slab.h

  1/* SPDX-License-Identifier: GPL-2.0 */
  2#ifndef MM_SLAB_H
  3#define MM_SLAB_H
  4
  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>
 14
 15/*
 16 * Internal slab definitions
 17 */
 18
 19#ifdef CONFIG_64BIT
 20# ifdef system_has_cmpxchg128
 21# define system_has_freelist_aba()	system_has_cmpxchg128()
 22# define try_cmpxchg_freelist		try_cmpxchg128
 23# endif
 24#define this_cpu_try_cmpxchg_freelist	this_cpu_try_cmpxchg128
 25typedef 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
 30# endif
 31#define this_cpu_try_cmpxchg_freelist	this_cpu_try_cmpxchg64
 32typedef u64 freelist_full_t;
 33#endif /* CONFIG_64BIT */
 34
 35#if defined(system_has_freelist_aba) && !defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
 36#undef system_has_freelist_aba
 37#endif
 38
 39/*
 40 * Freelist pointer and counter to cmpxchg together, avoids the typical ABA
 41 * problems with cmpxchg of just a pointer.
 42 */
 43struct freelist_counters {
 44	union {
 45		struct {
 46			void *freelist;
 47			union {
 48				unsigned long counters;
 49				struct {
 50					unsigned inuse:16;
 51					unsigned objects:15;
 52					/*
 53					 * If slab debugging is enabled then the
 54					 * frozen bit can be reused to indicate
 55					 * that the slab was corrupted
 56					 */
 57					unsigned frozen:1;
 58				};
 59			};
 60		};
 61#ifdef system_has_freelist_aba
 62		freelist_full_t freelist_counters;
 63#endif
 64	};
 65};
 66
 67/* Reuses the bits in struct page */
 68struct slab {
 69	memdesc_flags_t flags;
 70
 71	struct kmem_cache *slab_cache;
 72	union {
 73		struct {
 74			union {
 75				struct list_head slab_list;
 76				struct { /* For deferred deactivate_slab() */
 77					struct llist_node llnode;
 78					void *flush_freelist;
 79				};
 80#ifdef CONFIG_SLUB_CPU_PARTIAL
 81				struct {
 82					struct slab *next;
 83					int slabs;	/* Nr of slabs left */
 84				};
 85#endif
 86			};
 87			/* Double-word boundary */
 88			struct freelist_counters;
 89		};
 90		struct rcu_head rcu_head;
 91	};
 92
 93	unsigned int __page_type;
 94	atomic_t __page_refcount;
 95#ifdef CONFIG_SLAB_OBJ_EXT
 96	unsigned long obj_exts;
 97#endif
 98};
 99
100#define SLAB_MATCH(pg, sl)						\
101	static_assert(offsetof(struct page, pg) == offsetof(struct slab, sl))
102SLAB_MATCH(flags, flags);
103SLAB_MATCH(compound_head, slab_cache);	/* Ensure bit 0 is clear */
104SLAB_MATCH(_refcount, __page_refcount);
105#ifdef CONFIG_MEMCG
106SLAB_MATCH(memcg_data, obj_exts);
107#elif defined(CONFIG_SLAB_OBJ_EXT)
108SLAB_MATCH(_unused_slab_obj_exts, obj_exts);
109#endif
110#undef SLAB_MATCH
111static_assert(sizeof(struct slab) <= sizeof(struct page));
112#if defined(system_has_freelist_aba)
113static_assert(IS_ALIGNED(offsetof(struct slab, freelist), sizeof(struct freelist_counters)));
114#endif
115
116/**
117 * slab_folio - The folio allocated for a slab
118 * @s: The slab.
119 *
120 * Slabs are allocated as folios that contain the individual objects and are
121 * using some fields in the first struct page of the folio - those fields are
122 * now accessed by struct slab. It is occasionally necessary to convert back to
123 * a folio in order to communicate with the rest of the mm.  Please use this
124 * helper function instead of casting yourself, as the implementation may change
125 * in the future.
