1 /* SPDX-License-Identifier: GPL-2.0 */
5 * Internal slab definitions
8 /* Reuses the bits in struct page */
10 unsigned long __page_flags;
12 #if defined(CONFIG_SLAB)
15 struct list_head slab_list;
16 struct rcu_head rcu_head;
18 struct kmem_cache *slab_cache;
19 void *freelist; /* array of free object indexes */
20 void *s_mem; /* first object */
23 #elif defined(CONFIG_SLUB)
26 struct list_head slab_list;
27 struct rcu_head rcu_head;
28 #ifdef CONFIG_SLUB_CPU_PARTIAL
31 int slabs; /* Nr of slabs left */
35 struct kmem_cache *slab_cache;
36 /* Double-word boundary */
37 void *freelist; /* first free object */
39 unsigned long counters;
46 unsigned int __unused;
48 #elif defined(CONFIG_SLOB)
50 struct list_head slab_list;
52 void *freelist; /* first free block */
54 unsigned int __unused_2;
57 #error "Unexpected slab allocator configured"
60 atomic_t __page_refcount;
62 unsigned long memcg_data;
66 #define SLAB_MATCH(pg, sl) \
67 static_assert(offsetof(struct page, pg) == offsetof(struct slab, sl))
68 SLAB_MATCH(flags, __page_flags);
69 SLAB_MATCH(compound_head, slab_list); /* Ensure bit 0 is clear */
71 SLAB_MATCH(rcu_head, rcu_head);
73 SLAB_MATCH(_refcount, __page_refcount);
75 SLAB_MATCH(memcg_data, memcg_data);
78 static_assert(sizeof(struct slab) <= sizeof(struct page));
81 * folio_slab - Converts from folio to slab.
84 * Currently struct slab is a different representation of a folio where
85 * folio_test_slab() is true.
87 * Return: The slab which contains this folio.
89 #define folio_slab(folio) (_Generic((folio), \
90 const struct folio *: (const struct slab *)(folio), \
91 struct folio *: (struct slab *)(folio)))
94 * slab_folio - The folio allocated for a slab
97 * Slabs are allocated as folios that contain the individual objects and are
98 * using some fields in the first struct page of the folio - those fields are
99 * now accessed by struct slab. It is occasionally necessary to convert back to
100 * a folio in order to communicate with the rest of the mm. Please use this
101 * helper function instead of casting yourself, as the implementation may change
104 #define slab_folio(s) (_Generic((s), \
105 const struct slab *: (const struct folio *)s, \
106 struct slab *: (struct folio *)s))
109 * page_slab - Converts from first struct page to slab.
110 * @p: The first (either head of compound or single) page of slab.
112 * A temporary wrapper to convert struct page to struct slab in situations where
113 * we know the page is the compound head, or single order-0 page.
115 * Long-term ideally everything would work with struct slab directly or go
116 * through folio to struct slab.
118 * Return: The slab which contains this page
120 #define page_slab(p) (_Generic((p), \
121 const struct page *: (const struct slab *)(p), \
122 struct page *: (struct slab *)(p)))
125 * slab_page - The first struct page allocated for a slab
128 * A convenience wrapper for converting slab to the first struct page of the
129 * underlying folio, to communicate with code not yet converted to folio or
132 #define slab_page(s) folio_page(slab_folio(s), 0)
135 * If network-based swap is enabled, sl*b must keep track of whether pages
136 * were allocated from pfmemalloc reserves.
138 static inline bool slab_test_pfmemalloc(const struct slab *slab)
140 return folio_test_active((struct folio *)slab_folio(slab));
143 static inline void slab_set_pfmemalloc(struct slab *slab)
145 folio_set_active(slab_folio(slab));
148 static inline void slab_clear_pfmemalloc(struct slab *slab)
150 folio_clear_active(slab_folio(slab));
153 static inline void __slab_clear_pfmemalloc(struct slab *slab)
155 __folio_clear_active(slab_folio(slab));
158 static inline void *slab_address(const struct slab *slab)
160 return folio_address(slab_folio(slab));
163 static inline int slab_nid(const struct slab *slab)
165 return folio_nid(slab_folio(slab));
168 static inline pg_data_t *slab_pgdat(const struct slab *slab)
170 return folio_pgdat(slab_folio(slab));
173 static inline struct slab *virt_to_slab(const void *addr)
175 struct folio *folio = virt_to_folio(addr);
177 if (!folio_test_slab(folio))
180 return folio_slab(folio);
183 static inline int slab_order(const struct slab *slab)
185 return folio_order((struct folio *)slab_folio(slab));
188 static inline size_t slab_size(const struct slab *slab)
190 return PAGE_SIZE << slab_order(slab);
195 * Common fields provided in kmem_cache by all slab allocators
196 * This struct is either used directly by the allocator (SLOB)
197 * or the allocator must include definitions for all fields
198 * provided in kmem_cache_common in their definition of kmem_cache.
