Loading...
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 */
43typedef union {
44 struct {
45 void *freelist;
46 unsigned long counter;
47 };
48 freelist_full_t full;
49} freelist_aba_t;
50
51/* Reuses the bits in struct page */
52struct slab {
53 unsigned long __page_flags;
54
55 struct kmem_cache *slab_cache;
56 union {
57 struct {
58 union {
59 struct list_head slab_list;
60#ifdef CONFIG_SLUB_CPU_PARTIAL
61 struct {
62 struct slab *next;
63 int slabs; /* Nr of slabs left */
64 };
65#endif
66 };
67 /* Double-word boundary */
68 union {
69 struct {
70 void *freelist; /* first free object */
71 union {
72 unsigned long counters;
73 struct {
74 unsigned inuse:16;
75 unsigned objects:15;
76 unsigned frozen:1;
77 };
78 };
79 };
80#ifdef system_has_freelist_aba
81 freelist_aba_t freelist_counter;
82#endif
83 };
84 };
85 struct rcu_head rcu_head;
86 };
87 unsigned int __unused;
88
89 atomic_t __page_refcount;
90#ifdef CONFIG_MEMCG
91 unsigned long memcg_data;
92#endif
93};
94
95#define SLAB_MATCH(pg, sl) \
96 static_assert(offsetof(struct page, pg) == offsetof(struct slab, sl))
97SLAB_MATCH(flags, __page_flags);
98SLAB_MATCH(compound_head, slab_cache); /* Ensure bit 0 is clear */
99SLAB_MATCH(_refcount, __page_refcount);
100#ifdef CONFIG_MEMCG
101SLAB_MATCH(memcg_data, memcg_data);
102#endif
103#undef SLAB_MATCH
104static_assert(sizeof(struct slab) <= sizeof(struct page));
105#if defined(system_has_freelist_aba)
106static_assert(IS_ALIGNED(offsetof(struct slab, freelist), sizeof(freelist_aba_t)));
107#endif
108
109/**
110 * folio_slab - Converts from folio to slab.
111 * @folio: The folio.
112 *
113 * Currently struct slab is a different representation of a folio where
114 * folio_test_slab() is true.
115 *
116 * Return: The slab which contains this folio.
117 */
118#define folio_slab(folio) (_Generic((folio), \
119 const struct folio *: (const struct slab *)(folio), \
120 struct folio *: (struct slab *)(folio)))
121
122/**
123 * slab_folio - The folio allocated for a slab
124 * @slab: The slab.
125 *
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
131 * in the future.
132 */
133#define slab_folio(s) (_Generic((s), \
134 const struct slab *: (const struct folio *)s, \
135 struct slab *: (struct folio *)s))
136
137/**
138 * page_slab - Converts from first struct page to slab.
139 * @p: The first (either head of compound or single) page of slab.
140 *
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.
143 *
144 * Long-term ideally everything would work with struct slab directly or go
145 * through folio to struct slab.
146 *
147 * Return: The slab which contains this page
148 */
149#define page_slab(p) (_Generic((p), \
150 const struct page *: (const struct slab *)(p), \
151 struct page *: (struct slab *)(p)))
152
153/**
154 * slab_page - The first struct page allocated for a slab
155 * @slab: The slab.
156 *
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
159 * struct slab.
160 */
161#define slab_page(s) folio_page(slab_folio(s), 0)
162
163/*
164 * If network-based swap is enabled, sl*b must keep track of whether pages
165 * were allocated from pfmemalloc reserves.
166 */
167static inline bool slab_test_pfmemalloc(const struct slab *slab)
168{
169 return folio_test_active((struct folio *)slab_folio(slab));
170}
171
172static inline void slab_set_pfmemalloc(struct slab *slab)
173{
174 folio_set_active(slab_folio(slab));
175}
176
177static inline void slab_clear_pfmemalloc(struct slab *slab)
178{
179 folio_clear_active(slab_folio(slab));
180}
181
182static inline void __slab_clear_pfmemalloc(struct slab *slab)
183{
184 __folio_clear_active(slab_folio(slab));
185}
186
187static inline void *slab_address(const struct slab *slab)
188{
189 return folio_address(slab_folio(slab));
190}
191
192static inline int slab_nid(const struct slab *slab)
193{
194 return folio_nid(slab_folio(slab));
195}
196
197static inline pg_data_t *slab_pgdat(const struct slab *slab)
198{
199 return folio_pgdat(slab_folio(slab));
200}
201
202static inline struct slab *virt_to_slab(const void *addr)
203{
204 struct folio *folio = virt_to_folio(addr);
205
206 if (!folio_test_slab(folio))
207 return NULL;
208
209 return folio_slab(folio);
210}
211
212static inline int slab_order(const struct slab *slab)
213{
214 return folio_order((struct folio *)slab_folio(slab));
215}
216
217static inline size_t slab_size(const struct slab *slab)
218{
219 return PAGE_SIZE << slab_order(slab);
220}
221
222#ifdef CONFIG_SLUB_CPU_PARTIAL
223#define slub_percpu_partial(c) ((c)->partial)
224
225#define slub_set_percpu_partial(c, p) \
226({ \
227 slub_percpu_partial(c) = (p)->next; \
228})
229
230#define slub_percpu_partial_read_once(c) READ_ONCE(slub_percpu_partial(c))
231#else
232#define slub_percpu_partial(c) NULL
233
234#define slub_set_percpu_partial(c, p)
235
236#define slub_percpu_partial_read_once(c) NULL
237#endif // CONFIG_SLUB_CPU_PARTIAL
238
239/*
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.
