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 {
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);
93 /* Functions provided by the slab allocators */
94 int __kmem_cache_create(struct kmem_cache *, slab_flags_t flags);
96 struct kmem_cache *create_kmalloc_cache(const char *name, unsigned int size,
97 slab_flags_t flags, unsigned int useroffset,
98 unsigned int usersize);
99 extern void create_boot_cache(struct kmem_cache *, const char *name,
100 unsigned int size, slab_flags_t flags,
101 unsigned int useroffset, unsigned int usersize);
103 int slab_unmergeable(struct kmem_cache *s);
104 struct kmem_cache *find_mergeable(unsigned size, unsigned align,
105 slab_flags_t flags, const char *name, void (*ctor)(void *));
108 __kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
109 slab_flags_t flags, void (*ctor)(void *));
111 slab_flags_t kmem_cache_flags(unsigned int object_size,
112 slab_flags_t flags, const char *name,
113 void (*ctor)(void *));
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,
122 void (*ctor)(void *))
129 /* Legal flag mask for kmem_cache_create(), for various configurations */
130 #define SLAB_CORE_FLAGS (SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA | \
131 SLAB_CACHE_DMA32 | SLAB_PANIC | \
132 SLAB_TYPESAFE_BY_RCU | SLAB_DEBUG_OBJECTS )
134 #if defined(CONFIG_DEBUG_SLAB)
135 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
136 #elif defined(CONFIG_SLUB_DEBUG)
137 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
138 SLAB_TRACE | SLAB_CONSISTENCY_CHECKS)
140 #define SLAB_DEBUG_FLAGS (0)
143 #if defined(CONFIG_SLAB)
144 #define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \
145 SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \
147 #elif defined(CONFIG_SLUB)
148 #define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
149 SLAB_TEMPORARY | SLAB_ACCOUNT)
151 #define SLAB_CACHE_FLAGS (0)
154 /* Common flags available with current configuration */
155 #define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
157 /* Common flags permitted for kmem_cache_create */
158 #define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \
163 SLAB_CONSISTENCY_CHECKS | \
166 SLAB_RECLAIM_ACCOUNT | \
170 bool __kmem_cache_empty(struct kmem_cache *);
171 int __kmem_cache_shutdown(struct kmem_cache *);
172 void __kmem_cache_release(struct kmem_cache *);
173 int __kmem_cache_shrink(struct kmem_cache *);
174 void __kmemcg_cache_deactivate(struct kmem_cache *s);
175 void __kmemcg_cache_deactivate_after_rcu(struct kmem_cache *s);
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 : NR_SLAB_UNRECLAIMABLE;
214 #ifdef CONFIG_MEMCG_KMEM
216 /* List of all root caches. */
217 extern struct list_head slab_root_caches;
218 #define root_caches_node memcg_params.__root_caches_node
221 * Iterate over all memcg caches of the given root cache. The caller must hold
224 #define for_each_memcg_cache(iter, root) \
225 list_for_each_entry(iter, &(root)->memcg_params.children, \
226 memcg_params.children_node)
228 static inline bool is_root_cache(struct kmem_cache *s)
230 return !s->memcg_params.root_cache;
233 static inline bool slab_equal_or_root(struct kmem_cache *s,
234 struct kmem_cache *p)
236 return p == s || p == s->memcg_params.root_cache;
240 * We use suffixes to the name in memcg because we can't have caches
241 * created in the system with the same name. But when we print them
242 * locally, better refer to them with the base name
244 static inline const char *cache_name(struct kmem_cache *s)
246 if (!is_root_cache(s))
247 s = s->memcg_params.root_cache;
252 * Note, we protect with RCU only the memcg_caches array, not per-memcg caches.
253 * That said the caller must assure the memcg's cache won't go away by either
254 * taking a css reference to the owner cgroup, or holding the slab_mutex.