200 * Once we can do anonymous structs (C11 standard) we could put a
201 * anonymous struct definition in these allocators so that the
202 * separate allocations in the kmem_cache structure of SLAB and
203 * SLUB is no longer needed.
206 unsigned int object_size;/* The original size of the object */
207 unsigned int size; /* The aligned/padded/added on size */
208 unsigned int align; /* Alignment as calculated */
209 slab_flags_t flags; /* Active flags on the slab */
210 unsigned int useroffset;/* Usercopy region offset */
211 unsigned int usersize; /* Usercopy region size */
212 const char *name; /* Slab name for sysfs */
213 int refcount; /* Use counter */
214 void (*ctor)(void *); /* Called on object slot creation */
215 struct list_head list; /* List of all slab caches on the system */
218 #endif /* CONFIG_SLOB */
221 #include <linux/slab_def.h>
225 #include <linux/slub_def.h>
228 #include <linux/memcontrol.h>
229 #include <linux/fault-inject.h>
230 #include <linux/kasan.h>
231 #include <linux/kmemleak.h>
232 #include <linux/random.h>
233 #include <linux/sched/mm.h>
234 #include <linux/list_lru.h>
237 * State of the slab allocator.
239 * This is used to describe the states of the allocator during bootup.
240 * Allocators use this to gradually bootstrap themselves. Most allocators
241 * have the problem that the structures used for managing slab caches are
242 * allocated from slab caches themselves.
245 DOWN, /* No slab functionality yet */
246 PARTIAL, /* SLUB: kmem_cache_node available */
247 PARTIAL_NODE, /* SLAB: kmalloc size for node struct available */
248 UP, /* Slab caches usable but not all extras yet */
249 FULL /* Everything is working */
252 extern enum slab_state slab_state;
254 /* The slab cache mutex protects the management structures during changes */
255 extern struct mutex slab_mutex;
257 /* The list of all slab caches on the system */
258 extern struct list_head slab_caches;
260 /* The slab cache that manages slab cache information */
261 extern struct kmem_cache *kmem_cache;
263 /* A table of kmalloc cache names and sizes */
264 extern const struct kmalloc_info_struct {
265 const char *name[NR_KMALLOC_TYPES];
270 /* Kmalloc array related functions */
271 void setup_kmalloc_cache_index_table(void);
272 void create_kmalloc_caches(slab_flags_t);
274 /* Find the kmalloc slab corresponding for a certain size */
275 struct kmem_cache *kmalloc_slab(size_t, gfp_t);
278 gfp_t kmalloc_fix_flags(gfp_t flags);
280 /* Functions provided by the slab allocators */
281 int __kmem_cache_create(struct kmem_cache *, slab_flags_t flags);
283 struct kmem_cache *create_kmalloc_cache(const char *name, unsigned int size,
284 slab_flags_t flags, unsigned int useroffset,
285 unsigned int usersize);
286 extern void create_boot_cache(struct kmem_cache *, const char *name,
287 unsigned int size, slab_flags_t flags,
288 unsigned int useroffset, unsigned int usersize);
290 int slab_unmergeable(struct kmem_cache *s);
291 struct kmem_cache *find_mergeable(unsigned size, unsigned align,
292 slab_flags_t flags, const char *name, void (*ctor)(void *));
295 __kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
296 slab_flags_t flags, void (*ctor)(void *));
298 slab_flags_t kmem_cache_flags(unsigned int object_size,
299 slab_flags_t flags, const char *name);
301 static inline struct kmem_cache *
302 __kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
303 slab_flags_t flags, void (*ctor)(void *))
306 static inline slab_flags_t kmem_cache_flags(unsigned int object_size,
307 slab_flags_t flags, const char *name)
314 /* Legal flag mask for kmem_cache_create(), for various configurations */
315 #define SLAB_CORE_FLAGS (SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA | \
316 SLAB_CACHE_DMA32 | SLAB_PANIC | \
317 SLAB_TYPESAFE_BY_RCU | SLAB_DEBUG_OBJECTS )
319 #if defined(CONFIG_DEBUG_SLAB)
320 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