126 */
127#define slab_folio(s)		(_Generic((s),				\
128	const struct slab *:	(const struct folio *)s,		\
129	struct slab *:		(struct folio *)s))
130
131/**
132 * page_slab - Converts from struct page to its slab.
133 * @page: A page which may or may not belong to a slab.
134 *
135 * Return: The slab which contains this page or NULL if the page does
136 * not belong to a slab.  This includes pages returned from large kmalloc.
137 */
138static inline struct slab *page_slab(const struct page *page)
139{
140	unsigned long head;
141
142	head = READ_ONCE(page->compound_head);
143	if (head & 1)
144		page = (struct page *)(head - 1);
145	if (data_race(page->page_type >> 24) != PGTY_slab)
146		page = NULL;
147
148	return (struct slab *)page;
149}
150
151/**
152 * slab_page - The first struct page allocated for a slab
153 * @s: The slab.
154 *
155 * A convenience wrapper for converting slab to the first struct page of the
156 * underlying folio, to communicate with code not yet converted to folio or
157 * struct slab.
158 */
159#define slab_page(s) folio_page(slab_folio(s), 0)
160
161static inline void *slab_address(const struct slab *slab)
162{
163	return folio_address(slab_folio(slab));
164}
165
166static inline int slab_nid(const struct slab *slab)
167{
168	return memdesc_nid(slab->flags);
169}
170
171static inline pg_data_t *slab_pgdat(const struct slab *slab)
172{
173	return NODE_DATA(slab_nid(slab));
174}
175
176static inline struct slab *virt_to_slab(const void *addr)
177{
178	return page_slab(virt_to_page(addr));
179}
180
181static inline int slab_order(const struct slab *slab)
182{
183	return folio_order(slab_folio(slab));
184}
185
186static inline size_t slab_size(const struct slab *slab)
187{
188	return PAGE_SIZE << slab_order(slab);
189}
190
191#ifdef CONFIG_SLUB_CPU_PARTIAL
192#define slub_percpu_partial(c)			((c)->partial)
193
194#define slub_set_percpu_partial(c, p)		\
195({						\
196	slub_percpu_partial(c) = (p)->next;	\
197})
198
199#define slub_percpu_partial_read_once(c)	READ_ONCE(slub_percpu_partial(c))
200#else
201#define slub_percpu_partial(c)			NULL
202
203#define slub_set_percpu_partial(c, p)
204
205#define slub_percpu_partial_read_once(c)	NULL
206#endif // CONFIG_SLUB_CPU_PARTIAL
207
208/*
209 * Word size structure that can be atomically updated or read and that
210 * contains both the order and the number of objects that a slab of the
211 * given order would contain.
212 */
213struct kmem_cache_order_objects {
214	unsigned int x;
215};
216
217/*
218 * Slab cache management.
219 */
220struct kmem_cache {
221	struct kmem_cache_cpu __percpu *cpu_slab;
222	struct lock_class_key lock_key;
223	struct slub_percpu_sheaves __percpu *cpu_sheaves;
224	/* Used for retrieving partial slabs, etc. */
225	slab_flags_t flags;
226	unsigned long min_partial;
227	unsigned int size;		/* Object size including metadata */
228	unsigned int object_size;	/* Object size without metadata */
229	struct reciprocal_value reciprocal_size;
230	unsigned int offset;		/* Free pointer offset */
231#ifdef CONFIG_SLUB_CPU_PARTIAL
232	/* Number of per cpu partial objects to keep around */
233	unsigned int cpu_partial;
234	/* Number of per cpu partial slabs to keep around */
235	unsigned int cpu_partial_slabs;
236#endif
237	unsigned int sheaf_capacity;
238	struct kmem_cache_order_objects oo;
239
240	/* Allocation and freeing of slabs */
241	struct kmem_cache_order_objects min;
242	gfp_t allocflags;		/* gfp flags to use on each alloc */
243	int refcount;			/* Refcount for slab cache destroy */
244	void (*ctor)(void *object);	/* Object constructor */
245	unsigned int inuse;		/* Offset to metadata */
246	unsigned int align;		/* Alignment */
247	unsigned int red_left_pad;	/* Left redzone padding size */
248	const char *name;		/* Name (only for display!) */
249	struct list_head list;		/* List of slab caches */
250#ifdef CONFIG_SYSFS
251	struct kobject kobj;		/* For sysfs */
252#endif
253#ifdef CONFIG_SLAB_FREELIST_HARDENED
254	unsigned long random;
255#endif
256
257#ifdef CONFIG_NUMA
258	/*
259	 * Defragmentation by allocating from a remote node.
