Loading...
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/* Reuses the bits in struct page */
9struct slab {
10 unsigned long __page_flags;
11
12#if defined(CONFIG_SLAB)
13
14 struct kmem_cache *slab_cache;
15 union {
16 struct {
17 struct list_head slab_list;
18 void *freelist; /* array of free object indexes */
19 void *s_mem; /* first object */
20 };
21 struct rcu_head rcu_head;
22 };
23 unsigned int active;
24
25#elif defined(CONFIG_SLUB)
26
27 struct kmem_cache *slab_cache;
28 union {
29 struct {
30 union {
31 struct list_head slab_list;
32#ifdef CONFIG_SLUB_CPU_PARTIAL
33 struct {
34 struct slab *next;
35 int slabs; /* Nr of slabs left */
36 };
37#endif
38 };
39 /* Double-word boundary */
40 void *freelist; /* first free object */
41 union {
42 unsigned long counters;
43 struct {
44 unsigned inuse:16;
45 unsigned objects:15;
46 unsigned frozen:1;
47 };
48 };
49 };
50 struct rcu_head rcu_head;
51 };
52 unsigned int __unused;
53
54#elif defined(CONFIG_SLOB)
55
56 struct list_head slab_list;
57 void *__unused_1;
58 void *freelist; /* first free block */
59 long units;
60 unsigned int __unused_2;
61
62#else
63#error "Unexpected slab allocator configured"
64#endif
65
66 atomic_t __page_refcount;
67#ifdef CONFIG_MEMCG
68 unsigned long memcg_data;
69#endif
70};
71
72#define SLAB_MATCH(pg, sl) \
73 static_assert(offsetof(struct page, pg) == offsetof(struct slab, sl))
74SLAB_MATCH(flags, __page_flags);
75#ifndef CONFIG_SLOB
76SLAB_MATCH(compound_head, slab_cache); /* Ensure bit 0 is clear */
77#else
78SLAB_MATCH(compound_head, slab_list); /* Ensure bit 0 is clear */
79#endif
80SLAB_MATCH(_refcount, __page_refcount);
81#ifdef CONFIG_MEMCG
82SLAB_MATCH(memcg_data, memcg_data);
83#endif
84#undef SLAB_MATCH
85static_assert(sizeof(struct slab) <= sizeof(struct page));
86#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && defined(CONFIG_SLUB)
87static_assert(IS_ALIGNED(offsetof(struct slab, freelist), 2*sizeof(void *)));
88#endif
89
90/**
91 * folio_slab - Converts from folio to slab.
92 * @folio: The folio.
93 *
94 * Currently struct slab is a different representation of a folio where
95 * folio_test_slab() is true.
96 *
97 * Return: The slab which contains this folio.
98 */
99#define folio_slab(folio) (_Generic((folio), \
100 const struct folio *: (const struct slab *)(folio), \
101 struct folio *: (struct slab *)(folio)))
102
103/**
104 * slab_folio - The folio allocated for a slab
105 * @slab: The slab.
106 *
107 * Slabs are allocated as folios that contain the individual objects and are
108 * using some fields in the first struct page of the folio - those fields are
109 * now accessed by struct slab. It is occasionally necessary to convert back to
110 * a folio in order to communicate with the rest of the mm. Please use this
111 * helper function instead of casting yourself, as the implementation may change
112 * in the future.
113 */
114#define slab_folio(s) (_Generic((s), \
115 const struct slab *: (const struct folio *)s, \
116 struct slab *: (struct folio *)s))
117
118/**
119 * page_slab - Converts from first struct page to slab.
120 * @p: The first (either head of compound or single) page of slab.
121 *
122 * A temporary wrapper to convert struct page to struct slab in situations where
123 * we know the page is the compound head, or single order-0 page.
124 *
125 * Long-term ideally everything would work with struct slab directly or go
126 * through folio to struct slab.
127 *
128 * Return: The slab which contains this page
129 */
130#define page_slab(p) (_Generic((p), \
131 const struct page *: (const struct slab *)(p), \
132 struct page *: (struct slab *)(p)))
133
134/**
135 * slab_page - The first struct page allocated for a slab
136 * @slab: The slab.
137 *
138 * A convenience wrapper for converting slab to the first struct page of the
139 * underlying folio, to communicate with code not yet converted to folio or
140 * struct slab.
141 */
142#define slab_page(s) folio_page(slab_folio(s), 0)
143
144/*
145 * If network-based swap is enabled, sl*b must keep track of whether pages
146 * were allocated from pfmemalloc reserves.
