1 /* SPDX-License-Identifier: GPL-2.0 */
5 #include <linux/reciprocal_div.h>
6 #include <linux/list_lru.h>
7 #include <linux/local_lock.h>
8 #include <linux/random.h>
9 #include <linux/kobject.h>
10 #include <linux/sched/mm.h>
11 #include <linux/memcontrol.h>
12 #include <linux/kfence.h>
13 #include <linux/kasan.h>
16 * Internal slab definitions
20 # ifdef system_has_cmpxchg128
21 # define system_has_freelist_aba() system_has_cmpxchg128()
22 # define try_cmpxchg_freelist try_cmpxchg128
24 #define this_cpu_try_cmpxchg_freelist this_cpu_try_cmpxchg128
25 typedef u128 freelist_full_t;
26 #else /* CONFIG_64BIT */
27 # ifdef system_has_cmpxchg64
28 # define system_has_freelist_aba() system_has_cmpxchg64()
29 # define try_cmpxchg_freelist try_cmpxchg64
31 #define this_cpu_try_cmpxchg_freelist this_cpu_try_cmpxchg64
32 typedef u64 freelist_full_t;
33 #endif /* CONFIG_64BIT */
35 #if defined(system_has_freelist_aba) && !defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
36 #undef system_has_freelist_aba
40 * Freelist pointer and counter to cmpxchg together, avoids the typical ABA
41 * problems with cmpxchg of just a pointer.
46 unsigned long counter;
51 /* Reuses the bits in struct page */
53 unsigned long __page_flags;
55 struct kmem_cache *slab_cache;
59 struct list_head slab_list;
60 #ifdef CONFIG_SLUB_CPU_PARTIAL
63 int slabs; /* Nr of slabs left */
67 /* Double-word boundary */
70 void *freelist; /* first free object */
72 unsigned long counters;
80 #ifdef system_has_freelist_aba
81 freelist_aba_t freelist_counter;
85 struct rcu_head rcu_head;
87 unsigned int __unused;
89 atomic_t __page_refcount;
91 unsigned long memcg_data;
95 #define SLAB_MATCH(pg, sl) \
96 static_assert(offsetof(struct page, pg) == offsetof(struct slab, sl))
97 SLAB_MATCH(flags, __page_flags);
98 SLAB_MATCH(compound_head, slab_cache); /* Ensure bit 0 is clear */
99 SLAB_MATCH(_refcount, __page_refcount);
101 SLAB_MATCH(memcg_data, memcg_data);
104 static_assert(sizeof(struct slab) <= sizeof(struct page));
105 #if defined(system_has_freelist_aba)
106 static_assert(IS_ALIGNED(offsetof(struct slab, freelist), sizeof(freelist_aba_t)));
110 * folio_slab - Converts from folio to slab.
113 * Currently struct slab is a different representation of a folio where
114 * folio_test_slab() is true.
116 * Return: The slab which contains this folio.
118 #define folio_slab(folio) (_Generic((folio), \
119 const struct folio *: (const struct slab *)(folio), \
120 struct folio *: (struct slab *)(folio)))
123 * slab_folio - The folio allocated for a slab
126 * Slabs are allocated as folios that contain the individual objects and are
127 * using some fields in the first struct page of the folio - those fields are
128 * now accessed by struct slab. It is occasionally necessary to convert back to
129 * a folio in order to communicate with the rest of the mm. Please use this
130 * helper function instead of casting yourself, as the implementation may change
133 #define slab_folio(s) (_Generic((s), \
134 const struct slab *: (const struct folio *)s, \
135 struct slab *: (struct folio *)s))
138 * page_slab - Converts from first struct page to slab.
139 * @p: The first (either head of compound or single) page of slab.
141 * A temporary wrapper to convert struct page to struct slab in situations where
142 * we know the page is the compound head, or single order-0 page.
144 * Long-term ideally everything would work with struct slab directly or go
145 * through folio to struct slab.