260	 */
261	unsigned int remote_node_defrag_ratio;
262#endif
263
264#ifdef CONFIG_SLAB_FREELIST_RANDOM
265	unsigned int *random_seq;
266#endif
267
268#ifdef CONFIG_KASAN_GENERIC
269	struct kasan_cache kasan_info;
270#endif
271
272#ifdef CONFIG_HARDENED_USERCOPY
273	unsigned int useroffset;	/* Usercopy region offset */
274	unsigned int usersize;		/* Usercopy region size */
275#endif
276
277	struct kmem_cache_node *node[MAX_NUMNODES];
278};
279
280#if defined(CONFIG_SYSFS) && !defined(CONFIG_SLUB_TINY)
281#define SLAB_SUPPORTS_SYSFS 1
282void sysfs_slab_unlink(struct kmem_cache *s);
283void sysfs_slab_release(struct kmem_cache *s);
284#else
285static inline void sysfs_slab_unlink(struct kmem_cache *s) { }
286static inline void sysfs_slab_release(struct kmem_cache *s) { }
287#endif
288
289void *fixup_red_left(struct kmem_cache *s, void *p);
290
291static inline void *nearest_obj(struct kmem_cache *cache,
292				const struct slab *slab, void *x)
293{
294	void *object = x - (x - slab_address(slab)) % cache->size;
295	void *last_object = slab_address(slab) +
296		(slab->objects - 1) * cache->size;
297	void *result = (unlikely(object > last_object)) ? last_object : object;
298
299	result = fixup_red_left(cache, result);
300	return result;
301}
302
303/* Determine object index from a given position */
304static inline unsigned int __obj_to_index(const struct kmem_cache *cache,
305					  void *addr, void *obj)
306{
307	return reciprocal_divide(kasan_reset_tag(obj) - addr,
308				 cache->reciprocal_size);
309}
310
311static inline unsigned int obj_to_index(const struct kmem_cache *cache,
312					const struct slab *slab, void *obj)
313{
314	if (is_kfence_address(obj))
315		return 0;
316	return __obj_to_index(cache, slab_address(slab), obj);
317}
318
319static inline int objs_per_slab(const struct kmem_cache *cache,
320				const struct slab *slab)
321{
322	return slab->objects;
323}
324
325/*
326 * State of the slab allocator.
327 *
328 * This is used to describe the states of the allocator during bootup.
329 * Allocators use this to gradually bootstrap themselves. Most allocators
330 * have the problem that the structures used for managing slab caches are
331 * allocated from slab caches themselves.