147 */
148static inline bool slab_test_pfmemalloc(const struct slab *slab)
149{
150 return folio_test_active((struct folio *)slab_folio(slab));
151}
152
153static inline void slab_set_pfmemalloc(struct slab *slab)
154{
155 folio_set_active(slab_folio(slab));
156}
157
158static inline void slab_clear_pfmemalloc(struct slab *slab)
159{
160 folio_clear_active(slab_folio(slab));
161}
162
163static inline void __slab_clear_pfmemalloc(struct slab *slab)
164{
165 __folio_clear_active(slab_folio(slab));
166}
167
168static inline void *slab_address(const struct slab *slab)
169{
170 return folio_address(slab_folio(slab));
171}
172
173static inline int slab_nid(const struct slab *slab)
174{
175 return folio_nid(slab_folio(slab));
176}
177
178static inline pg_data_t *slab_pgdat(const struct slab *slab)
179{
180 return folio_pgdat(slab_folio(slab));
181}
182
183static inline struct slab *virt_to_slab(const void *addr)
184{
185 struct folio *folio = virt_to_folio(addr);
186
187 if (!folio_test_slab(folio))
188 return NULL;
189
190 return folio_slab(folio);
191}
192
193static inline int slab_order(const struct slab *slab)
194{
195 return folio_order((struct folio *)slab_folio(slab));
196}
197
198static inline size_t slab_size(const struct slab *slab)
199{
200 return PAGE_SIZE << slab_order(slab);
201}
202
203#ifdef CONFIG_SLOB
204/*
205 * Common fields provided in kmem_cache by all slab allocators
206 * This struct is either used directly by the allocator (SLOB)
207 * or the allocator must include definitions for all fields
208 * provided in kmem_cache_common in their definition of kmem_cache.
209 *
210 * Once we can do anonymous structs (C11 standard) we could put a
211 * anonymous struct definition in these allocators so that the
212 * separate allocations in the kmem_cache structure of SLAB and
213 * SLUB is no longer needed.
214 */
215struct kmem_cache {
216 unsigned int object_size;/* The original size of the object */
217 unsigned int size; /* The aligned/padded/added on size */
218 unsigned int align; /* Alignment as calculated */
219 slab_flags_t flags; /* Active flags on the slab */
220 const char *name; /* Slab name for sysfs */
221 int refcount; /* Use counter */
222 void (*ctor)(void *); /* Called on object slot creation */
223 struct list_head list; /* List of all slab caches on the system */
224};
225
226#endif /* CONFIG_SLOB */
227
228#ifdef CONFIG_SLAB
229#include <linux/slab_def.h>
230#endif
231
232#ifdef CONFIG_SLUB
233#include <linux/slub_def.h>
234#endif
235
236#include <linux/memcontrol.h>
237#include <linux/fault-inject.h>
238#include <linux/kasan.h>
239#include <linux/kmemleak.h>
240#include <linux/random.h>
241#include <linux/sched/mm.h>
242#include <linux/list_lru.h>
243
244/*
245 * State of the slab allocator.
246 *
247 * This is used to describe the states of the allocator during bootup.
248 * Allocators use this to gradually bootstrap themselves. Most allocators
249 * have the problem that the structures used for managing slab caches are
250 * allocated from slab caches themselves.
251 */
252enum slab_state {
253 DOWN, /* No slab functionality yet */
254 PARTIAL, /* SLUB: kmem_cache_node available */
255 PARTIAL_NODE, /* SLAB: kmalloc size for node struct available */
256 UP, /* Slab caches usable but not all extras yet */
257 FULL /* Everything is working */
258};
259
260extern enum slab_state slab_state;
261
262/* The slab cache mutex protects the management structures during changes */
263extern struct mutex slab_mutex;
264
265/* The list of all slab caches on the system */
266extern struct list_head slab_caches;
267
268/* The slab cache that manages slab cache information */
269extern struct kmem_cache *kmem_cache;
270
271/* A table of kmalloc cache names and sizes */
272extern const struct kmalloc_info_struct {
273 const char *name[NR_KMALLOC_TYPES];
274 unsigned int size;
275} kmalloc_info[];
276
277#ifndef CONFIG_SLOB
278/* Kmalloc array related functions */
279void setup_kmalloc_cache_index_table(void);
280void create_kmalloc_caches(slab_flags_t);
281
282/* Find the kmalloc slab corresponding for a certain size */
283struct kmem_cache *kmalloc_slab(size_t, gfp_t);
284
285void *__kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags,
286 int node, size_t orig_size,
287 unsigned long caller);
288void __kmem_cache_free(struct kmem_cache *s, void *x, unsigned long caller);
289#endif
290
291gfp_t kmalloc_fix_flags(gfp_t flags);
292
293/* Functions provided by the slab allocators */
294int __kmem_cache_create(struct kmem_cache *, slab_flags_t flags);
295
296struct kmem_cache *create_kmalloc_cache(const char *name, unsigned int size,
297 slab_flags_t flags, unsigned int useroffset,
298 unsigned int usersize);
299extern void create_boot_cache(struct kmem_cache *, const char *name,
300 unsigned int size, slab_flags_t flags,
301 unsigned int useroffset, unsigned int usersize);
302
303int slab_unmergeable(struct kmem_cache *s);
304struct kmem_cache *find_mergeable(unsigned size, unsigned align,
305 slab_flags_t flags, const char *name, void (*ctor)(void *));
306#ifndef CONFIG_SLOB
307struct kmem_cache *
308__kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
309 slab_flags_t flags, void (*ctor)(void *));
310
311slab_flags_t kmem_cache_flags(unsigned int object_size,