321 #elif defined(CONFIG_SLUB_DEBUG)
322 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
323 SLAB_TRACE | SLAB_CONSISTENCY_CHECKS)
325 #define SLAB_DEBUG_FLAGS (0)
328 #if defined(CONFIG_SLAB)
329 #define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \
330 SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \
332 #elif defined(CONFIG_SLUB)
333 #define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
334 SLAB_TEMPORARY | SLAB_ACCOUNT | SLAB_NO_USER_FLAGS)
336 #define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE)
339 /* Common flags available with current configuration */
340 #define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
342 /* Common flags permitted for kmem_cache_create */
343 #define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \
348 SLAB_CONSISTENCY_CHECKS | \
351 SLAB_RECLAIM_ACCOUNT | \
356 bool __kmem_cache_empty(struct kmem_cache *);
357 int __kmem_cache_shutdown(struct kmem_cache *);
358 void __kmem_cache_release(struct kmem_cache *);
359 int __kmem_cache_shrink(struct kmem_cache *);
360 void slab_kmem_cache_release(struct kmem_cache *);
366 unsigned long active_objs;
367 unsigned long num_objs;
368 unsigned long active_slabs;
369 unsigned long num_slabs;
370 unsigned long shared_avail;
372 unsigned int batchcount;
374 unsigned int objects_per_slab;
375 unsigned int cache_order;
378 void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo);
379 void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s);
380 ssize_t slabinfo_write(struct file *file, const char __user *buffer,
381 size_t count, loff_t *ppos);
383 static inline enum node_stat_item cache_vmstat_idx(struct kmem_cache *s)
385 return (s->flags & SLAB_RECLAIM_ACCOUNT) ?
386 NR_SLAB_RECLAIMABLE_B : NR_SLAB_UNRECLAIMABLE_B;
389 #ifdef CONFIG_SLUB_DEBUG
390 #ifdef CONFIG_SLUB_DEBUG_ON
391 DECLARE_STATIC_KEY_TRUE(slub_debug_enabled);
393 DECLARE_STATIC_KEY_FALSE(slub_debug_enabled);
395 extern void print_tracking(struct kmem_cache *s, void *object);
396 long validate_slab_cache(struct kmem_cache *s);
397 static inline bool __slub_debug_enabled(void)
399 return static_branch_unlikely(&slub_debug_enabled);
402 static inline void print_tracking(struct kmem_cache *s, void *object)
405 static inline bool __slub_debug_enabled(void)
412 * Returns true if any of the specified slub_debug flags is enabled for the
413 * cache. Use only for flags parsed by setup_slub_debug() as it also enables
416 static inline bool kmem_cache_debug_flags(struct kmem_cache *s, slab_flags_t flags)
418 if (IS_ENABLED(CONFIG_SLUB_DEBUG))
419 VM_WARN_ON_ONCE(!(flags & SLAB_DEBUG_FLAGS));
420 if (__slub_debug_enabled())
421 return s->flags & flags;
425 #ifdef CONFIG_MEMCG_KMEM
427 * slab_objcgs - get the object cgroups vector associated with a slab
428 * @slab: a pointer to the slab struct
430 * Returns a pointer to the object cgroups vector associated with the slab,
431 * or NULL if no such vector has been associated yet.
433 static inline struct obj_cgroup **slab_objcgs(struct slab *slab)
435 unsigned long memcg_data = READ_ONCE(slab->memcg_data);
437 VM_BUG_ON_PAGE(memcg_data && !(memcg_data & MEMCG_DATA_OBJCGS),
439 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_KMEM, slab_page(slab));
441 return (struct obj_cgroup **)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
444 int memcg_alloc_slab_cgroups(struct slab *slab, struct kmem_cache *s,
445 gfp_t gfp, bool new_slab);
446 void mod_objcg_state(struct obj_cgroup *objcg, struct pglist_data *pgdat,
447 enum node_stat_item idx, int nr);
449 static inline void memcg_free_slab_cgroups(struct slab *slab)
451 kfree(slab_objcgs(slab));
452 slab->memcg_data = 0;
455 static inline size_t obj_full_size(struct kmem_cache *s)
458 * For each accounted object there is an extra space which is used
459 * to store obj_cgroup membership. Charge it too.
461 return s->size + sizeof(struct obj_cgroup *);
465 * Returns false if the allocation should fail.