332 */
333enum slab_state {
334	DOWN,			/* No slab functionality yet */
335	PARTIAL,		/* SLUB: kmem_cache_node available */
336	UP,			/* Slab caches usable but not all extras yet */
337	FULL			/* Everything is working */
338};
339
340extern enum slab_state slab_state;
341
342/* The slab cache mutex protects the management structures during changes */
343extern struct mutex slab_mutex;
344
345/* The list of all slab caches on the system */
346extern struct list_head slab_caches;
347
348/* The slab cache that manages slab cache information */
349extern struct kmem_cache *kmem_cache;
350
351/* A table of kmalloc cache names and sizes */
352extern const struct kmalloc_info_struct {
353	const char *name[NR_KMALLOC_TYPES];
354	unsigned int size;
355} kmalloc_info[];
356
357/* Kmalloc array related functions */
358void setup_kmalloc_cache_index_table(void);
359void create_kmalloc_caches(void);
360
361extern u8 kmalloc_size_index[24];
362
363static inline unsigned int size_index_elem(unsigned int bytes)
364{
365	return (bytes - 1) / 8;
366}
367
368/*
369 * Find the kmem_cache structure that serves a given size of
370 * allocation
371 *
372 * This assumes size is larger than zero and not larger than
373 * KMALLOC_MAX_CACHE_SIZE and the caller must check that.
374 */
375static inline struct kmem_cache *
376kmalloc_slab(size_t size, kmem_buckets *b, gfp_t flags, unsigned long caller)
377{
378	unsigned int index;
379
380	if (!b)
381		b = &kmalloc_caches[kmalloc_type(flags, caller)];
382	if (size <= 192)
383		index = kmalloc_size_index[size_index_elem(size)];
384	else
385		index = fls(size - 1);
386
387	return (*b)[index];
388}
389
390gfp_t kmalloc_fix_flags(gfp_t flags);
391
392/* Functions provided by the slab allocators */
393int do_kmem_cache_create(struct kmem_cache *s, const char *name,
394			 unsigned int size, struct kmem_cache_args *args,
395			 slab_flags_t flags);
396
397void __init kmem_cache_init(void);
398extern void create_boot_cache(struct kmem_cache *, const char *name,
399			unsigned int size, slab_flags_t flags,
400			unsigned int useroffset, unsigned int usersize);
401
402int slab_unmergeable(struct kmem_cache *s);
403struct kmem_cache *find_mergeable(unsigned size, unsigned align,
404		slab_flags_t flags, const char *name, void (*ctor)(void *));
405struct kmem_cache *
406__kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
407		   slab_flags_t flags, void (*ctor)(void *));
408
409slab_flags_t kmem_cache_flags(slab_flags_t flags, const char *name);
410
411static inline bool is_kmalloc_cache(struct kmem_cache *s)
412{
413	return (s->flags & SLAB_KMALLOC);
414}
415
416static inline bool is_kmalloc_normal(struct kmem_cache *s)
417{
418	if (!is_kmalloc_cache(s))
419		return false;
420	return !(s->flags & (SLAB_CACHE_DMA|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT));
421}
422
423bool __kfree_rcu_sheaf(struct kmem_cache *s, void *obj);
424void flush_all_rcu_sheaves(void);
425void flush_rcu_sheaves_on_cache(struct kmem_cache *s);
426
427#define SLAB_CORE_FLAGS (SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA | \
428			 SLAB_CACHE_DMA32 | SLAB_PANIC | \
429			 SLAB_TYPESAFE_BY_RCU | SLAB_DEBUG_OBJECTS | \
430			 SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
431			 SLAB_TEMPORARY | SLAB_ACCOUNT | \
432			 SLAB_NO_USER_FLAGS | SLAB_KMALLOC | SLAB_NO_MERGE)
433
434#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
435			  SLAB_TRACE | SLAB_CONSISTENCY_CHECKS)
436
437#define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS)
438
439bool __kmem_cache_empty(struct kmem_cache *);
440int __kmem_cache_shutdown(struct kmem_cache *);
441void __kmem_cache_release(struct kmem_cache *);
442int __kmem_cache_shrink(struct kmem_cache *);
443void slab_kmem_cache_release(struct kmem_cache *);
444
445struct seq_file;
446struct file;
447
448struct slabinfo {
449	unsigned long active_objs;
450	unsigned long num_objs;
451	unsigned long active_slabs;
452	unsigned long num_slabs;
453	unsigned long shared_avail;
454	unsigned int limit;
455	unsigned int batchcount;
456	unsigned int shared;
457	unsigned int objects_per_slab;
458	unsigned int cache_order;
459};
460
461void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo);
462
463#ifdef CONFIG_SLUB_DEBUG
464#ifdef CONFIG_SLUB_DEBUG_ON
465DECLARE_STATIC_KEY_TRUE(slub_debug_enabled);
466#else
467DECLARE_STATIC_KEY_FALSE(slub_debug_enabled);
468#endif
469extern void print_tracking(struct kmem_cache *s, void *object);
470long validate_slab_cache(struct kmem_cache *s);
471static inline bool __slub_debug_enabled(void)
472{
473	return static_branch_unlikely(&slub_debug_enabled);
474}
475#else
476static inline void print_tracking(struct kmem_cache *s, void *object)
477{
478}
479static inline bool __slub_debug_enabled(void)
480{
481	return false;
482}
483#endif
484
485/*
486 * Returns true if any of the specified slab_debug flags is enabled for the
487 * cache. Use only for flags parsed by setup_slub_debug() as it also enables
488 * the static key.