312 slab_flags_t flags, const char *name);
313#else
314static inline struct kmem_cache *
315__kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
316 slab_flags_t flags, void (*ctor)(void *))
317{ return NULL; }
318
319static inline slab_flags_t kmem_cache_flags(unsigned int object_size,
320 slab_flags_t flags, const char *name)
321{
322 return flags;
323}
324#endif
325
326
327/* Legal flag mask for kmem_cache_create(), for various configurations */
328#define SLAB_CORE_FLAGS (SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA | \
329 SLAB_CACHE_DMA32 | SLAB_PANIC | \
330 SLAB_TYPESAFE_BY_RCU | SLAB_DEBUG_OBJECTS )
331
332#if defined(CONFIG_DEBUG_SLAB)
333#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
334#elif defined(CONFIG_SLUB_DEBUG)
335#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
336 SLAB_TRACE | SLAB_CONSISTENCY_CHECKS)
337#else
338#define SLAB_DEBUG_FLAGS (0)
339#endif
340
341#if defined(CONFIG_SLAB)
342#define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \
343 SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \
344 SLAB_ACCOUNT)
345#elif defined(CONFIG_SLUB)
346#define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
347 SLAB_TEMPORARY | SLAB_ACCOUNT | \
348 SLAB_NO_USER_FLAGS | SLAB_KMALLOC)
349#else
350#define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE)
351#endif
352
353/* Common flags available with current configuration */
354#define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
355
356/* Common flags permitted for kmem_cache_create */
357#define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \
358 SLAB_RED_ZONE | \
359 SLAB_POISON | \
360 SLAB_STORE_USER | \
361 SLAB_TRACE | \
362 SLAB_CONSISTENCY_CHECKS | \
363 SLAB_MEM_SPREAD | \
364 SLAB_NOLEAKTRACE | \
365 SLAB_RECLAIM_ACCOUNT | \
366 SLAB_TEMPORARY | \
367 SLAB_ACCOUNT | \
368 SLAB_KMALLOC | \
369 SLAB_NO_USER_FLAGS)
370
371bool __kmem_cache_empty(struct kmem_cache *);
372int __kmem_cache_shutdown(struct kmem_cache *);
373void __kmem_cache_release(struct kmem_cache *);
374int __kmem_cache_shrink(struct kmem_cache *);
375void slab_kmem_cache_release(struct kmem_cache *);
376
377struct seq_file;
378struct file;
379
380struct slabinfo {
381 unsigned long active_objs;
382 unsigned long num_objs;
383 unsigned long active_slabs;
384 unsigned long num_slabs;
385 unsigned long shared_avail;
386 unsigned int limit;
387 unsigned int batchcount;
388 unsigned int shared;
389 unsigned int objects_per_slab;
390 unsigned int cache_order;
391};
392
393void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo);
394void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s);
395ssize_t slabinfo_write(struct file *file, const char __user *buffer,
396 size_t count, loff_t *ppos);
397
398static inline enum node_stat_item cache_vmstat_idx(struct kmem_cache *s)
399{
400 return (s->flags & SLAB_RECLAIM_ACCOUNT) ?
401 NR_SLAB_RECLAIMABLE_B : NR_SLAB_UNRECLAIMABLE_B;
402}
403
404#ifdef CONFIG_SLUB_DEBUG
405#ifdef CONFIG_SLUB_DEBUG_ON
406DECLARE_STATIC_KEY_TRUE(slub_debug_enabled);
407#else
408DECLARE_STATIC_KEY_FALSE(slub_debug_enabled);
409#endif
410extern void print_tracking(struct kmem_cache *s, void *object);
411long validate_slab_cache(struct kmem_cache *s);
412static inline bool __slub_debug_enabled(void)
413{
414 return static_branch_unlikely(&slub_debug_enabled);
415}
416#else
417static inline void print_tracking(struct kmem_cache *s, void *object)
418{
419}
420static inline bool __slub_debug_enabled(void)
421{
422 return false;
423}
424#endif
425
426/*
427 * Returns true if any of the specified slub_debug flags is enabled for the
428 * cache. Use only for flags parsed by setup_slub_debug() as it also enables
429 * the static key.
430 */
431static inline bool kmem_cache_debug_flags(struct kmem_cache *s, slab_flags_t flags)
432{
433 if (IS_ENABLED(CONFIG_SLUB_DEBUG))
434 VM_WARN_ON_ONCE(!(flags & SLAB_DEBUG_FLAGS));
435 if (__slub_debug_enabled())
436 return s->flags & flags;
437 return false;
438}
439
440#ifdef CONFIG_MEMCG_KMEM
441/*
442 * slab_objcgs - get the object cgroups vector associated with a slab
443 * @slab: a pointer to the slab struct
444 *
445 * Returns a pointer to the object cgroups vector associated with the slab,
446 * or NULL if no such vector has been associated yet.
447 */
448static inline struct obj_cgroup **slab_objcgs(struct slab *slab)
449{
450 unsigned long memcg_data = READ_ONCE(slab->memcg_data);
451
452 VM_BUG_ON_PAGE(memcg_data && !(memcg_data & MEMCG_DATA_OBJCGS),
453 slab_page(slab));
454 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_KMEM, slab_page(slab));
455
456 return (struct obj_cgroup **)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
457}
458
459int memcg_alloc_slab_cgroups(struct slab *slab, struct kmem_cache *s,
460 gfp_t gfp, bool new_slab);
461void mod_objcg_state(struct obj_cgroup *objcg, struct pglist_data *pgdat,
462 enum node_stat_item idx, int nr);
463
464static inline void memcg_free_slab_cgroups(struct slab *slab)
465{
466 kfree(slab_objcgs(slab));
467 slab->memcg_data = 0;
468}
469
470static inline size_t obj_full_size(struct kmem_cache *s)
471{
472 /*
473 * For each accounted object there is an extra space which is used
474 * to store obj_cgroup membership. Charge it too.