243 */
244struct kmem_cache_order_objects {
245 unsigned int x;
246};
247
248/*
249 * Slab cache management.
250 */
251struct kmem_cache {
252#ifndef CONFIG_SLUB_TINY
253 struct kmem_cache_cpu __percpu *cpu_slab;
254#endif
255 /* Used for retrieving partial slabs, etc. */
256 slab_flags_t flags;
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;
267#endif
268 struct kmem_cache_order_objects oo;
269
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 */
280#ifdef CONFIG_SYSFS
281 struct kobject kobj; /* For sysfs */
282#endif
283#ifdef CONFIG_SLAB_FREELIST_HARDENED
284 unsigned long random;
285#endif
286
287#ifdef CONFIG_NUMA
288 /*
289 * Defragmentation by allocating from a remote node.
290 */
291 unsigned int remote_node_defrag_ratio;
292#endif
293
294#ifdef CONFIG_SLAB_FREELIST_RANDOM
295 unsigned int *random_seq;
296#endif
297
298#ifdef CONFIG_KASAN_GENERIC
299 struct kasan_cache kasan_info;
300#endif
301
302#ifdef CONFIG_HARDENED_USERCOPY
303 unsigned int useroffset; /* Usercopy region offset */
304 unsigned int usersize; /* Usercopy region size */
305#endif
306
307 struct kmem_cache_node *node[MAX_NUMNODES];
308};
309
310#if defined(CONFIG_SYSFS) && !defined(CONFIG_SLUB_TINY)
311#define SLAB_SUPPORTS_SYSFS
312void sysfs_slab_unlink(struct kmem_cache *s);
313void sysfs_slab_release(struct kmem_cache *s);
314#else
315static inline void sysfs_slab_unlink(struct kmem_cache *s) { }
316static inline void sysfs_slab_release(struct kmem_cache *s) { }
317#endif
318
319void *fixup_red_left(struct kmem_cache *s, void *p);
320
321static inline void *nearest_obj(struct kmem_cache *cache,
322 const struct slab *slab, void *x)
323{
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;
328
329 result = fixup_red_left(cache, result);
330 return result;
331}
332
333/* Determine object index from a given position */
334static inline unsigned int __obj_to_index(const struct kmem_cache *cache,
335 void *addr, void *obj)
336{
337 return reciprocal_divide(kasan_reset_tag(obj) - addr,
338 cache->reciprocal_size);
339}
340
341static inline unsigned int obj_to_index(const struct kmem_cache *cache,
342 const struct slab *slab, void *obj)
343{
344 if (is_kfence_address(obj))
345 return 0;
346 return __obj_to_index(cache, slab_address(slab), obj);
347}
348
349static inline int objs_per_slab(const struct kmem_cache *cache,
350 const struct slab *slab)
351{
352 return slab->objects;
353}
354
355/*
356 * State of the slab allocator.
357 *
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.
362 */
363enum slab_state {
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 */
369};
370
371extern enum slab_state slab_state;
372
373/* The slab cache mutex protects the management structures during changes */
374extern struct mutex slab_mutex;
375
376/* The list of all slab caches on the system */
377extern struct list_head slab_caches;
378
379/* The slab cache that manages slab cache information */
380extern struct kmem_cache *kmem_cache;
381
382/* A table of kmalloc cache names and sizes */
383extern const struct kmalloc_info_struct {
384 const char *name[NR_KMALLOC_TYPES];
385 unsigned int size;
386} kmalloc_info[];
387
388/* Kmalloc array related functions */
389void setup_kmalloc_cache_index_table(void);
390void create_kmalloc_caches(slab_flags_t);
391
392extern u8 kmalloc_size_index[24];
393
394static inline unsigned int size_index_elem(unsigned int bytes)
395{
396 return (bytes - 1) / 8;
397}
398
399/*
400 * Find the kmem_cache structure that serves a given size of
401 * allocation
402 *
403 * This assumes size is larger than zero and not larger than
404 * KMALLOC_MAX_CACHE_SIZE and the caller must check that.
