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#else /* !CONFIG_SLOB */
34
35struct memcg_cache_array {
36 struct rcu_head rcu;
37 struct kmem_cache *entries[0];
38};
39
40/*
41 * This is the main placeholder for memcg-related information in kmem caches.
42 * Both the root cache and the child caches will have it. For the root cache,
43 * this will hold a dynamically allocated array large enough to hold
44 * information about the currently limited memcgs in the system. To allow the
45 * array to be accessed without taking any locks, on relocation we free the old
46 * version only after a grace period.
47 *
48 * Root and child caches hold different metadata.
49 *
50 * @root_cache: Common to root and child caches. NULL for root, pointer to
51 * the root cache for children.
52 *
53 * The following fields are specific to root caches.
54 *
55 * @memcg_caches: kmemcg ID indexed table of child caches. This table is
56 * used to index child cachces during allocation and cleared
57 * early during shutdown.
58 *
59 * @root_caches_node: List node for slab_root_caches list.
60 *
61 * @children: List of all child caches. While the child caches are also
62 * reachable through @memcg_caches, a child cache remains on
63 * this list until it is actually destroyed.
64 *
65 * The following fields are specific to child caches.
66 *
67 * @memcg: Pointer to the memcg this cache belongs to.
68 *
69 * @children_node: List node for @root_cache->children list.
70 *
71 * @kmem_caches_node: List node for @memcg->kmem_caches list.
72 */
73struct memcg_cache_params {
74 struct kmem_cache *root_cache;
75 union {
76 struct {
77 struct memcg_cache_array __rcu *memcg_caches;
78 struct list_head __root_caches_node;
79 struct list_head children;
80 bool dying;
81 };
82 struct {
83 struct mem_cgroup *memcg;
84 struct list_head children_node;
85 struct list_head kmem_caches_node;
86 struct percpu_ref refcnt;
87
88 void (*work_fn)(struct kmem_cache *);
89 union {
90 struct rcu_head rcu_head;
91 struct work_struct work;
92 };
93 };
94 };
95};
96#endif /* CONFIG_SLOB */
97
98#ifdef CONFIG_SLAB
99#include <linux/slab_def.h>
100#endif
101
102#ifdef CONFIG_SLUB
103#include <linux/slub_def.h>
104#endif
105
106#include <linux/memcontrol.h>
107#include <linux/fault-inject.h>
108#include <linux/kasan.h>
109#include <linux/kmemleak.h>
110#include <linux/random.h>
111#include <linux/sched/mm.h>
112
113/*
114 * State of the slab allocator.
115 *
116 * This is used to describe the states of the allocator during bootup.
117 * Allocators use this to gradually bootstrap themselves. Most allocators
118 * have the problem that the structures used for managing slab caches are
119 * allocated from slab caches themselves.
120 */
121enum slab_state {
122 DOWN, /* No slab functionality yet */
123 PARTIAL, /* SLUB: kmem_cache_node available */
124 PARTIAL_NODE, /* SLAB: kmalloc size for node struct available */
125 UP, /* Slab caches usable but not all extras yet */
126 FULL /* Everything is working */
127};
128
129extern enum slab_state slab_state;
130
131/* The slab cache mutex protects the management structures during changes */
132extern struct mutex slab_mutex;
133
134/* The list of all slab caches on the system */
135extern struct list_head slab_caches;
136
137/* The slab cache that manages slab cache information */
138extern struct kmem_cache *kmem_cache;
139
140/* A table of kmalloc cache names and sizes */
141extern const struct kmalloc_info_struct {
142 const char *name;
143 unsigned int size;
144} kmalloc_info[];
145
146#ifndef CONFIG_SLOB
147/* Kmalloc array related functions */
148void setup_kmalloc_cache_index_table(void);
149void create_kmalloc_caches(slab_flags_t);
150
151/* Find the kmalloc slab corresponding for a certain size */
152struct kmem_cache *kmalloc_slab(size_t, gfp_t);
153#endif
154
155
156/* Functions provided by the slab allocators */
157int __kmem_cache_create(struct kmem_cache *, slab_flags_t flags);
158
159struct kmem_cache *create_kmalloc_cache(const char *name, unsigned int size,
160 slab_flags_t flags, unsigned int useroffset,
161 unsigned int usersize);
162extern void create_boot_cache(struct kmem_cache *, const char *name,