147 * Return: The slab which contains this page
149 #define page_slab(p) (_Generic((p), \
150 const struct page *: (const struct slab *)(p), \
151 struct page *: (struct slab *)(p)))
154 * slab_page - The first struct page allocated for a slab
157 * A convenience wrapper for converting slab to the first struct page of the
158 * underlying folio, to communicate with code not yet converted to folio or
161 #define slab_page(s) folio_page(slab_folio(s), 0)
164 * If network-based swap is enabled, sl*b must keep track of whether pages
165 * were allocated from pfmemalloc reserves.
167 static inline bool slab_test_pfmemalloc(const struct slab *slab)
169 return folio_test_active((struct folio *)slab_folio(slab));
172 static inline void slab_set_pfmemalloc(struct slab *slab)
174 folio_set_active(slab_folio(slab));
177 static inline void slab_clear_pfmemalloc(struct slab *slab)
179 folio_clear_active(slab_folio(slab));
182 static inline void __slab_clear_pfmemalloc(struct slab *slab)
184 __folio_clear_active(slab_folio(slab));
187 static inline void *slab_address(const struct slab *slab)
189 return folio_address(slab_folio(slab));
192 static inline int slab_nid(const struct slab *slab)
194 return folio_nid(slab_folio(slab));
197 static inline pg_data_t *slab_pgdat(const struct slab *slab)
199 return folio_pgdat(slab_folio(slab));
202 static inline struct slab *virt_to_slab(const void *addr)
204 struct folio *folio = virt_to_folio(addr);
206 if (!folio_test_slab(folio))
209 return folio_slab(folio);
212 static inline int slab_order(const struct slab *slab)
214 return folio_order((struct folio *)slab_folio(slab));
217 static inline size_t slab_size(const struct slab *slab)
219 return PAGE_SIZE << slab_order(slab);
222 #ifdef CONFIG_SLUB_CPU_PARTIAL
223 #define slub_percpu_partial(c) ((c)->partial)
225 #define slub_set_percpu_partial(c, p) \
227 slub_percpu_partial(c) = (p)->next; \
230 #define slub_percpu_partial_read_once(c) READ_ONCE(slub_percpu_partial(c))
232 #define slub_percpu_partial(c) NULL
234 #define slub_set_percpu_partial(c, p)
236 #define slub_percpu_partial_read_once(c) NULL
237 #endif // CONFIG_SLUB_CPU_PARTIAL
240 * Word size structure that can be atomically updated or read and that
241 * contains both the order and the number of objects that a slab of the
242 * given order would contain.
244 struct kmem_cache_order_objects {
249 * Slab cache management.
252 #ifndef CONFIG_SLUB_TINY
253 struct kmem_cache_cpu __percpu *cpu_slab;
255 /* Used for retrieving partial slabs, etc. */
257 unsigned long min_partial;
258 unsigned int size; /* Object size including metadata */
259 unsigned int object_size; /* Object size without metadata */
260 struct reciprocal_value reciprocal_size;
261 unsigned int offset; /* Free pointer offset */
262 #ifdef CONFIG_SLUB_CPU_PARTIAL
263 /* Number of per cpu partial objects to keep around */
264 unsigned int cpu_partial;
265 /* Number of per cpu partial slabs to keep around */
266 unsigned int cpu_partial_slabs;
268 struct kmem_cache_order_objects oo;
270 /* Allocation and freeing of slabs */
271 struct kmem_cache_order_objects min;
272 gfp_t allocflags; /* gfp flags to use on each alloc */
273 int refcount; /* Refcount for slab cache destroy */
274 void (*ctor)(void *object); /* Object constructor */
275 unsigned int inuse; /* Offset to metadata */
276 unsigned int align; /* Alignment */
277 unsigned int red_left_pad; /* Left redzone padding size */
278 const char *name; /* Name (only for display!) */
279 struct list_head list; /* List of slab caches */
281 struct kobject kobj; /* For sysfs */
283 #ifdef CONFIG_SLAB_FREELIST_HARDENED
284 unsigned long random;
289 * Defragmentation by allocating from a remote node.