405 */
406static inline struct kmem_cache *
407kmalloc_slab(size_t size, gfp_t flags, unsigned long caller)
408{
409 unsigned int index;
410
411 if (size <= 192)
412 index = kmalloc_size_index[size_index_elem(size)];
413 else
414 index = fls(size - 1);
415
416 return kmalloc_caches[kmalloc_type(flags, caller)][index];
417}
418
419gfp_t kmalloc_fix_flags(gfp_t flags);
420
421/* Functions provided by the slab allocators */
422int __kmem_cache_create(struct kmem_cache *, slab_flags_t flags);
423
424void __init kmem_cache_init(void);
425void __init new_kmalloc_cache(int idx, enum kmalloc_cache_type type,
426 slab_flags_t flags);
427extern 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);
430
431int slab_unmergeable(struct kmem_cache *s);
432struct kmem_cache *find_mergeable(unsigned size, unsigned align,
433 slab_flags_t flags, const char *name, void (*ctor)(void *));
434struct kmem_cache *
435__kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
436 slab_flags_t flags, void (*ctor)(void *));
437
438slab_flags_t kmem_cache_flags(unsigned int object_size,
439 slab_flags_t flags, const char *name);
440
441static inline bool is_kmalloc_cache(struct kmem_cache *s)
442{
443 return (s->flags & SLAB_KMALLOC);
444}
445
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 )
450
451#ifdef CONFIG_SLUB_DEBUG
452#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
453 SLAB_TRACE | SLAB_CONSISTENCY_CHECKS)
454#else
455#define SLAB_DEBUG_FLAGS (0)
456#endif
457
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)
461
462/* Common flags available with current configuration */
463#define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
464
465/* Common flags permitted for kmem_cache_create */
466#define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \
467 SLAB_RED_ZONE | \
468 SLAB_POISON | \
469 SLAB_STORE_USER | \
470 SLAB_TRACE | \
471 SLAB_CONSISTENCY_CHECKS | \
472 SLAB_MEM_SPREAD | \
473 SLAB_NOLEAKTRACE | \
474 SLAB_RECLAIM_ACCOUNT | \
475 SLAB_TEMPORARY | \
476 SLAB_ACCOUNT | \
477 SLAB_KMALLOC | \
478 SLAB_NO_MERGE | \
479 SLAB_NO_USER_FLAGS)
480
481bool __kmem_cache_empty(struct kmem_cache *);
482int __kmem_cache_shutdown(struct kmem_cache *);
483void __kmem_cache_release(struct kmem_cache *);
484int __kmem_cache_shrink(struct kmem_cache *);
485void slab_kmem_cache_release(struct kmem_cache *);
486
487struct seq_file;
488struct file;
489
490struct slabinfo {
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;
496 unsigned int limit;
497 unsigned int batchcount;
498 unsigned int shared;
499 unsigned int objects_per_slab;
500 unsigned int cache_order;
501};
502
503void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo);
504void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s);
505ssize_t slabinfo_write(struct file *file, const char __user *buffer,
506 size_t count, loff_t *ppos);
507
508#ifdef CONFIG_SLUB_DEBUG
509#ifdef CONFIG_SLUB_DEBUG_ON
510DECLARE_STATIC_KEY_TRUE(slub_debug_enabled);
511#else
512DECLARE_STATIC_KEY_FALSE(slub_debug_enabled);
513#endif
514extern void print_tracking(struct kmem_cache *s, void *object);
515long validate_slab_cache(struct kmem_cache *s);
516static inline bool __slub_debug_enabled(void)
517{
518 return static_branch_unlikely(&slub_debug_enabled);
519}
520#else
521static inline void print_tracking(struct kmem_cache *s, void *object)
522{
523}
524static inline bool __slub_debug_enabled(void)
525{
526 return false;
527}
528#endif
529
530/*
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
533 * the static key.
534 */
535static inline bool kmem_cache_debug_flags(struct kmem_cache *s, slab_flags_t flags)
536{
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;
541 return false;
542}
543
544#ifdef CONFIG_MEMCG_KMEM
545/*
546 * slab_objcgs - get the object cgroups vector associated with a slab
547 * @slab: a pointer to the slab struct
548 *
549 * Returns a pointer to the object cgroups vector associated with the slab,
550 * or NULL if no such vector has been associated yet.