163 unsigned int size, slab_flags_t flags,
164 unsigned int useroffset, unsigned int usersize);
165
166int slab_unmergeable(struct kmem_cache *s);
167struct kmem_cache *find_mergeable(unsigned size, unsigned align,
168 slab_flags_t flags, const char *name, void (*ctor)(void *));
169#ifndef CONFIG_SLOB
170struct kmem_cache *
171__kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
172 slab_flags_t flags, void (*ctor)(void *));
173
174slab_flags_t kmem_cache_flags(unsigned int object_size,
175 slab_flags_t flags, const char *name,
176 void (*ctor)(void *));
177#else
178static inline struct kmem_cache *
179__kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
180 slab_flags_t flags, void (*ctor)(void *))
181{ return NULL; }
182
183static inline slab_flags_t kmem_cache_flags(unsigned int object_size,
184 slab_flags_t flags, const char *name,
185 void (*ctor)(void *))
186{
187 return flags;
188}
189#endif
190
191
192/* Legal flag mask for kmem_cache_create(), for various configurations */
193#define SLAB_CORE_FLAGS (SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA | \
194 SLAB_CACHE_DMA32 | SLAB_PANIC | \
195 SLAB_TYPESAFE_BY_RCU | SLAB_DEBUG_OBJECTS )
196
197#if defined(CONFIG_DEBUG_SLAB)
198#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
199#elif defined(CONFIG_SLUB_DEBUG)
200#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
201 SLAB_TRACE | SLAB_CONSISTENCY_CHECKS)
202#else
203#define SLAB_DEBUG_FLAGS (0)
204#endif
205
206#if defined(CONFIG_SLAB)
207#define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \
208 SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \
209 SLAB_ACCOUNT)
210#elif defined(CONFIG_SLUB)
211#define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
212 SLAB_TEMPORARY | SLAB_ACCOUNT)
213#else
214#define SLAB_CACHE_FLAGS (0)
215#endif
216
217/* Common flags available with current configuration */
218#define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
219
220/* Common flags permitted for kmem_cache_create */
221#define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \
222 SLAB_RED_ZONE | \
223 SLAB_POISON | \
224 SLAB_STORE_USER | \
225 SLAB_TRACE | \
226 SLAB_CONSISTENCY_CHECKS | \
227 SLAB_MEM_SPREAD | \
228 SLAB_NOLEAKTRACE | \
229 SLAB_RECLAIM_ACCOUNT | \
230 SLAB_TEMPORARY | \
231 SLAB_ACCOUNT)
232
233bool __kmem_cache_empty(struct kmem_cache *);
234int __kmem_cache_shutdown(struct kmem_cache *);
235void __kmem_cache_release(struct kmem_cache *);
236int __kmem_cache_shrink(struct kmem_cache *);
237void __kmemcg_cache_deactivate(struct kmem_cache *s);
238void __kmemcg_cache_deactivate_after_rcu(struct kmem_cache *s);
239void slab_kmem_cache_release(struct kmem_cache *);
240void kmem_cache_shrink_all(struct kmem_cache *s);
241
242struct seq_file;
243struct file;
244
245struct slabinfo {
246 unsigned long active_objs;
247 unsigned long num_objs;
248 unsigned long active_slabs;
249 unsigned long num_slabs;
250 unsigned long shared_avail;
251 unsigned int limit;
252 unsigned int batchcount;
253 unsigned int shared;
254 unsigned int objects_per_slab;
255 unsigned int cache_order;
256};
257
258void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo);
259void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s);
260ssize_t slabinfo_write(struct file *file, const char __user *buffer,
261 size_t count, loff_t *ppos);
262
263/*
264 * Generic implementation of bulk operations
265 * These are useful for situations in which the allocator cannot
266 * perform optimizations. In that case segments of the object listed
267 * may be allocated or freed using these operations.
268 */
269void __kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
270int __kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
271
272static inline int cache_vmstat_idx(struct kmem_cache *s)
273{
274 return (s->flags & SLAB_RECLAIM_ACCOUNT) ?
275 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE;
276}
277
278#ifdef CONFIG_MEMCG_KMEM
279
280/* List of all root caches. */
281extern struct list_head slab_root_caches;
282#define root_caches_node memcg_params.__root_caches_node
283
284/*
285 * Iterate over all memcg caches of the given root cache. The caller must hold
286 * slab_mutex.