475 */
476 return s->size + sizeof(struct obj_cgroup *);
477}
478
479/*
480 * Returns false if the allocation should fail.
481 */
482static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s,
483 struct list_lru *lru,
484 struct obj_cgroup **objcgp,
485 size_t objects, gfp_t flags)
486{
487 struct obj_cgroup *objcg;
488
489 if (!memcg_kmem_enabled())
490 return true;
491
492 if (!(flags & __GFP_ACCOUNT) && !(s->flags & SLAB_ACCOUNT))
493 return true;
494
495 objcg = get_obj_cgroup_from_current();
496 if (!objcg)
497 return true;
498
499 if (lru) {
500 int ret;
501 struct mem_cgroup *memcg;
502
503 memcg = get_mem_cgroup_from_objcg(objcg);
504 ret = memcg_list_lru_alloc(memcg, lru, flags);
505 css_put(&memcg->css);
506
507 if (ret)
508 goto out;
509 }
510
511 if (obj_cgroup_charge(objcg, flags, objects * obj_full_size(s)))
512 goto out;
513
514 *objcgp = objcg;
515 return true;
516out:
517 obj_cgroup_put(objcg);
518 return false;
519}
520
521static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
522 struct obj_cgroup *objcg,
523 gfp_t flags, size_t size,
524 void **p)
525{
526 struct slab *slab;
527 unsigned long off;
528 size_t i;
529
530 if (!memcg_kmem_enabled() || !objcg)
531 return;
532
533 for (i = 0; i < size; i++) {
534 if (likely(p[i])) {
535 slab = virt_to_slab(p[i]);
536
537 if (!slab_objcgs(slab) &&
538 memcg_alloc_slab_cgroups(slab, s, flags,
539 false)) {
540 obj_cgroup_uncharge(objcg, obj_full_size(s));
541 continue;
542 }
543
544 off = obj_to_index(s, slab, p[i]);
545 obj_cgroup_get(objcg);
546 slab_objcgs(slab)[off] = objcg;
547 mod_objcg_state(objcg, slab_pgdat(slab),
548 cache_vmstat_idx(s), obj_full_size(s));
549 } else {
550 obj_cgroup_uncharge(objcg, obj_full_size(s));
551 }
552 }
553 obj_cgroup_put(objcg);
554}
555
556static inline void memcg_slab_free_hook(struct kmem_cache *s, struct slab *slab,
557 void **p, int objects)
558{
559 struct obj_cgroup **objcgs;
560 int i;
561
562 if (!memcg_kmem_enabled())
563 return;
564
565 objcgs = slab_objcgs(slab);
566 if (!objcgs)
567 return;
568
569 for (i = 0; i < objects; i++) {
570 struct obj_cgroup *objcg;
571 unsigned int off;
572
573 off = obj_to_index(s, slab, p[i]);
574 objcg = objcgs[off];
575 if (!objcg)
576 continue;
577
578 objcgs[off] = NULL;
579 obj_cgroup_uncharge(objcg, obj_full_size(s));
580 mod_objcg_state(objcg, slab_pgdat(slab), cache_vmstat_idx(s),
581 -obj_full_size(s));
582 obj_cgroup_put(objcg);
583 }
584}
585
586#else /* CONFIG_MEMCG_KMEM */
587static inline struct obj_cgroup **slab_objcgs(struct slab *slab)
588{
589 return NULL;
590}
591
592static inline struct mem_cgroup *memcg_from_slab_obj(void *ptr)
593{
594 return NULL;
595}
596
597static inline int memcg_alloc_slab_cgroups(struct slab *slab,
598 struct kmem_cache *s, gfp_t gfp,
599 bool new_slab)
600{
601 return 0;
602}
603
604static inline void memcg_free_slab_cgroups(struct slab *slab)
605{
606}
607
608static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s,
609 struct list_lru *lru,
610 struct obj_cgroup **objcgp,
611 size_t objects, gfp_t flags)
612{
613 return true;
614}
615
616static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
617 struct obj_cgroup *objcg,
618 gfp_t flags, size_t size,
619 void **p)
620{
621}
622
623static inline void memcg_slab_free_hook(struct kmem_cache *s, struct slab *slab,
624 void **p, int objects)
625{
626}
627#endif /* CONFIG_MEMCG_KMEM */
628
629#ifndef CONFIG_SLOB
630static inline struct kmem_cache *virt_to_cache(const void *obj)
631{
632 struct slab *slab;
633
634 slab = virt_to_slab(obj);
635 if (WARN_ONCE(!slab, "%s: Object is not a Slab page!\n",
636 __func__))
637 return NULL;
638 return slab->slab_cache;
639}
640
641static __always_inline void account_slab(struct slab *slab, int order,
642 struct kmem_cache *s, gfp_t gfp)
643{
644 if (memcg_kmem_enabled() && (s->flags & SLAB_ACCOUNT))
645 memcg_alloc_slab_cgroups(slab, s, gfp, true);
646
647 mod_node_page_state(slab_pgdat(slab), cache_vmstat_idx(s),
648 PAGE_SIZE << order);
649}
650
651static __always_inline void unaccount_slab(struct slab *slab, int order,
652 struct kmem_cache *s)
653{
654 if (memcg_kmem_enabled())
655 memcg_free_slab_cgroups(slab);
656
657 mod_node_page_state(slab_pgdat(slab), cache_vmstat_idx(s),
658 -(PAGE_SIZE << order));
659}
660
661static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x)
662{
663 struct kmem_cache *cachep;
664
665 if (!IS_ENABLED(CONFIG_SLAB_FREELIST_HARDENED) &&
666 !kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS))
667 return s;
668
669 cachep = virt_to_cache(x);
670 if (WARN(cachep && cachep != s,
671 "%s: Wrong slab cache. %s but object is from %s\n",
672 __func__, s->name, cachep->name))
673 print_tracking(cachep, x);
674 return cachep;
675}
676
677void free_large_kmalloc(struct folio *folio, void *object);
678
679#endif /* CONFIG_SLOB */
680
681size_t __ksize(const void *objp);
682
683static inline size_t slab_ksize(const struct kmem_cache *s)
684{
685#ifndef CONFIG_SLUB
686 return s->object_size;
687
688#else /* CONFIG_SLUB */
689# ifdef CONFIG_SLUB_DEBUG
690 /*
691 * Debugging requires use of the padding between object
692 * and whatever may come after it.