256 static inline struct kmem_cache *
257 cache_from_memcg_idx(struct kmem_cache *s, int idx)
259 struct kmem_cache *cachep;
260 struct memcg_cache_array *arr;
263 arr = rcu_dereference(s->memcg_params.memcg_caches);
266 * Make sure we will access the up-to-date value. The code updating
267 * memcg_caches issues a write barrier to match this (see
268 * memcg_create_kmem_cache()).
270 cachep = READ_ONCE(arr->entries[idx]);
276 static inline struct kmem_cache *memcg_root_cache(struct kmem_cache *s)
278 if (is_root_cache(s))
280 return s->memcg_params.root_cache;
283 static __always_inline int memcg_charge_slab(struct page *page,
284 gfp_t gfp, int order,
285 struct kmem_cache *s)
287 if (is_root_cache(s))
289 return memcg_kmem_charge_memcg(page, gfp, order, s->memcg_params.memcg);
292 static __always_inline void memcg_uncharge_slab(struct page *page, int order,
293 struct kmem_cache *s)
295 memcg_kmem_uncharge(page, order);
298 extern void slab_init_memcg_params(struct kmem_cache *);
299 extern void memcg_link_cache(struct kmem_cache *s, struct mem_cgroup *memcg);
301 #else /* CONFIG_MEMCG_KMEM */
303 /* If !memcg, all caches are root. */
304 #define slab_root_caches slab_caches
305 #define root_caches_node list
307 #define for_each_memcg_cache(iter, root) \
308 for ((void)(iter), (void)(root); 0; )
310 static inline bool is_root_cache(struct kmem_cache *s)
315 static inline bool slab_equal_or_root(struct kmem_cache *s,
316 struct kmem_cache *p)
321 static inline const char *cache_name(struct kmem_cache *s)
326 static inline struct kmem_cache *
327 cache_from_memcg_idx(struct kmem_cache *s, int idx)
332 static inline struct kmem_cache *memcg_root_cache(struct kmem_cache *s)
337 static inline int memcg_charge_slab(struct page *page, gfp_t gfp, int order,
338 struct kmem_cache *s)
343 static inline void memcg_uncharge_slab(struct page *page, int order,
344 struct kmem_cache *s)
348 static inline void slab_init_memcg_params(struct kmem_cache *s)
352 static inline void memcg_link_cache(struct kmem_cache *s,
353 struct mem_cgroup *memcg)
357 #endif /* CONFIG_MEMCG_KMEM */
359 static inline struct kmem_cache *virt_to_cache(const void *obj)
363 page = virt_to_head_page(obj);
364 if (WARN_ONCE(!PageSlab(page), "%s: Object is not a Slab page!\n",
367 return page->slab_cache;
370 static __always_inline int charge_slab_page(struct page *page,
371 gfp_t gfp, int order,
372 struct kmem_cache *s)
374 int ret = memcg_charge_slab(page, gfp, order, s);
377 mod_lruvec_page_state(page, cache_vmstat_idx(s), 1 << order);
382 static __always_inline void uncharge_slab_page(struct page *page, int order,
383 struct kmem_cache *s)
385 mod_lruvec_page_state(page, cache_vmstat_idx(s), -(1 << order));
386 memcg_uncharge_slab(page, order, s);
389 static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x)
391 struct kmem_cache *cachep;
394 * When kmemcg is not being used, both assignments should return the
395 * same value. but we don't want to pay the assignment price in that
396 * case. If it is not compiled in, the compiler should be smart enough
397 * to not do even the assignment. In that case, slab_equal_or_root
398 * will also be a constant.