287 */
288#define for_each_memcg_cache(iter, root) \
289 list_for_each_entry(iter, &(root)->memcg_params.children, \
290 memcg_params.children_node)
291
292static inline bool is_root_cache(struct kmem_cache *s)
293{
294 return !s->memcg_params.root_cache;
295}
296
297static inline bool slab_equal_or_root(struct kmem_cache *s,
298 struct kmem_cache *p)
299{
300 return p == s || p == s->memcg_params.root_cache;
301}
302
303/*
304 * We use suffixes to the name in memcg because we can't have caches
305 * created in the system with the same name. But when we print them
306 * locally, better refer to them with the base name
307 */
308static inline const char *cache_name(struct kmem_cache *s)
309{
310 if (!is_root_cache(s))
311 s = s->memcg_params.root_cache;
312 return s->name;
313}
314
315static inline struct kmem_cache *memcg_root_cache(struct kmem_cache *s)
316{
317 if (is_root_cache(s))
318 return s;
319 return s->memcg_params.root_cache;
320}
321
322/*
323 * Expects a pointer to a slab page. Please note, that PageSlab() check
324 * isn't sufficient, as it returns true also for tail compound slab pages,
325 * which do not have slab_cache pointer set.
326 * So this function assumes that the page can pass PageSlab() && !PageTail()
327 * check.
328 *
329 * The kmem_cache can be reparented asynchronously. The caller must ensure
330 * the memcg lifetime, e.g. by taking rcu_read_lock() or cgroup_mutex.
331 */
332static inline struct mem_cgroup *memcg_from_slab_page(struct page *page)
333{
334 struct kmem_cache *s;
335
336 s = READ_ONCE(page->slab_cache);
337 if (s && !is_root_cache(s))
338 return READ_ONCE(s->memcg_params.memcg);
339
340 return NULL;
341}
342
343/*
344 * Charge the slab page belonging to the non-root kmem_cache.
345 * Can be called for non-root kmem_caches only.
346 */
347static __always_inline int memcg_charge_slab(struct page *page,
348 gfp_t gfp, int order,
349 struct kmem_cache *s)
350{
351 struct mem_cgroup *memcg;
352 struct lruvec *lruvec;
353 int ret;
354
355 rcu_read_lock();
356 memcg = READ_ONCE(s->memcg_params.memcg);
357 while (memcg && !css_tryget_online(&memcg->css))
358 memcg = parent_mem_cgroup(memcg);
359 rcu_read_unlock();
360
361 if (unlikely(!memcg || mem_cgroup_is_root(memcg))) {
362 mod_node_page_state(page_pgdat(page), cache_vmstat_idx(s),
363 (1 << order));
364 percpu_ref_get_many(&s->memcg_params.refcnt, 1 << order);
365 return 0;
366 }
367
368 ret = memcg_kmem_charge_memcg(page, gfp, order, memcg);
369 if (ret)
370 goto out;
371
372 lruvec = mem_cgroup_lruvec(page_pgdat(page), memcg);
373 mod_lruvec_state(lruvec, cache_vmstat_idx(s), 1 << order);
374
375 /* transer try_charge() page references to kmem_cache */
376 percpu_ref_get_many(&s->memcg_params.refcnt, 1 << order);
377 css_put_many(&memcg->css, 1 << order);
378out:
379 css_put(&memcg->css);
380 return ret;
381}
382
383/*
384 * Uncharge a slab page belonging to a non-root kmem_cache.
385 * Can be called for non-root kmem_caches only.