693 */
694 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
695 return s->object_size;
696# endif
697 if (s->flags & SLAB_KASAN)
698 return s->object_size;
699 /*
700 * If we have the need to store the freelist pointer
701 * back there or track user information then we can
702 * only use the space before that information.
703 */
704 if (s->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_STORE_USER))
705 return s->inuse;
706 /*
707 * Else we can use all the padding etc for the allocation
708 */
709 return s->size;
710#endif
711}
712
713static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s,
714 struct list_lru *lru,
715 struct obj_cgroup **objcgp,
716 size_t size, gfp_t flags)
717{
718 flags &= gfp_allowed_mask;
719
720 might_alloc(flags);
721
722 if (should_failslab(s, flags))
723 return NULL;
724
725 if (!memcg_slab_pre_alloc_hook(s, lru, objcgp, size, flags))
726 return NULL;
727
728 return s;
729}
730
731static inline void slab_post_alloc_hook(struct kmem_cache *s,
732 struct obj_cgroup *objcg, gfp_t flags,
733 size_t size, void **p, bool init,
734 unsigned int orig_size)
735{
736 unsigned int zero_size = s->object_size;
737 size_t i;
738
739 flags &= gfp_allowed_mask;
740
741 /*
742 * For kmalloc object, the allocated memory size(object_size) is likely
743 * larger than the requested size(orig_size). If redzone check is
744 * enabled for the extra space, don't zero it, as it will be redzoned
745 * soon. The redzone operation for this extra space could be seen as a
746 * replacement of current poisoning under certain debug option, and
747 * won't break other sanity checks.
748 */
749 if (kmem_cache_debug_flags(s, SLAB_STORE_USER | SLAB_RED_ZONE) &&
750 (s->flags & SLAB_KMALLOC))
751 zero_size = orig_size;
752
753 /*
754 * As memory initialization might be integrated into KASAN,
755 * kasan_slab_alloc and initialization memset must be
756 * kept together to avoid discrepancies in behavior.
757 *
758 * As p[i] might get tagged, memset and kmemleak hook come after KASAN.
759 */
760 for (i = 0; i < size; i++) {
761 p[i] = kasan_slab_alloc(s, p[i], flags, init);
762 if (p[i] && init && !kasan_has_integrated_init())
763 memset(p[i], 0, zero_size);
764 kmemleak_alloc_recursive(p[i], s->object_size, 1,
765 s->flags, flags);
766 kmsan_slab_alloc(s, p[i], flags);
767 }
768
769 memcg_slab_post_alloc_hook(s, objcg, flags, size, p);
770}
771
772#ifndef CONFIG_SLOB
773/*
774 * The slab lists for all objects.