400 if (!memcg_kmem_enabled() &&
401 !IS_ENABLED(CONFIG_SLAB_FREELIST_HARDENED) &&
402 !unlikely(s->flags & SLAB_CONSISTENCY_CHECKS))
405 cachep = virt_to_cache(x);
406 WARN_ONCE(cachep && !slab_equal_or_root(cachep, s),
407 "%s: Wrong slab cache. %s but object is from %s\n",
408 __func__, s->name, cachep->name);
412 static inline size_t slab_ksize(const struct kmem_cache *s)
415 return s->object_size;
417 #else /* CONFIG_SLUB */
418 # ifdef CONFIG_SLUB_DEBUG
420 * Debugging requires use of the padding between object
421 * and whatever may come after it.
423 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
424 return s->object_size;
426 if (s->flags & SLAB_KASAN)
427 return s->object_size;
429 * If we have the need to store the freelist pointer
430 * back there or track user information then we can
431 * only use the space before that information.
433 if (s->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_STORE_USER))
436 * Else we can use all the padding etc for the allocation
442 static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s,
445 flags &= gfp_allowed_mask;
447 fs_reclaim_acquire(flags);
448 fs_reclaim_release(flags);
450 might_sleep_if(gfpflags_allow_blocking(flags));
452 if (should_failslab(s, flags))
455 if (memcg_kmem_enabled() &&
456 ((flags & __GFP_ACCOUNT) || (s->flags & SLAB_ACCOUNT)))
457 return memcg_kmem_get_cache(s);
462 static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags,
463 size_t size, void **p)
467 flags &= gfp_allowed_mask;
468 for (i = 0; i < size; i++) {
469 p[i] = kasan_slab_alloc(s, p[i], flags);
470 /* As p[i] might get tagged, call kmemleak hook after KASAN. */
471 kmemleak_alloc_recursive(p[i], s->object_size, 1,
475 if (memcg_kmem_enabled())
476 memcg_kmem_put_cache(s);
481 * The slab lists for all objects.
483 struct kmem_cache_node {
484 spinlock_t list_lock;
487 struct list_head slabs_partial; /* partial list first, better asm code */
488 struct list_head slabs_full;
489 struct list_head slabs_free;
490 unsigned long total_slabs; /* length of all slab lists */
491 unsigned long free_slabs; /* length of free slab list only */
492 unsigned long free_objects;
493 unsigned int free_limit;
494 unsigned int colour_next; /* Per-node cache coloring */
495 struct array_cache *shared; /* shared per node */
496 struct alien_cache **alien; /* on other nodes */
497 unsigned long next_reap; /* updated without locking */
498 int free_touched; /* updated without locking */
502 unsigned long nr_partial;
503 struct list_head partial;
504 #ifdef CONFIG_SLUB_DEBUG
505 atomic_long_t nr_slabs;
506 atomic_long_t total_objects;
507 struct list_head full;
513 static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
515 return s->node[node];
519 * Iterator over all nodes. The body will be executed for each node that has
520 * a kmem_cache_node structure allocated (which is true for all online nodes)
522 #define for_each_kmem_cache_node(__s, __node, __n) \
523 for (__node = 0; __node < nr_node_ids; __node++) \
524 if ((__n = get_node(__s, __node)))
528 void *slab_start(struct seq_file *m, loff_t *pos);
529 void *slab_next(struct seq_file *m, void *p, loff_t *pos);
530 void slab_stop(struct seq_file *m, void *p);
531 void *memcg_slab_start(struct seq_file *m, loff_t *pos);
532 void *memcg_slab_next(struct seq_file *m, void *p, loff_t *pos);
533 void memcg_slab_stop(struct seq_file *m, void *p);
534 int memcg_slab_show(struct seq_file *m, void *p);
536 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
537 void dump_unreclaimable_slab(void);
539 static inline void dump_unreclaimable_slab(void)
544 void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr);
546 #ifdef CONFIG_SLAB_FREELIST_RANDOM
547 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
549 void cache_random_seq_destroy(struct kmem_cache *cachep);
551 static inline int cache_random_seq_create(struct kmem_cache *cachep,
552 unsigned int count, gfp_t gfp)
556 static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { }
557 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
559 #endif /* MM_SLAB_H */