551 */
552static inline struct obj_cgroup **slab_objcgs(struct slab *slab)
553{
554 unsigned long memcg_data = READ_ONCE(slab->memcg_data);
555
556 VM_BUG_ON_PAGE(memcg_data && !(memcg_data & MEMCG_DATA_OBJCGS),
557 slab_page(slab));
558 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_KMEM, slab_page(slab));
559
560 return (struct obj_cgroup **)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
561}
562
563int memcg_alloc_slab_cgroups(struct slab *slab, struct kmem_cache *s,
564 gfp_t gfp, bool new_slab);
565void mod_objcg_state(struct obj_cgroup *objcg, struct pglist_data *pgdat,
566 enum node_stat_item idx, int nr);
567#else /* CONFIG_MEMCG_KMEM */
568static inline struct obj_cgroup **slab_objcgs(struct slab *slab)
569{
570 return NULL;
571}
572
573static inline int memcg_alloc_slab_cgroups(struct slab *slab,
574 struct kmem_cache *s, gfp_t gfp,
575 bool new_slab)
576{
577 return 0;
578}
579#endif /* CONFIG_MEMCG_KMEM */
580
581size_t __ksize(const void *objp);
582
583static inline size_t slab_ksize(const struct kmem_cache *s)
584{
585#ifdef CONFIG_SLUB_DEBUG
586 /*
587 * Debugging requires use of the padding between object
588 * and whatever may come after it.
589 */
590 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
591 return s->object_size;
592#endif
593 if (s->flags & SLAB_KASAN)
594 return s->object_size;
595 /*
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.
599 */
600 if (s->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_STORE_USER))
601 return s->inuse;
602 /*
603 * Else we can use all the padding etc for the allocation
604 */
605 return s->size;
606}
607
608#ifdef CONFIG_SLUB_DEBUG
609void dump_unreclaimable_slab(void);
610#else
611static inline void dump_unreclaimable_slab(void)
612{
613}
614#endif
615
616void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr);
617
618#ifdef CONFIG_SLAB_FREELIST_RANDOM
619int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
620 gfp_t gfp);
621void cache_random_seq_destroy(struct kmem_cache *cachep);
622#else
623static inline int cache_random_seq_create(struct kmem_cache *cachep,
624 unsigned int count, gfp_t gfp)
625{
626 return 0;
627}
628static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { }
629#endif /* CONFIG_SLAB_FREELIST_RANDOM */
630
631static inline bool slab_want_init_on_alloc(gfp_t flags, struct kmem_cache *c)
632{
633 if (static_branch_maybe(CONFIG_INIT_ON_ALLOC_DEFAULT_ON,
634 &init_on_alloc)) {
635 if (c->ctor)
636 return false;
637 if (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON))
638 return flags & __GFP_ZERO;
639 return true;
640 }
641 return flags & __GFP_ZERO;
642}
643
644static inline bool slab_want_init_on_free(struct kmem_cache *c)
645{
646 if (static_branch_maybe(CONFIG_INIT_ON_FREE_DEFAULT_ON,
647 &init_on_free))
648 return !(c->ctor ||
649 (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)));
650 return false;
651}
652
653#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG)
654void debugfs_slab_release(struct kmem_cache *);
655#else
656static inline void debugfs_slab_release(struct kmem_cache *s) { }
657#endif
658
659#ifdef CONFIG_PRINTK
660#define KS_ADDRS_COUNT 16
661struct kmem_obj_info {
662 void *kp_ptr;
663 struct slab *kp_slab;
664 void *kp_objp;
665 unsigned long kp_data_offset;
666 struct kmem_cache *kp_slab_cache;
667 void *kp_ret;
668 void *kp_stack[KS_ADDRS_COUNT];
669 void *kp_free_stack[KS_ADDRS_COUNT];
670};
671void __kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab);
672#endif
673
674void __check_heap_object(const void *ptr, unsigned long n,
675 const struct slab *slab, bool to_user);
676
677#ifdef CONFIG_SLUB_DEBUG
678void skip_orig_size_check(struct kmem_cache *s, const void *object);
679#endif
680
681#endif /* MM_SLAB_H */
1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef MM_SLAB_H
3#define MM_SLAB_H
4/*
5 * Internal slab definitions
6 */
7
8#ifdef CONFIG_SLOB
9/*
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.
14 *
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.
19 */
20struct kmem_cache {
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 */
31};
32
33#endif /* CONFIG_SLOB */
34
35#ifdef CONFIG_SLAB
36#include <linux/slab_def.h>
37#endif
38
39#ifdef CONFIG_SLUB
40#include <linux/slub_def.h>
41#endif
42
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>
49#include <linux/kmemleak.h>
50
51/*
52 * State of the slab allocator.