291 unsigned int remote_node_defrag_ratio;
294 #ifdef CONFIG_SLAB_FREELIST_RANDOM
295 unsigned int *random_seq;
298 #ifdef CONFIG_KASAN_GENERIC
299 struct kasan_cache kasan_info;
302 #ifdef CONFIG_HARDENED_USERCOPY
303 unsigned int useroffset; /* Usercopy region offset */
304 unsigned int usersize; /* Usercopy region size */
307 struct kmem_cache_node *node[MAX_NUMNODES];
310 #if defined(CONFIG_SYSFS) && !defined(CONFIG_SLUB_TINY)
311 #define SLAB_SUPPORTS_SYSFS
312 void sysfs_slab_unlink(struct kmem_cache *s);
313 void sysfs_slab_release(struct kmem_cache *s);
315 static inline void sysfs_slab_unlink(struct kmem_cache *s) { }
316 static inline void sysfs_slab_release(struct kmem_cache *s) { }
319 void *fixup_red_left(struct kmem_cache *s, void *p);
321 static inline void *nearest_obj(struct kmem_cache *cache,
322 const struct slab *slab, void *x)
324 void *object = x - (x - slab_address(slab)) % cache->size;
325 void *last_object = slab_address(slab) +
326 (slab->objects - 1) * cache->size;
327 void *result = (unlikely(object > last_object)) ? last_object : object;
329 result = fixup_red_left(cache, result);
333 /* Determine object index from a given position */
334 static inline unsigned int __obj_to_index(const struct kmem_cache *cache,
335 void *addr, void *obj)
337 return reciprocal_divide(kasan_reset_tag(obj) - addr,
338 cache->reciprocal_size);
341 static inline unsigned int obj_to_index(const struct kmem_cache *cache,
342 const struct slab *slab, void *obj)
344 if (is_kfence_address(obj))
346 return __obj_to_index(cache, slab_address(slab), obj);
349 static inline int objs_per_slab(const struct kmem_cache *cache,
350 const struct slab *slab)
352 return slab->objects;
356 * State of the slab allocator.
358 * This is used to describe the states of the allocator during bootup.
359 * Allocators use this to gradually bootstrap themselves. Most allocators
360 * have the problem that the structures used for managing slab caches are
361 * allocated from slab caches themselves.
364 DOWN, /* No slab functionality yet */
365 PARTIAL, /* SLUB: kmem_cache_node available */
366 PARTIAL_NODE, /* SLAB: kmalloc size for node struct available */
367 UP, /* Slab caches usable but not all extras yet */
368 FULL /* Everything is working */
371 extern enum slab_state slab_state;
373 /* The slab cache mutex protects the management structures during changes */
374 extern struct mutex slab_mutex;
376 /* The list of all slab caches on the system */
377 extern struct list_head slab_caches;
379 /* The slab cache that manages slab cache information */
380 extern struct kmem_cache *kmem_cache;
382 /* A table of kmalloc cache names and sizes */
383 extern const struct kmalloc_info_struct {
384 const char *name[NR_KMALLOC_TYPES];
388 /* Kmalloc array related functions */
389 void setup_kmalloc_cache_index_table(void);
390 void create_kmalloc_caches(slab_flags_t);
392 extern u8 kmalloc_size_index[24];
394 static inline unsigned int size_index_elem(unsigned int bytes)
396 return (bytes - 1) / 8;
400 * Find the kmem_cache structure that serves a given size of
403 * This assumes size is larger than zero and not larger than
404 * KMALLOC_MAX_CACHE_SIZE and the caller must check that.