489 */
490static inline bool kmem_cache_debug_flags(struct kmem_cache *s, slab_flags_t flags)
491{
492	if (IS_ENABLED(CONFIG_SLUB_DEBUG))
493		VM_WARN_ON_ONCE(!(flags & SLAB_DEBUG_FLAGS));
494	if (__slub_debug_enabled())
495		return s->flags & flags;
496	return false;
497}
498
499#if IS_ENABLED(CONFIG_SLUB_DEBUG) && IS_ENABLED(CONFIG_KUNIT)
500bool slab_in_kunit_test(void);
501#else
502static inline bool slab_in_kunit_test(void) { return false; }
503#endif
504
505#ifdef CONFIG_SLAB_OBJ_EXT
506
507/*
508 * slab_obj_exts - get the pointer to the slab object extension vector
509 * associated with a slab.
510 * @slab: a pointer to the slab struct
511 *
512 * Returns a pointer to the object extension vector associated with the slab,
513 * or NULL if no such vector has been associated yet.
514 */
515static inline struct slabobj_ext *slab_obj_exts(struct slab *slab)
516{
517	unsigned long obj_exts = READ_ONCE(slab->obj_exts);
518
519#ifdef CONFIG_MEMCG
520	/*
521	 * obj_exts should be either NULL, a valid pointer with
522	 * MEMCG_DATA_OBJEXTS bit set or be equal to OBJEXTS_ALLOC_FAIL.
523	 */
524	VM_BUG_ON_PAGE(obj_exts && !(obj_exts & MEMCG_DATA_OBJEXTS) &&
525		       obj_exts != OBJEXTS_ALLOC_FAIL, slab_page(slab));
526	VM_BUG_ON_PAGE(obj_exts & MEMCG_DATA_KMEM, slab_page(slab));
527#endif
528	return (struct slabobj_ext *)(obj_exts & ~OBJEXTS_FLAGS_MASK);
529}
530
531int alloc_slab_obj_exts(struct slab *slab, struct kmem_cache *s,
532                        gfp_t gfp, bool new_slab);
533
534#else /* CONFIG_SLAB_OBJ_EXT */
535
536static inline struct slabobj_ext *slab_obj_exts(struct slab *slab)
537{
538	return NULL;
539}
540
541#endif /* CONFIG_SLAB_OBJ_EXT */
542
543static inline enum node_stat_item cache_vmstat_idx(struct kmem_cache *s)
544{
545	return (s->flags & SLAB_RECLAIM_ACCOUNT) ?