53 *
54 * This is used to describe the states of the allocator during bootup.
55 * Allocators use this to gradually bootstrap themselves. Most allocators
56 * have the problem that the structures used for managing slab caches are
57 * allocated from slab caches themselves.
58 */
59enum slab_state {
60 DOWN, /* No slab functionality yet */
61 PARTIAL, /* SLUB: kmem_cache_node available */
62 PARTIAL_NODE, /* SLAB: kmalloc size for node struct available */
63 UP, /* Slab caches usable but not all extras yet */
64 FULL /* Everything is working */
65};
66
67extern enum slab_state slab_state;
68
69/* The slab cache mutex protects the management structures during changes */
70extern struct mutex slab_mutex;
71
72/* The list of all slab caches on the system */
73extern struct list_head slab_caches;
74
75/* The slab cache that manages slab cache information */
76extern struct kmem_cache *kmem_cache;
77
78/* A table of kmalloc cache names and sizes */
79extern const struct kmalloc_info_struct {
80 const char *name[NR_KMALLOC_TYPES];
81 unsigned int size;
82} kmalloc_info[];
83
84#ifndef CONFIG_SLOB
85/* Kmalloc array related functions */
86void setup_kmalloc_cache_index_table(void);
87void create_kmalloc_caches(slab_flags_t);
88
89/* Find the kmalloc slab corresponding for a certain size */
90struct kmem_cache *kmalloc_slab(size_t, gfp_t);
91#endif
92
93gfp_t kmalloc_fix_flags(gfp_t flags);
94
95/* Functions provided by the slab allocators */
96int __kmem_cache_create(struct kmem_cache *, slab_flags_t flags);
97
98struct kmem_cache *create_kmalloc_cache(const char *name, unsigned int size,
99 slab_flags_t flags, unsigned int useroffset,
100 unsigned int usersize);
101extern void create_boot_cache(struct kmem_cache *, const char *name,
102 unsigned int size, slab_flags_t flags,
103 unsigned int useroffset, unsigned int usersize);
104
105int slab_unmergeable(struct kmem_cache *s);
106struct kmem_cache *find_mergeable(unsigned size, unsigned align,
107 slab_flags_t flags, const char *name, void (*ctor)(void *));
108#ifndef CONFIG_SLOB
109struct kmem_cache *
110__kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
111 slab_flags_t flags, void (*ctor)(void *));
112
113slab_flags_t kmem_cache_flags(unsigned int object_size,
114 slab_flags_t flags, const char *name,
115 void (*ctor)(void *));
116#else
117static inline struct kmem_cache *
118__kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
119 slab_flags_t flags, void (*ctor)(void *))
120{ return NULL; }
121
122static inline slab_flags_t kmem_cache_flags(unsigned int object_size,
123 slab_flags_t flags, const char *name,
124 void (*ctor)(void *))
125{
126 return flags;
127}
128#endif
129
130
131/* Legal flag mask for kmem_cache_create(), for various configurations */
132#define SLAB_CORE_FLAGS (SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA | \
133 SLAB_CACHE_DMA32 | SLAB_PANIC | \
134 SLAB_TYPESAFE_BY_RCU | SLAB_DEBUG_OBJECTS )
135
136#if defined(CONFIG_DEBUG_SLAB)
137#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
138#elif defined(CONFIG_SLUB_DEBUG)
139#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
140 SLAB_TRACE | SLAB_CONSISTENCY_CHECKS)
141#else
142#define SLAB_DEBUG_FLAGS (0)
143#endif
144
145#if defined(CONFIG_SLAB)
146#define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \
147 SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \
148 SLAB_ACCOUNT)
149#elif defined(CONFIG_SLUB)
150#define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
151 SLAB_TEMPORARY | SLAB_ACCOUNT)
152#else
153#define SLAB_CACHE_FLAGS (0)
154#endif
155
156/* Common flags available with current configuration */
157#define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
158
159/* Common flags permitted for kmem_cache_create */
160#define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \
161 SLAB_RED_ZONE | \
162 SLAB_POISON | \
163 SLAB_STORE_USER | \
164 SLAB_TRACE | \
165 SLAB_CONSISTENCY_CHECKS | \
166 SLAB_MEM_SPREAD | \
167 SLAB_NOLEAKTRACE | \
168 SLAB_RECLAIM_ACCOUNT | \
169 SLAB_TEMPORARY | \
170 SLAB_ACCOUNT)
171
172bool __kmem_cache_empty(struct kmem_cache *);
173int __kmem_cache_shutdown(struct kmem_cache *);
174void __kmem_cache_release(struct kmem_cache *);
175int __kmem_cache_shrink(struct kmem_cache *);
176void slab_kmem_cache_release(struct kmem_cache *);
177
178struct seq_file;
179struct file;
180
181struct slabinfo {
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;
187 unsigned int limit;
188 unsigned int batchcount;
189 unsigned int shared;
190 unsigned int objects_per_slab;
191 unsigned int cache_order;
192};
193
194void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo);
195void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s);
196ssize_t slabinfo_write(struct file *file, const char __user *buffer,
197 size_t count, loff_t *ppos);
198
199/*
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.
