroup **objcgp,
412 size_t objects, gfp_t flags)
417 static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
418 struct obj_cgroup *objcg,
419 gfp_t flags, size_t size,
424 static inline void memcg_slab_free_hook(struct kmem_cache *s,
425 void **p, int objects)
428 #endif /* CONFIG_MEMCG_KMEM */
430 static inline struct kmem_cache *virt_to_cache(const void *obj)
434 page = virt_to_head_page(obj);
435 if (WARN_ONCE(!PageSlab(page), "%s: Object is not a Slab page!\n",
438 return page->slab_cache;
441 static __always_inline void account_slab_page(struct page *page, int order,
442 struct kmem_cache *s)
444 mod_node_page_state(page_pgdat(page), cache_vmstat_idx(s),
448 static __always_inline void unaccount_slab_page(struct page *page, int order,
449 struct kmem_cache *s)
451 if (memcg_kmem_enabled())
452 memcg_free_page_obj_cgroups(page);
454 mod_node_page_state(page_pgdat(page), cache_vmstat_idx(s),
455 -(PAGE_SIZE << order));
458 static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x)
460 struct kmem_cache *cachep;
462 if (!IS_ENABLED(CONFIG_SLAB_FREELIST_HARDENED) &&
463 !kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS))
466 cachep = virt_to_cache(x);
467 if (WARN(cachep && cachep != s,
468 "%s: Wrong slab cache. %s but object is from %s\n",
469 __func__, s->name, cachep->name))
470 print_tracking(cachep, x);
474 static inline size_t slab_ksize(const struct kmem_cache *s)
477 return s->object_size;
479 #else /* CONFIG_SLUB */
480 # ifdef CONFIG_SLUB_DEBUG
482 * Debugging requires use of the padding between object
483 * and whatever may come after it.
485 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
486 return s->object_size;
488 if (s->flags & SLAB_KASAN)
489 return s->object_size;
491 * If we have the need to store the freelist pointer
492 * back there or track user information then we can
493 * only use the space before that information.
495 if (s->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_STORE_USER))
498 * Else we can use all the padding etc for the allocation
504 static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s,
505 struct obj_cgroup **objcgp,
506 size_t size, gfp_t flags)
508 flags &= gfp_allowed_mask;
510 fs_reclaim_acquire(flags);
511 fs_reclaim_release(flags);
513 might_sleep_if(gfpflags_allow_blocking(flags));
515 if (should_failslab(s, flags))
518 if (!memcg_slab_pre_alloc_hook(s, objcgp, size, flags))
524 static inline void slab_post_alloc_hook(struct kmem_cache *s,
525 struct obj_cgroup *objcg,
526 gfp_t flags, size_t size, void **p)
530 flags &= gfp_allowed_mask;
531 for (i = 0; i < size; i++) {
532 p[i] = kasan_slab_alloc(s, p[i], flags);
533 /* As p[i] might get tagged, call kmemleak hook after KASAN. */
534 kmemleak_alloc_recursive(p[i], s->object_size, 1,
538 memcg_slab_post_alloc_hook(s, objcg, flags, size, p);
543 * The slab lists for all objects.
545 struct kmem_cache_node {
546 spinlock_t list_lock;
549 struct list_head slabs_partial; /* partial list first, better asm code */
550 struct list_head slabs_full;
551 struct list_head slabs_free;
552 unsigned long total_slabs; /* length of all slab lists */
553 unsigned long free_slabs; /* length of free slab list only */
554 unsigned long free_objects;
555 unsigned int free_limit;
556 unsigned int colour_next; /* Per-node cache coloring */
557 struct array_cache *shared; /* shared per node */
558 struct alien_cache **alien; /* on other nodes */
559 unsigned long next_reap; /* updated without locking */
560 int free_touched; /* updated without locking */
564 unsigned long nr_partial;
565 struct list_head partial;
566 #ifdef CONFIG_SLUB_DEBUG
567 atomic_long_t nr_slabs;
568 atomic_long_t total_objects;
569 struct list_head full;
575 static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
577 return s->node[node];
581 * Iterator over all nodes. The body will be executed for each node that has
582 * a kmem_cache_node structure allocated (which is true for all online nodes)
584 #define for_each_kmem_cache_node(__s, __node, __n) \
585 for (__node = 0; __node < nr_node_ids; __node++) \
586 if ((__n = get_node(__s, __node)))
590 void *slab_start(struct seq_file *m, loff_t *pos);
591 void *slab_next(struct seq_file *m, void *p, loff_t *pos);
592 void slab_stop(struct seq_file *m, void *p);
593 int memcg_slab_show(struct seq_file *m, void *p);
595 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
596 void dump_unreclaimable_slab(void);
598 static inline void dump_unreclaimable_slab(void)
603 void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr);
605 #ifdef CONFIG_SLAB_FREELIST_RANDOM
606 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
608 void cache_random_seq_destroy(struct kmem_cache *cachep);
610 static inline int cache_random_seq_create(struct kmem_cache *cachep,
611 unsigned int count, gfp_t gfp)
615 static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { }
616 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
618 static inline bool slab_want_init_on_alloc(gfp_t flags, struct kmem_cache *c)
620 if (static_branch_unlikely(&init_on_alloc)) {
623 if (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON))
624 return flags & __GFP_ZERO;
627 return flags & __GFP_ZERO;
630 static inline bool slab_want_init_on_free(struct kmem_cache *c)
632 if (static_branch_unlikely(&init_on_free))
634 (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)));
638 #endif /* MM_SLAB_H */