386 */
387static __always_inline void memcg_uncharge_slab(struct page *page, int order,
388 struct kmem_cache *s)
389{
390 struct mem_cgroup *memcg;
391 struct lruvec *lruvec;
392
393 rcu_read_lock();
394 memcg = READ_ONCE(s->memcg_params.memcg);
395 if (likely(!mem_cgroup_is_root(memcg))) {
396 lruvec = mem_cgroup_lruvec(page_pgdat(page), memcg);
397 mod_lruvec_state(lruvec, cache_vmstat_idx(s), -(1 << order));
398 memcg_kmem_uncharge_memcg(page, order, memcg);
399 } else {
400 mod_node_page_state(page_pgdat(page), cache_vmstat_idx(s),
401 -(1 << order));
402 }
403 rcu_read_unlock();
404
405 percpu_ref_put_many(&s->memcg_params.refcnt, 1 << order);
406}
407
408extern void slab_init_memcg_params(struct kmem_cache *);
409extern void memcg_link_cache(struct kmem_cache *s, struct mem_cgroup *memcg);
410
411#else /* CONFIG_MEMCG_KMEM */
412
413/* If !memcg, all caches are root. */
414#define slab_root_caches slab_caches
415#define root_caches_node list
416
417#define for_each_memcg_cache(iter, root) \
418 for ((void)(iter), (void)(root); 0; )
419
420static inline bool is_root_cache(struct kmem_cache *s)
421{
422 return true;
423}
424
425static inline bool slab_equal_or_root(struct kmem_cache *s,
426 struct kmem_cache *p)
427{
428 return s == p;
429}
430
431static inline const char *cache_name(struct kmem_cache *s)
432{
433 return s->name;
434}
435
436static inline struct kmem_cache *memcg_root_cache(struct kmem_cache *s)
437{
438 return s;
439}
440
441static inline struct mem_cgroup *memcg_from_slab_page(struct page *page)
442{
443 return NULL;
444}
445
446static inline int memcg_charge_slab(struct page *page, gfp_t gfp, int order,
447 struct kmem_cache *s)
448{
449 return 0;
450}
451
452static inline void memcg_uncharge_slab(struct page *page, int order,
453 struct kmem_cache *s)
454{
455}
456
457static inline void slab_init_memcg_params(struct kmem_cache *s)
458{
459}
460
461static inline void memcg_link_cache(struct kmem_cache *s,
462 struct mem_cgroup *memcg)
463{
464}
465
466#endif /* CONFIG_MEMCG_KMEM */
467
468static inline struct kmem_cache *virt_to_cache(const void *obj)
469{
470 struct page *page;
471
472 page = virt_to_head_page(obj);
473 if (WARN_ONCE(!PageSlab(page), "%s: Object is not a Slab page!\n",
474 __func__))
475 return NULL;
476 return page->slab_cache;
477}
478
479static __always_inline int charge_slab_page(struct page *page,
480 gfp_t gfp, int order,
481 struct kmem_cache *s)
482{
483 if (is_root_cache(s)) {
484 mod_node_page_state(page_pgdat(page), cache_vmstat_idx(s),
485 1 << order);
486 return 0;
487 }
488
489 return memcg_charge_slab(page, gfp, order, s);
490}
491
492static __always_inline void uncharge_slab_page(struct page *page, int order,
493 struct kmem_cache *s)
494{
495 if (is_root_cache(s)) {
496 mod_node_page_state(page_pgdat(page), cache_vmstat_idx(s),
497 -(1 << order));
498 return;
499 }
500
501 memcg_uncharge_slab(page, order, s);
502}
503
504static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x)
505{
506 struct kmem_cache *cachep;
507
508 /*
509 * When kmemcg is not being used, both assignments should return the
510 * same value. but we don't want to pay the assignment price in that
511 * case. If it is not compiled in, the compiler should be smart enough
512 * to not do even the assignment. In that case, slab_equal_or_root
513 * will also be a constant.
514 */
515 if (!memcg_kmem_enabled() &&
516 !IS_ENABLED(CONFIG_SLAB_FREELIST_HARDENED) &&
517 !unlikely(s->flags & SLAB_CONSISTENCY_CHECKS))
518 return s;
519
520 cachep = virt_to_cache(x);
521 WARN_ONCE(cachep && !slab_equal_or_root(cachep, s),
522 "%s: Wrong slab cache. %s but object is from %s\n",
523 __func__, s->name, cachep->name);
524 return cachep;
525}
526
527static inline size_t slab_ksize(const struct kmem_cache *s)
528{
529#ifndef CONFIG_SLUB
530 return s->object_size;
531
532#else /* CONFIG_SLUB */
533# ifdef CONFIG_SLUB_DEBUG
534 /*
535 * Debugging requires use of the padding between object
536 * and whatever may come after it.
537 */
538 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
539 return s->object_size;
540# endif
541 if (s->flags & SLAB_KASAN)
542 return s->object_size;
543 /*
544 * If we have the need to store the freelist pointer
545 * back there or track user information then we can
546 * only use the space before that information.