467 static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s,
468 struct list_lru *lru,
469 struct obj_cgroup **objcgp,
470 size_t objects, gfp_t flags)
472 struct obj_cgroup *objcg;
474 if (!memcg_kmem_enabled())
477 if (!(flags & __GFP_ACCOUNT) && !(s->flags & SLAB_ACCOUNT))
480 objcg = get_obj_cgroup_from_current();
486 struct mem_cgroup *memcg;
488 memcg = get_mem_cgroup_from_objcg(objcg);
489 ret = memcg_list_lru_alloc(memcg, lru, flags);
490 css_put(&memcg->css);
496 if (obj_cgroup_charge(objcg, flags, objects * obj_full_size(s)))
502 obj_cgroup_put(objcg);
506 static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
507 struct obj_cgroup *objcg,
508 gfp_t flags, size_t size,
515 if (!memcg_kmem_enabled() || !objcg)
518 for (i = 0; i < size; i++) {
520 slab = virt_to_slab(p[i]);
522 if (!slab_objcgs(slab) &&
523 memcg_alloc_slab_cgroups(slab, s, flags,
525 obj_cgroup_uncharge(objcg, obj_full_size(s));
529 off = obj_to_index(s, slab, p[i]);
530 obj_cgroup_get(objcg);
531 slab_objcgs(slab)[off] = objcg;
532 mod_objcg_state(objcg, slab_pgdat(slab),
533 cache_vmstat_idx(s), obj_full_size(s));
535 obj_cgroup_uncharge(objcg, obj_full_size(s));
538 obj_cgroup_put(objcg);
541 static inline void memcg_slab_free_hook(struct kmem_cache *s, struct slab *slab,
542 void **p, int objects)
544 struct obj_cgroup **objcgs;
547 if (!memcg_kmem_enabled())
550 objcgs = slab_objcgs(slab);
554 for (i = 0; i < objects; i++) {
555 struct obj_cgroup *objcg;
558 off = obj_to_index(s, slab, p[i]);
564 obj_cgroup_uncharge(objcg, obj_full_size(s));
565 mod_objcg_state(objcg, slab_pgdat(slab), cache_vmstat_idx(s),
567 obj_cgroup_put(objcg);
571 #else /* CONFIG_MEMCG_KMEM */
572 static inline struct obj_cgroup **slab_objcgs(struct slab *slab)
577 static inline struct mem_cgroup *memcg_from_slab_obj(void *ptr)
582 static inline int memcg_alloc_slab_cgroups(struct slab *slab,
583 struct kmem_cache *s, gfp_t gfp,
589 static inline void memcg_free_slab_cgroups(struct slab *slab)
593 static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s,
594 struct list_lru *lru,
595 struct obj_cgroup **objcgp,
596 size_t objects, gfp_t flags)
601 static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
602 struct obj_cgroup *objcg,
603 gfp_t flags, size_t size,
608 static inline void memcg_slab_free_hook(struct kmem_cache *s, struct slab *slab,
609 void **p, int objects)
612 #endif /* CONFIG_MEMCG_KMEM */
615 static inline struct kmem_cache *virt_to_cache(const void *obj)
619 slab = virt_to_slab(obj);
620 if (WARN_ONCE(!slab, "%s: Object is not a Slab page!\n",
623 return slab->slab_cache;
626 static __always_inline void account_slab(struct slab *slab, int order,
627 struct kmem_cache *s, gfp_t gfp)
629 if (memcg_kmem_enabled() && (s->flags & SLAB_ACCOUNT))
630 memcg_alloc_slab_cgroups(slab, s, gfp, true);
632 mod_node_page_state(slab_pgdat(slab), cache_vmstat_idx(s),
636 static __always_inline void unaccount_slab(struct slab *slab, int order,
637 struct kmem_cache *s)
639 if (memcg_kmem_enabled())
640 memcg_free_slab_cgroups(slab);
642 mod_node_page_state(slab_pgdat(slab), cache_vmstat_idx(s),
643 -(PAGE_SIZE << order));
646 static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x)
648 struct kmem_cache *cachep;
650 if (!IS_ENABLED(CONFIG_SLAB_FREELIST_HARDENED) &&
651 !kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS))
654 cachep = virt_to_cache(x);
655 if (WARN(cachep && cachep != s,
656 "%s: Wrong slab cache. %s but object is from %s\n",
657 __func__, s->name, cachep->name))
658 print_tracking(cachep, x);
661 #endif /* CONFIG_SLOB */
663 static inline size_t slab_ksize(const struct kmem_cache *s)
666 return s->object_size;
668 #else /* CONFIG_SLUB */
669 # ifdef CONFIG_SLUB_DEBUG
671 * Debugging requires use of the padding between object
672 * and whatever may come after it.