775 */
776struct kmem_cache_node {
777#ifdef CONFIG_SLAB
778 raw_spinlock_t list_lock;
779 struct list_head slabs_partial; /* partial list first, better asm code */
780 struct list_head slabs_full;
781 struct list_head slabs_free;
782 unsigned long total_slabs; /* length of all slab lists */
783 unsigned long free_slabs; /* length of free slab list only */
784 unsigned long free_objects;
785 unsigned int free_limit;
786 unsigned int colour_next; /* Per-node cache coloring */
787 struct array_cache *shared; /* shared per node */
788 struct alien_cache **alien; /* on other nodes */
789 unsigned long next_reap; /* updated without locking */
790 int free_touched; /* updated without locking */
791#endif
792
793#ifdef CONFIG_SLUB
794 spinlock_t list_lock;
795 unsigned long nr_partial;
796 struct list_head partial;
797#ifdef CONFIG_SLUB_DEBUG
798 atomic_long_t nr_slabs;
799 atomic_long_t total_objects;
800 struct list_head full;
801#endif
802#endif
803
804};
805
806static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
807{
808 return s->node[node];
809}
810
811/*
812 * Iterator over all nodes. The body will be executed for each node that has
813 * a kmem_cache_node structure allocated (which is true for all online nodes)
814 */
815#define for_each_kmem_cache_node(__s, __node, __n) \
816 for (__node = 0; __node < nr_node_ids; __node++) \
817 if ((__n = get_node(__s, __node)))
818
819#endif
820
821#if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
822void dump_unreclaimable_slab(void);
823#else
824static inline void dump_unreclaimable_slab(void)
825{
826}
827#endif
828
829void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr);
830
831#ifdef CONFIG_SLAB_FREELIST_RANDOM
832int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
833 gfp_t gfp);
834void cache_random_seq_destroy(struct kmem_cache *cachep);
835#else
836static inline int cache_random_seq_create(struct kmem_cache *cachep,
837 unsigned int count, gfp_t gfp)
838{
839 return 0;
840}
841static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { }
842#endif /* CONFIG_SLAB_FREELIST_RANDOM */
843
844static inline bool slab_want_init_on_alloc(gfp_t flags, struct kmem_cache *c)
845{
846 if (static_branch_maybe(CONFIG_INIT_ON_ALLOC_DEFAULT_ON,
847 &init_on_alloc)) {
848 if (c->ctor)
849 return false;
850 if (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON))
851 return flags & __GFP_ZERO;
852 return true;
853 }
854 return flags & __GFP_ZERO;
855}
856
857static inline bool slab_want_init_on_free(struct kmem_cache *c)
858{
859 if (static_branch_maybe(CONFIG_INIT_ON_FREE_DEFAULT_ON,
860 &init_on_free))
861 return !(c->ctor ||
862 (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)));
863 return false;
864}
865
866#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG)
867void debugfs_slab_release(struct kmem_cache *);
868#else
869static inline void debugfs_slab_release(struct kmem_cache *s) { }
870#endif
871
872#ifdef CONFIG_PRINTK
873#define KS_ADDRS_COUNT 16
874struct kmem_obj_info {
875 void *kp_ptr;
876 struct slab *kp_slab;
877 void *kp_objp;
878 unsigned long kp_data_offset;
879 struct kmem_cache *kp_slab_cache;
880 void *kp_ret;
881 void *kp_stack[KS_ADDRS_COUNT];
882 void *kp_free_stack[KS_ADDRS_COUNT];
883};
884void __kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab);
885#endif
886
887#ifdef CONFIG_HAVE_HARDENED_USERCOPY_ALLOCATOR
888void __check_heap_object(const void *ptr, unsigned long n,
889 const struct slab *slab, bool to_user);
890#else
891static inline
892void __check_heap_object(const void *ptr, unsigned long n,
893 const struct slab *slab, bool to_user)
894{
895}
896#endif
897
898#ifdef CONFIG_SLUB_DEBUG
899void skip_orig_size_check(struct kmem_cache *s, const void *object);
900#endif
901
902#endif /* MM_SLAB_H */
1#ifndef MM_SLAB_H
2#define MM_SLAB_H
3/*
4 * Internal slab definitions
5 */
6
7#ifdef CONFIG_SLOB
8/*
9 * Common fields provided in kmem_cache by all slab allocators
10 * This struct is either used directly by the allocator (SLOB)
11 * or the allocator must include definitions for all fields
12 * provided in kmem_cache_common in their definition of kmem_cache.
13 *
14 * Once we can do anonymous structs (C11 standard) we could put a
15 * anonymous struct definition in these allocators so that the
16 * separate allocations in the kmem_cache structure of SLAB and
17 * SLUB is no longer needed.
18 */
19struct kmem_cache {
20 unsigned int object_size;/* The original size of the object */
21 unsigned int size; /* The aligned/padded/added on size */
22 unsigned int align; /* Alignment as calculated */
23 unsigned long flags; /* Active flags on the slab */
24 const char *name; /* Slab name for sysfs */
25 int refcount; /* Use counter */
26 void (*ctor)(void *); /* Called on object slot creation */
27 struct list_head list; /* List of all slab caches on the system */
28};
29
30#endif /* CONFIG_SLOB */
31
32#ifdef CONFIG_SLAB
33#include <linux/slab_def.h>
34#endif
35
36#ifdef CONFIG_SLUB
37#include <linux/slub_def.h>
38#endif
39
40#include <linux/memcontrol.h>
41#include <linux/fault-inject.h>
42#include <linux/kmemcheck.h>
43#include <linux/kasan.h>
44#include <linux/kmemleak.h>
45
46/*
47 * State of the slab allocator.
48 *
49 * This is used to describe the states of the allocator during bootup.
50 * Allocators use this to gradually bootstrap themselves. Most allocators
51 * have the problem that the structures used for managing slab caches are
52 * allocated from slab caches themselves.