547 */
548 if (s->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_STORE_USER))
549 return s->inuse;
550 /*
551 * Else we can use all the padding etc for the allocation
552 */
553 return s->size;
554#endif
555}
556
557static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s,
558 gfp_t flags)
559{
560 flags &= gfp_allowed_mask;
561
562 fs_reclaim_acquire(flags);
563 fs_reclaim_release(flags);
564
565 might_sleep_if(gfpflags_allow_blocking(flags));
566
567 if (should_failslab(s, flags))
568 return NULL;
569
570 if (memcg_kmem_enabled() &&
571 ((flags & __GFP_ACCOUNT) || (s->flags & SLAB_ACCOUNT)))
572 return memcg_kmem_get_cache(s);
573
574 return s;
575}
576
577static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags,
578 size_t size, void **p)
579{
580 size_t i;
581
582 flags &= gfp_allowed_mask;
583 for (i = 0; i < size; i++) {
584 p[i] = kasan_slab_alloc(s, p[i], flags);
585 /* As p[i] might get tagged, call kmemleak hook after KASAN. */
586 kmemleak_alloc_recursive(p[i], s->object_size, 1,
587 s->flags, flags);
588 }
589
590 if (memcg_kmem_enabled())
591 memcg_kmem_put_cache(s);
592}
593
594#ifndef CONFIG_SLOB
595/*
596 * The slab lists for all objects.
597 */
598struct kmem_cache_node {
599 spinlock_t list_lock;
600
601#ifdef CONFIG_SLAB
602 struct list_head slabs_partial; /* partial list first, better asm code */
603 struct list_head slabs_full;
604 struct list_head slabs_free;
605 unsigned long total_slabs; /* length of all slab lists */
606 unsigned long free_slabs; /* length of free slab list only */
607 unsigned long free_objects;
608 unsigned int free_limit;
609 unsigned int colour_next; /* Per-node cache coloring */
610 struct array_cache *shared; /* shared per node */
611 struct alien_cache **alien; /* on other nodes */
612 unsigned long next_reap; /* updated without locking */
613 int free_touched; /* updated without locking */
614#endif
615
616#ifdef CONFIG_SLUB
617 unsigned long nr_partial;
618 struct list_head partial;
619#ifdef CONFIG_SLUB_DEBUG
620 atomic_long_t nr_slabs;
621 atomic_long_t total_objects;
622 struct list_head full;
623#endif
624#endif
625
626};
627
628static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
629{
630 return s->node[node];
631}
632
633/*
634 * Iterator over all nodes. The body will be executed for each node that has
635 * a kmem_cache_node structure allocated (which is true for all online nodes)
636 */
637#define for_each_kmem_cache_node(__s, __node, __n) \
638 for (__node = 0; __node < nr_node_ids; __node++) \
639 if ((__n = get_node(__s, __node)))
640
641#endif
642
643void *slab_start(struct seq_file *m, loff_t *pos);
644void *slab_next(struct seq_file *m, void *p, loff_t *pos);
645void slab_stop(struct seq_file *m, void *p);
646void *memcg_slab_start(struct seq_file *m, loff_t *pos);
647void *memcg_slab_next(struct seq_file *m, void *p, loff_t *pos);
648void memcg_slab_stop(struct seq_file *m, void *p);
649int memcg_slab_show(struct seq_file *m, void *p);
650
651#if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
652void dump_unreclaimable_slab(void);
653#else
654static inline void dump_unreclaimable_slab(void)
655{
656}
657#endif
658
659void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr);
660
661#ifdef CONFIG_SLAB_FREELIST_RANDOM
662int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
663 gfp_t gfp);
664void cache_random_seq_destroy(struct kmem_cache *cachep);
665#else
666static inline int cache_random_seq_create(struct kmem_cache *cachep,
667 unsigned int count, gfp_t gfp)
668{
669 return 0;
670}
671static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { }
672#endif /* CONFIG_SLAB_FREELIST_RANDOM */
673
674static inline bool slab_want_init_on_alloc(gfp_t flags, struct kmem_cache *c)
675{
676 if (static_branch_unlikely(&init_on_alloc)) {
677 if (c->ctor)
678 return false;
679 if (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON))
680 return flags & __GFP_ZERO;
681 return true;
682 }
683 return flags & __GFP_ZERO;
684}
685
686static inline bool slab_want_init_on_free(struct kmem_cache *c)
687{
688 if (static_branch_unlikely(&init_on_free))
689 return !(c->ctor ||
690 (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)));
691 return false;
692}
693
694#endif /* MM_SLAB_H */