674 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
675 return s->object_size;
677 if (s->flags & SLAB_KASAN)
678 return s->object_size;
680 * If we have the need to store the freelist pointer
681 * back there or track user information then we can
682 * only use the space before that information.
684 if (s->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_STORE_USER))
687 * Else we can use all the padding etc for the allocation
693 static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s,
694 struct list_lru *lru,
695 struct obj_cgroup **objcgp,
696 size_t size, gfp_t flags)
698 flags &= gfp_allowed_mask;
702 if (should_failslab(s, flags))
705 if (!memcg_slab_pre_alloc_hook(s, lru, objcgp, size, flags))
711 static inline void slab_post_alloc_hook(struct kmem_cache *s,
712 struct obj_cgroup *objcg, gfp_t flags,
713 size_t size, void **p, bool init)
717 flags &= gfp_allowed_mask;
720 * As memory initialization might be integrated into KASAN,
721 * kasan_slab_alloc and initialization memset must be
722 * kept together to avoid discrepancies in behavior.
724 * As p[i] might get tagged, memset and kmemleak hook come after KASAN.
726 for (i = 0; i < size; i++) {
727 p[i] = kasan_slab_alloc(s, p[i], flags, init);
728 if (p[i] && init && !kasan_has_integrated_init())
729 memset(p[i], 0, s->object_size);
730 kmemleak_alloc_recursive(p[i], s->object_size, 1,
734 memcg_slab_post_alloc_hook(s, objcg, flags, size, p);
739 * The slab lists for all objects.
741 struct kmem_cache_node {
742 spinlock_t list_lock;
745 struct list_head slabs_partial; /* partial list first, better asm code */
746 struct list_head slabs_full;
747 struct list_head slabs_free;
748 unsigned long total_slabs; /* length of all slab lists */
749 unsigned long free_slabs; /* length of free slab list only */
750 unsigned long free_objects;
751 unsigned int free_limit;
752 unsigned int colour_next; /* Per-node cache coloring */
753 struct array_cache *shared; /* shared per node */
754 struct alien_cache **alien; /* on other nodes */
755 unsigned long next_reap; /* updated without locking */
756 int free_touched; /* updated without locking */
760 unsigned long nr_partial;
761 struct list_head partial;
762 #ifdef CONFIG_SLUB_DEBUG
763 atomic_long_t nr_slabs;
764 atomic_long_t total_objects;
765 struct list_head full;
771 static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
773 return s->node[node];
777 * Iterator over all nodes. The body will be executed for each node that has
778 * a kmem_cache_node structure allocated (which is true for all online nodes)
780 #define for_each_kmem_cache_node(__s, __node, __n) \
781 for (__node = 0; __node < nr_node_ids; __node++) \
782 if ((__n = get_node(__s, __node)))
786 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
787 void dump_unreclaimable_slab(void);
789 static inline void dump_unreclaimable_slab(void)
794 void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr);
796 #ifdef CONFIG_SLAB_FREELIST_RANDOM
797 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
799 void cache_random_seq_destroy(struct kmem_cache *cachep);
801 static inline int cache_random_seq_create(struct kmem_cache *cachep,
802 unsigned int count, gfp_t gfp)
806 static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { }
807 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
809 static inline bool slab_want_init_on_alloc(gfp_t flags, struct kmem_cache *c)
811 if (static_branch_maybe(CONFIG_INIT_ON_ALLOC_DEFAULT_ON,
815 if (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON))
816 return flags & __GFP_ZERO;
819 return flags & __GFP_ZERO;
822 static inline bool slab_want_init_on_free(struct kmem_cache *c)
824 if (static_branch_maybe(CONFIG_INIT_ON_FREE_DEFAULT_ON,
827 (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)));
831 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG)
832 void debugfs_slab_release(struct kmem_cache *);
834 static inline void debugfs_slab_release(struct kmem_cache *s) { }
838 #define KS_ADDRS_COUNT 16
839 struct kmem_obj_info {
841 struct slab *kp_slab;
843 unsigned long kp_data_offset;
844 struct kmem_cache *kp_slab_cache;
846 void *kp_stack[KS_ADDRS_COUNT];
847 void *kp_free_stack[KS_ADDRS_COUNT];
849 void __kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab);
852 #ifdef CONFIG_HAVE_HARDENED_USERCOPY_ALLOCATOR
853 void __check_heap_object(const void *ptr, unsigned long n,
854 const struct slab *slab, bool to_user);
857 void __check_heap_object(const void *ptr, unsigned long n,
858 const struct slab *slab, bool to_user)
863 #endif /* MM_SLAB_H */