406 static inline struct kmem_cache *
407 kmalloc_slab(size_t size, gfp_t flags, unsigned long caller)
412 index = kmalloc_size_index[size_index_elem(size)];
414 index = fls(size - 1);
416 return kmalloc_caches[kmalloc_type(flags, caller)][index];
419 gfp_t kmalloc_fix_flags(gfp_t flags);
421 /* Functions provided by the slab allocators */
422 int __kmem_cache_create(struct kmem_cache *, slab_flags_t flags);
424 void __init kmem_cache_init(void);
425 void __init new_kmalloc_cache(int idx, enum kmalloc_cache_type type,
427 extern void create_boot_cache(struct kmem_cache *, const char *name,
428 unsigned int size, slab_flags_t flags,
429 unsigned int useroffset, unsigned int usersize);
431 int slab_unmergeable(struct kmem_cache *s);
432 struct kmem_cache *find_mergeable(unsigned size, unsigned align,
433 slab_flags_t flags, const char *name, void (*ctor)(void *));
435 __kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
436 slab_flags_t flags, void (*ctor)(void *));
438 slab_flags_t kmem_cache_flags(unsigned int object_size,
439 slab_flags_t flags, const char *name);
441 static inline bool is_kmalloc_cache(struct kmem_cache *s)
443 return (s->flags & SLAB_KMALLOC);
446 /* Legal flag mask for kmem_cache_create(), for various configurations */
447 #define SLAB_CORE_FLAGS (SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA | \
448 SLAB_CACHE_DMA32 | SLAB_PANIC | \
449 SLAB_TYPESAFE_BY_RCU | SLAB_DEBUG_OBJECTS )
451 #ifdef CONFIG_SLUB_DEBUG
452 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
453 SLAB_TRACE | SLAB_CONSISTENCY_CHECKS)
455 #define SLAB_DEBUG_FLAGS (0)
458 #define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
459 SLAB_TEMPORARY | SLAB_ACCOUNT | \
460 SLAB_NO_USER_FLAGS | SLAB_KMALLOC | SLAB_NO_MERGE)
462 /* Common flags available with current configuration */
463 #define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
465 /* Common flags permitted for kmem_cache_create */
466 #define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \
471 SLAB_CONSISTENCY_CHECKS | \
474 SLAB_RECLAIM_ACCOUNT | \
481 bool __kmem_cache_empty(struct kmem_cache *);
482 int __kmem_cache_shutdown(struct kmem_cache *);
483 void __kmem_cache_release(struct kmem_cache *);
484 int __kmem_cache_shrink(struct kmem_cache *);
485 void slab_kmem_cache_release(struct kmem_cache *);
491 unsigned long active_objs;
492 unsigned long num_objs;
493 unsigned long active_slabs;
494 unsigned long num_slabs;
495 unsigned long shared_avail;
497 unsigned int batchcount;
499 unsigned int objects_per_slab;
500 unsigned int cache_order;
503 void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo);
504 void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s);
505 ssize_t slabinfo_write(struct file *file, const char __user *buffer,
506 size_t count, loff_t *ppos);
508 #ifdef CONFIG_SLUB_DEBUG
509 #ifdef CONFIG_SLUB_DEBUG_ON
510 DECLARE_STATIC_KEY_TRUE(slub_debug_enabled);
512 DECLARE_STATIC_KEY_FALSE(slub_debug_enabled);
514 extern void print_tracking(struct kmem_cache *s, void *object);
515 long validate_slab_cache(struct kmem_cache *s);
516 static inline bool __slub_debug_enabled(void)
518 return static_branch_unlikely(&slub_debug_enabled);
521 static inline void print_tracking(struct kmem_cache *s, void *object)
524 static inline bool __slub_debug_enabled(void)
531 * Returns true if any of the specified slub_debug flags is enabled for the
532 * cache. Use only for flags parsed by setup_slub_debug() as it also enables
535 static inline bool kmem_cache_debug_flags(struct kmem_cache *s, slab_flags_t flags)
537 if (IS_ENABLED(CONFIG_SLUB_DEBUG))
538 VM_WARN_ON_ONCE(!(flags & SLAB_DEBUG_FLAGS));
539 if (__slub_debug_enabled())
540 return s->flags & flags;
544 #ifdef CONFIG_MEMCG_KMEM
546 * slab_objcgs - get the object cgroups vector associated with a slab
547 * @slab: a pointer to the slab struct
549 * Returns a pointer to the object cgroups vector associated with the slab,
550 * or NULL if no such vector has been associated yet.