204 */
205void __kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
206int __kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
207
208static inline int cache_vmstat_idx(struct kmem_cache *s)
209{
210 return (s->flags & SLAB_RECLAIM_ACCOUNT) ?
211 NR_SLAB_RECLAIMABLE_B : NR_SLAB_UNRECLAIMABLE_B;
212}
213
214#ifdef CONFIG_SLUB_DEBUG
215#ifdef CONFIG_SLUB_DEBUG_ON
216DECLARE_STATIC_KEY_TRUE(slub_debug_enabled);
217#else
218DECLARE_STATIC_KEY_FALSE(slub_debug_enabled);
219#endif
220extern void print_tracking(struct kmem_cache *s, void *object);
221#else
222static inline void print_tracking(struct kmem_cache *s, void *object)
223{
224}
225#endif
226
227/*
228 * Returns true if any of the specified slub_debug flags is enabled for the
229 * cache. Use only for flags parsed by setup_slub_debug() as it also enables
230 * the static key.
231 */
232static inline bool kmem_cache_debug_flags(struct kmem_cache *s, slab_flags_t flags)
233{
234#ifdef CONFIG_SLUB_DEBUG
235 VM_WARN_ON_ONCE(!(flags & SLAB_DEBUG_FLAGS));
236 if (static_branch_unlikely(&slub_debug_enabled))
237 return s->flags & flags;
238#endif
239 return false;
240}
241
242#ifdef CONFIG_MEMCG_KMEM
243static inline struct obj_cgroup **page_obj_cgroups(struct page *page)
244{
245 /*
246 * page->mem_cgroup and page->obj_cgroups are sharing the same
247 * space. To distinguish between them in case we don't know for sure
248 * that the page is a slab page (e.g. page_cgroup_ino()), let's
249 * always set the lowest bit of obj_cgroups.
250 */
251 return (struct obj_cgroup **)
252 ((unsigned long)page->obj_cgroups & ~0x1UL);
253}
254
255static inline bool page_has_obj_cgroups(struct page *page)
256{
257 return ((unsigned long)page->obj_cgroups & 0x1UL);
258}
259
260int memcg_alloc_page_obj_cgroups(struct page *page, struct kmem_cache *s,
261 gfp_t gfp);
262
263static inline void memcg_free_page_obj_cgroups(struct page *page)
264{
265 kfree(page_obj_cgroups(page));
266 page->obj_cgroups = NULL;
267}
268
269static inline size_t obj_full_size(struct kmem_cache *s)
270{
271 /*
272 * For each accounted object there is an extra space which is used
273 * to store obj_cgroup membership. Charge it too.
274 */
275 return s->size + sizeof(struct obj_cgroup *);
276}
277
278static inline struct obj_cgroup *memcg_slab_pre_alloc_hook(struct kmem_cache *s,
279 size_t objects,
280 gfp_t flags)
281{
282 struct obj_cgroup *objcg;
283
284 if (memcg_kmem_bypass())
285 return NULL;
286
287 objcg = get_obj_cgroup_from_current();
288 if (!objcg)
289 return NULL;
290
291 if (obj_cgroup_charge(objcg, flags, objects * obj_full_size(s))) {
292 obj_cgroup_put(objcg);
293 return NULL;
294 }
295
296 return objcg;
297}
298
299static inline void mod_objcg_state(struct obj_cgroup *objcg,
300 struct pglist_data *pgdat,
301 int idx, int nr)
302{
303 struct mem_cgroup *memcg;
304 struct lruvec *lruvec;
305
306 rcu_read_lock();
307 memcg = obj_cgroup_memcg(objcg);
308 lruvec = mem_cgroup_lruvec(memcg, pgdat);
309 mod_memcg_lruvec_state(lruvec, idx, nr);
310 rcu_read_unlock();
311}
312
313static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
314 struct obj_cgroup *objcg,
315 gfp_t flags, size_t size,
316 void **p)
317{
318 struct page *page;
319 unsigned long off;
320 size_t i;
321
322 if (!objcg)
323 return;
324
325 flags &= ~__GFP_ACCOUNT;
326 for (i = 0; i < size; i++) {
327 if (likely(p[i])) {
328 page = virt_to_head_page(p[i]);
329
330 if (!page_has_obj_cgroups(page) &&
331 memcg_alloc_page_obj_cgroups(page, s, flags)) {
332 obj_cgroup_uncharge(objcg, obj_full_size(s));
333 continue;
334 }
335
336 off = obj_to_index(s, page, p[i]);