546		NR_SLAB_RECLAIMABLE_B : NR_SLAB_UNRECLAIMABLE_B;
547}
548
549#ifdef CONFIG_MEMCG
550bool __memcg_slab_post_alloc_hook(struct kmem_cache *s, struct list_lru *lru,
551				  gfp_t flags, size_t size, void **p);
552void __memcg_slab_free_hook(struct kmem_cache *s, struct slab *slab,
553			    void **p, int objects, struct slabobj_ext *obj_exts);
554#endif
555
556void kvfree_rcu_cb(struct rcu_head *head);
557
558size_t __ksize(const void *objp);
559
560static inline size_t slab_ksize(const struct kmem_cache *s)
561{
562#ifdef CONFIG_SLUB_DEBUG
563	/*
564	 * Debugging requires use of the padding between object
565	 * and whatever may come after it.
566	 */
567	if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
568		return s->object_size;
569#endif
570	if (s->flags & SLAB_KASAN)
571		return s->object_size;
572	/*
573	 * If we have the need to store the freelist pointer
574	 * back there or track user information then we can
575	 * only use the space before that information.
576	 */
577	if (s->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_STORE_USER))
578		return s->inuse;
579	/*
580	 * Else we can use all the padding etc for the allocation
581	 */
582	return s->size;
583}
584
585static inline unsigned int large_kmalloc_order(const struct page *page)
586{
587	return page[1].flags.f & 0xff;
588}
589
590static inline size_t large_kmalloc_size(const struct page *page)
591{
592	return PAGE_SIZE << large_kmalloc_order(page);
593}
594
595#ifdef CONFIG_SLUB_DEBUG
596void dump_unreclaimable_slab(void);
597#else
598static inline void dump_unreclaimable_slab(void)
599{
600}
601#endif
602
603void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr);
604
605#ifdef CONFIG_SLAB_FREELIST_RANDOM
606int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
607			gfp_t gfp);
608void cache_random_seq_destroy(struct kmem_cache *cachep);
609#else
610static inline int cache_random_seq_create(struct kmem_cache *cachep,
611					unsigned int count, gfp_t gfp)
612{
613	return 0;
614}
615static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { }
616#endif /* CONFIG_SLAB_FREELIST_RANDOM */
617
618static inline bool slab_want_init_on_alloc(gfp_t flags, struct kmem_cache *c)
619{
620	if (static_branch_maybe(CONFIG_INIT_ON_ALLOC_DEFAULT_ON,
621				&init_on_alloc)) {
622		if (c->ctor)
623			return false;
624		if (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON))
625			return flags & __GFP_ZERO;
626		return true;
627	}
628	return flags & __GFP_ZERO;
629}
630
631static inline bool slab_want_init_on_free(struct kmem_cache *c)
632{
633	if (static_branch_maybe(CONFIG_INIT_ON_FREE_DEFAULT_ON,
634				&init_on_free))
635		return !(c->ctor ||
636			 (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)));
637	return false;
638}
639
640#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG)
641void debugfs_slab_release(struct kmem_cache *);
642#else
643static inline void debugfs_slab_release(struct kmem_cache *s) { }
644#endif
645
646#ifdef CONFIG_PRINTK
647#define KS_ADDRS_COUNT 16
648struct kmem_obj_info {
649	void *kp_ptr;
650	struct slab *kp_slab;
651	void *kp_objp;
652	unsigned long kp_data_offset;
653	struct kmem_cache *kp_slab_cache;
654	void *kp_ret;
655	void *kp_stack[KS_ADDRS_COUNT];
656	void *kp_free_stack[KS_ADDRS_COUNT];
657};
658void __kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab);
659#endif
660
661void __check_heap_object(const void *ptr, unsigned long n,
662			 const struct slab *slab, bool to_user);
663
664void defer_free_barrier(void);
665
666static inline bool slub_debug_orig_size(struct kmem_cache *s)
667{
668	return (kmem_cache_debug_flags(s, SLAB_STORE_USER) &&
669			(s->flags & SLAB_KMALLOC));
670}
671
672#ifdef CONFIG_SLUB_DEBUG
673void skip_orig_size_check(struct kmem_cache *s, const void *object);
674#endif
675
676#endif /* MM_SLAB_H */