552 static inline struct obj_cgroup **slab_objcgs(struct slab *slab)
554 unsigned long memcg_data = READ_ONCE(slab->memcg_data);
556 VM_BUG_ON_PAGE(memcg_data && !(memcg_data & MEMCG_DATA_OBJCGS),
558 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_KMEM, slab_page(slab));
560 return (struct obj_cgroup **)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
563 int memcg_alloc_slab_cgroups(struct slab *slab, struct kmem_cache *s,
564 gfp_t gfp, bool new_slab);
565 void mod_objcg_state(struct obj_cgroup *objcg, struct pglist_data *pgdat,
566 enum node_stat_item idx, int nr);
567 #else /* CONFIG_MEMCG_KMEM */
568 static inline struct obj_cgroup **slab_objcgs(struct slab *slab)
573 static inline int memcg_alloc_slab_cgroups(struct slab *slab,
574 struct kmem_cache *s, gfp_t gfp,
579 #endif /* CONFIG_MEMCG_KMEM */
581 size_t __ksize(const void *objp);
583 static inline size_t slab_ksize(const struct kmem_cache *s)
585 #ifdef CONFIG_SLUB_DEBUG
587 * Debugging requires use of the padding between object
588 * and whatever may come after it.
590 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
591 return s->object_size;
593 if (s->flags & SLAB_KASAN)
594 return s->object_size;
596 * If we have the need to store the freelist pointer
597 * back there or track user information then we can
598 * only use the space before that information.
600 if (s->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_STORE_USER))
603 * Else we can use all the padding etc for the allocation
608 #ifdef CONFIG_SLUB_DEBUG
609 void dump_unreclaimable_slab(void);
611 static inline void dump_unreclaimable_slab(void)
616 void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr);
618 #ifdef CONFIG_SLAB_FREELIST_RANDOM
619 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
621 void cache_random_seq_destroy(struct kmem_cache *cachep);
623 static inline int cache_random_seq_create(struct kmem_cache *cachep,
624 unsigned int count, gfp_t gfp)
628 static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { }
629 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
631 static inline bool slab_want_init_on_alloc(gfp_t flags, struct kmem_cache *c)
633 if (static_branch_maybe(CONFIG_INIT_ON_ALLOC_DEFAULT_ON,
637 if (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON))
638 return flags & __GFP_ZERO;
641 return flags & __GFP_ZERO;
644 static inline bool slab_want_init_on_free(struct kmem_cache *c)
646 if (static_branch_maybe(CONFIG_INIT_ON_FREE_DEFAULT_ON,
649 (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)));
653 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG)
654 void debugfs_slab_release(struct kmem_cache *);
656 static inline void debugfs_slab_release(struct kmem_cache *s) { }
660 #define KS_ADDRS_COUNT 16
661 struct kmem_obj_info {
663 struct slab *kp_slab;
665 unsigned long kp_data_offset;
666 struct kmem_cache *kp_slab_cache;
668 void *kp_stack[KS_ADDRS_COUNT];
669 void *kp_free_stack[KS_ADDRS_COUNT];
671 void __kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab);
674 void __check_heap_object(const void *ptr, unsigned long n,
675 const struct slab *slab, bool to_user);
677 #ifdef CONFIG_SLUB_DEBUG
678 void skip_orig_size_check(struct kmem_cache *s, const void *object);
681 #endif /* MM_SLAB_H */