337 obj_cgroup_get(objcg);
338 page_obj_cgroups(page)[off] = objcg;
339 mod_objcg_state(objcg, page_pgdat(page),
340 cache_vmstat_idx(s), obj_full_size(s));
341 } else {
342 obj_cgroup_uncharge(objcg, obj_full_size(s));
343 }
344 }
345 obj_cgroup_put(objcg);
346}
347
348static inline void memcg_slab_free_hook(struct kmem_cache *s, struct page *page,
349 void *p)
350{
351 struct obj_cgroup *objcg;
352 unsigned int off;
353
354 if (!memcg_kmem_enabled())
355 return;
356
357 if (!page_has_obj_cgroups(page))
358 return;
359
360 off = obj_to_index(s, page, p);
361 objcg = page_obj_cgroups(page)[off];
362 page_obj_cgroups(page)[off] = NULL;
363
364 if (!objcg)
365 return;
366
367 obj_cgroup_uncharge(objcg, obj_full_size(s));
368 mod_objcg_state(objcg, page_pgdat(page), cache_vmstat_idx(s),
369 -obj_full_size(s));
370
371 obj_cgroup_put(objcg);
372}
373
374#else /* CONFIG_MEMCG_KMEM */
375static inline bool page_has_obj_cgroups(struct page *page)
376{
377 return false;
378}
379
380static inline struct mem_cgroup *memcg_from_slab_obj(void *ptr)
381{
382 return NULL;
383}
384
385static inline int memcg_alloc_page_obj_cgroups(struct page *page,
386 struct kmem_cache *s, gfp_t gfp)
387{
388 return 0;
389}
390
391static inline void memcg_free_page_obj_cgroups(struct page *page)
392{
393}
394
395static inline struct obj_cgroup *memcg_slab_pre_alloc_hook(struct kmem_cache *s,
396 size_t objects,
397 gfp_t flags)
398{
399 return NULL;
400}
401
402static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
403 struct obj_cgroup *objcg,
404 gfp_t flags, size_t size,
405 void **p)
406{
407}
408
409static inline void memcg_slab_free_hook(struct kmem_cache *s, struct page *page,
410 void *p)
411{
412}
413#endif /* CONFIG_MEMCG_KMEM */
414
415static inline struct kmem_cache *virt_to_cache(const void *obj)
416{
417 struct page *page;
418
419 page = virt_to_head_page(obj);
420 if (WARN_ONCE(!PageSlab(page), "%s: Object is not a Slab page!\n",
421 __func__))
422 return NULL;
423 return page->slab_cache;
424}
425
426static __always_inline void account_slab_page(struct page *page, int order,
427 struct kmem_cache *s)
428{
429 mod_node_page_state(page_pgdat(page), cache_vmstat_idx(s),
430 PAGE_SIZE << order);
431}
432
433static __always_inline void unaccount_slab_page(struct page *page, int order,
434 struct kmem_cache *s)
435{
436 if (memcg_kmem_enabled())
437 memcg_free_page_obj_cgroups(page);
438
439 mod_node_page_state(page_pgdat(page), cache_vmstat_idx(s),
440 -(PAGE_SIZE << order));
441}
442
443static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x)
444{
445 struct kmem_cache *cachep;
446
447 if (!IS_ENABLED(CONFIG_SLAB_FREELIST_HARDENED) &&
448 !kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS))
449 return s;
450
451 cachep = virt_to_cache(x);
452 if (WARN(cachep && cachep != s,
453 "%s: Wrong slab cache. %s but object is from %s\n",
454 __func__, s->name, cachep->name))
455 print_tracking(cachep, x);
456 return cachep;
457}
458
459static inline size_t slab_ksize(const struct kmem_cache *s)
460{
461#ifndef CONFIG_SLUB
462 return s->object_size;
463
464#else /* CONFIG_SLUB */
465# ifdef CONFIG_SLUB_DEBUG
466 /*
467 * Debugging requires use of the padding between object
468 * and whatever may come after it.
469 */
470 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
471 return s->object_size;
472# endif
473 if (s->flags & SLAB_KASAN)
474 return s->object_size;
475 /*
476 * If we have the need to store the freelist pointer
477 * back there or track user information then we can
478 * only use the space before that information.