53 */
54enum slab_state {
55 DOWN, /* No slab functionality yet */
56 PARTIAL, /* SLUB: kmem_cache_node available */
57 PARTIAL_NODE, /* SLAB: kmalloc size for node struct available */
58 UP, /* Slab caches usable but not all extras yet */
59 FULL /* Everything is working */
60};
61
62extern enum slab_state slab_state;
63
64/* The slab cache mutex protects the management structures during changes */
65extern struct mutex slab_mutex;
66
67/* The list of all slab caches on the system */
68extern struct list_head slab_caches;
69
70/* The slab cache that manages slab cache information */
71extern struct kmem_cache *kmem_cache;
72
73unsigned long calculate_alignment(unsigned long flags,
74 unsigned long align, unsigned long size);
75
76#ifndef CONFIG_SLOB
77/* Kmalloc array related functions */
78void setup_kmalloc_cache_index_table(void);
79void create_kmalloc_caches(unsigned long);
80
81/* Find the kmalloc slab corresponding for a certain size */
82struct kmem_cache *kmalloc_slab(size_t, gfp_t);
83#endif
84
85
86/* Functions provided by the slab allocators */
87extern int __kmem_cache_create(struct kmem_cache *, unsigned long flags);
88
89extern struct kmem_cache *create_kmalloc_cache(const char *name, size_t size,
90 unsigned long flags);
91extern void create_boot_cache(struct kmem_cache *, const char *name,
92 size_t size, unsigned long flags);
93
94int slab_unmergeable(struct kmem_cache *s);
95struct kmem_cache *find_mergeable(size_t size, size_t align,
96 unsigned long flags, const char *name, void (*ctor)(void *));
97#ifndef CONFIG_SLOB
98struct kmem_cache *
99__kmem_cache_alias(const char *name, size_t size, size_t align,
100 unsigned long flags, void (*ctor)(void *));
101
102unsigned long kmem_cache_flags(unsigned long object_size,
103 unsigned long flags, const char *name,
104 void (*ctor)(void *));
105#else
106static inline struct kmem_cache *
107__kmem_cache_alias(const char *name, size_t size, size_t align,
108 unsigned long flags, void (*ctor)(void *))
109{ return NULL; }
110
111static inline unsigned long kmem_cache_flags(unsigned long object_size,
112 unsigned long flags, const char *name,
113 void (*ctor)(void *))
114{
115 return flags;
116}
117#endif
118
119
120/* Legal flag mask for kmem_cache_create(), for various configurations */
121#define SLAB_CORE_FLAGS (SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA | SLAB_PANIC | \
122 SLAB_DESTROY_BY_RCU | SLAB_DEBUG_OBJECTS )
123
124#if defined(CONFIG_DEBUG_SLAB)
125#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
126#elif defined(CONFIG_SLUB_DEBUG)
127#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
128 SLAB_TRACE | SLAB_CONSISTENCY_CHECKS)
129#else
130#define SLAB_DEBUG_FLAGS (0)
131#endif
132
133#if defined(CONFIG_SLAB)
134#define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \
135 SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \
136 SLAB_NOTRACK | SLAB_ACCOUNT)
137#elif defined(CONFIG_SLUB)
138#define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
139 SLAB_TEMPORARY | SLAB_NOTRACK | SLAB_ACCOUNT)
140#else
141#define SLAB_CACHE_FLAGS (0)
142#endif
143
144#define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
145
146int __kmem_cache_shutdown(struct kmem_cache *);
147void __kmem_cache_release(struct kmem_cache *);
148int __kmem_cache_shrink(struct kmem_cache *, bool);
149void slab_kmem_cache_release(struct kmem_cache *);
150
151struct seq_file;
152struct file;
153
154struct slabinfo {
155 unsigned long active_objs;
156 unsigned long num_objs;
157 unsigned long active_slabs;
158 unsigned long num_slabs;
159 unsigned long shared_avail;
160 unsigned int limit;
161 unsigned int batchcount;
162 unsigned int shared;
163 unsigned int objects_per_slab;
164 unsigned int cache_order;
165};
166
167void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo);
168void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s);
169ssize_t slabinfo_write(struct file *file, const char __user *buffer,
170 size_t count, loff_t *ppos);
171
172/*
173 * Generic implementation of bulk operations
174 * These are useful for situations in which the allocator cannot
175 * perform optimizations. In that case segments of the object listed
176 * may be allocated or freed using these operations.
177 */
178void __kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
179int __kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
180
181#if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
182/*
183 * Iterate over all memcg caches of the given root cache. The caller must hold
184 * slab_mutex.
185 */
186#define for_each_memcg_cache(iter, root) \
187 list_for_each_entry(iter, &(root)->memcg_params.list, \
188 memcg_params.list)
189
190static inline bool is_root_cache(struct kmem_cache *s)
191{
192 return s->memcg_params.is_root_cache;
193}
194
195static inline bool slab_equal_or_root(struct kmem_cache *s,
196 struct kmem_cache *p)
197{
198 return p == s || p == s->memcg_params.root_cache;
199}
200
201/*
202 * We use suffixes to the name in memcg because we can't have caches
203 * created in the system with the same name. But when we print them
204 * locally, better refer to them with the base name
205 */
206static inline const char *cache_name(struct kmem_cache *s)
207{
208 if (!is_root_cache(s))
209 s = s->memcg_params.root_cache;
210 return s->name;
211}
212
213/*
214 * Note, we protect with RCU only the memcg_caches array, not per-memcg caches.
215 * That said the caller must assure the memcg's cache won't go away by either
216 * taking a css reference to the owner cgroup, or holding the slab_mutex.