479 */
480 if (s->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_STORE_USER))
481 return s->inuse;
482 /*
483 * Else we can use all the padding etc for the allocation
484 */
485 return s->size;
486#endif
487}
488
489static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s,
490 struct obj_cgroup **objcgp,
491 size_t size, gfp_t flags)
492{
493 flags &= gfp_allowed_mask;
494
495 fs_reclaim_acquire(flags);
496 fs_reclaim_release(flags);
497
498 might_sleep_if(gfpflags_allow_blocking(flags));
499
500 if (should_failslab(s, flags))
501 return NULL;
502
503 if (memcg_kmem_enabled() &&
504 ((flags & __GFP_ACCOUNT) || (s->flags & SLAB_ACCOUNT)))
505 *objcgp = memcg_slab_pre_alloc_hook(s, size, flags);
506
507 return s;
508}
509
510static inline void slab_post_alloc_hook(struct kmem_cache *s,
511 struct obj_cgroup *objcg,
512 gfp_t flags, size_t size, void **p)
513{
514 size_t i;
515
516 flags &= gfp_allowed_mask;
517 for (i = 0; i < size; i++) {
518 p[i] = kasan_slab_alloc(s, p[i], flags);
519 /* As p[i] might get tagged, call kmemleak hook after KASAN. */
520 kmemleak_alloc_recursive(p[i], s->object_size, 1,
521 s->flags, flags);
522 }
523
524 if (memcg_kmem_enabled())
525 memcg_slab_post_alloc_hook(s, objcg, flags, size, p);
526}
527
528#ifndef CONFIG_SLOB
529/*
530 * The slab lists for all objects.
531 */
532struct kmem_cache_node {
533 spinlock_t list_lock;
534
535#ifdef CONFIG_SLAB
536 struct list_head slabs_partial; /* partial list first, better asm code */
537 struct list_head slabs_full;
538 struct list_head slabs_free;
539 unsigned long total_slabs; /* length of all slab lists */
540 unsigned long free_slabs; /* length of free slab list only */
541 unsigned long free_objects;
542 unsigned int free_limit;
543 unsigned int colour_next; /* Per-node cache coloring */
544 struct array_cache *shared; /* shared per node */
545 struct alien_cache **alien; /* on other nodes */
546 unsigned long next_reap; /* updated without locking */
547 int free_touched; /* updated without locking */
548#endif
549
550#ifdef CONFIG_SLUB
551 unsigned long nr_partial;
552 struct list_head partial;
553#ifdef CONFIG_SLUB_DEBUG
554 atomic_long_t nr_slabs;
555 atomic_long_t total_objects;
556 struct list_head full;
557#endif
558#endif
559
560};
561
562static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
563{
564 return s->node[node];
565}
566
567/*
568 * Iterator over all nodes. The body will be executed for each node that has
569 * a kmem_cache_node structure allocated (which is true for all online nodes)
570 */
571#define for_each_kmem_cache_node(__s, __node, __n) \
572 for (__node = 0; __node < nr_node_ids; __node++) \
573 if ((__n = get_node(__s, __node)))
574
575#endif
576
577void *slab_start(struct seq_file *m, loff_t *pos);
578void *slab_next(struct seq_file *m, void *p, loff_t *pos);
579void slab_stop(struct seq_file *m, void *p);
580int memcg_slab_show(struct seq_file *m, void *p);
581
582#if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
583void dump_unreclaimable_slab(void);
584#else
585static inline void dump_unreclaimable_slab(void)
586{
587}
588#endif
589
590void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr);
591
592#ifdef CONFIG_SLAB_FREELIST_RANDOM
593int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
594 gfp_t gfp);
595void cache_random_seq_destroy(struct kmem_cache *cachep);
596#else
597static inline int cache_random_seq_create(struct kmem_cache *cachep,
598 unsigned int count, gfp_t gfp)
599{
600 return 0;
601}
602static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { }
603#endif /* CONFIG_SLAB_FREELIST_RANDOM */
604
605static inline bool slab_want_init_on_alloc(gfp_t flags, struct kmem_cache *c)
606{
607 if (static_branch_unlikely(&init_on_alloc)) {
608 if (c->ctor)
609 return false;
610 if (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON))
611 return flags & __GFP_ZERO;
612 return true;
613 }
614 return flags & __GFP_ZERO;
615}
616
617static inline bool slab_want_init_on_free(struct kmem_cache *c)
618{
619 if (static_branch_unlikely(&init_on_free))
620 return !(c->ctor ||
621 (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)));
622 return false;
623}
624
625#endif /* MM_SLAB_H */