217 */
218static inline struct kmem_cache *
219cache_from_memcg_idx(struct kmem_cache *s, int idx)
220{
221 struct kmem_cache *cachep;
222 struct memcg_cache_array *arr;
223
224 rcu_read_lock();
225 arr = rcu_dereference(s->memcg_params.memcg_caches);
226
227 /*
228 * Make sure we will access the up-to-date value. The code updating
229 * memcg_caches issues a write barrier to match this (see
230 * memcg_create_kmem_cache()).
231 */
232 cachep = lockless_dereference(arr->entries[idx]);
233 rcu_read_unlock();
234
235 return cachep;
236}
237
238static inline struct kmem_cache *memcg_root_cache(struct kmem_cache *s)
239{
240 if (is_root_cache(s))
241 return s;
242 return s->memcg_params.root_cache;
243}
244
245static __always_inline int memcg_charge_slab(struct page *page,
246 gfp_t gfp, int order,
247 struct kmem_cache *s)
248{
249 int ret;
250
251 if (!memcg_kmem_enabled())
252 return 0;
253 if (is_root_cache(s))
254 return 0;
255
256 ret = __memcg_kmem_charge_memcg(page, gfp, order,
257 s->memcg_params.memcg);
258 if (ret)
259 return ret;
260
261 memcg_kmem_update_page_stat(page,
262 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
263 MEMCG_SLAB_RECLAIMABLE : MEMCG_SLAB_UNRECLAIMABLE,
264 1 << order);
265 return 0;
266}
267
268static __always_inline void memcg_uncharge_slab(struct page *page, int order,
269 struct kmem_cache *s)
270{
271 memcg_kmem_update_page_stat(page,
272 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
273 MEMCG_SLAB_RECLAIMABLE : MEMCG_SLAB_UNRECLAIMABLE,
274 -(1 << order));
275 memcg_kmem_uncharge(page, order);
276}
277
278extern void slab_init_memcg_params(struct kmem_cache *);
279
280#else /* CONFIG_MEMCG && !CONFIG_SLOB */
281
282#define for_each_memcg_cache(iter, root) \
283 for ((void)(iter), (void)(root); 0; )
284
285static inline bool is_root_cache(struct kmem_cache *s)
286{
287 return true;
288}
289
290static inline bool slab_equal_or_root(struct kmem_cache *s,
291 struct kmem_cache *p)
292{
293 return true;
294}
295
296static inline const char *cache_name(struct kmem_cache *s)
297{
298 return s->name;
299}
300
301static inline struct kmem_cache *
302cache_from_memcg_idx(struct kmem_cache *s, int idx)
303{
304 return NULL;
305}
306
307static inline struct kmem_cache *memcg_root_cache(struct kmem_cache *s)
308{
309 return s;
310}
311
312static inline int memcg_charge_slab(struct page *page, gfp_t gfp, int order,
313 struct kmem_cache *s)
314{
315 return 0;
316}
317
318static inline void memcg_uncharge_slab(struct page *page, int order,
319 struct kmem_cache *s)
320{
321}
322
323static inline void slab_init_memcg_params(struct kmem_cache *s)
324{
325}
326#endif /* CONFIG_MEMCG && !CONFIG_SLOB */
327
328static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x)
329{
330 struct kmem_cache *cachep;
331 struct page *page;
332
333 /*
334 * When kmemcg is not being used, both assignments should return the
335 * same value. but we don't want to pay the assignment price in that
336 * case. If it is not compiled in, the compiler should be smart enough
337 * to not do even the assignment. In that case, slab_equal_or_root
338 * will also be a constant.
339 */
340 if (!memcg_kmem_enabled() &&
341 !unlikely(s->flags & SLAB_CONSISTENCY_CHECKS))
342 return s;
343
344 page = virt_to_head_page(x);
345 cachep = page->slab_cache;
346 if (slab_equal_or_root(cachep, s))
347 return cachep;
348
349 pr_err("%s: Wrong slab cache. %s but object is from %s\n",
350 __func__, s->name, cachep->name);
351 WARN_ON_ONCE(1);
352 return s;
353}
354
355static inline size_t slab_ksize(const struct kmem_cache *s)
356{
357#ifndef CONFIG_SLUB
358 return s->object_size;
359
360#else /* CONFIG_SLUB */
361# ifdef CONFIG_SLUB_DEBUG
362 /*
363 * Debugging requires use of the padding between object
364 * and whatever may come after it.
365 */
366 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
367 return s->object_size;
368# endif
369 /*
370 * If we have the need to store the freelist pointer
371 * back there or track user information then we can
372 * only use the space before that information.
373 */
374 if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER))
375 return s->inuse;
376 /*
377 * Else we can use all the padding etc for the allocation
378 */
379 return s->size;
380#endif
381}
382
383static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s,
384 gfp_t flags)
385{
386 flags &= gfp_allowed_mask;
387 lockdep_trace_alloc(flags);
388 might_sleep_if(gfpflags_allow_blocking(flags));
389
390 if (should_failslab(s, flags))
391 return NULL;
392
393 return memcg_kmem_get_cache(s, flags);
394}
395
396static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags,
397 size_t size, void **p)
398{
399 size_t i;
400
401 flags &= gfp_allowed_mask;
402 for (i = 0; i < size; i++) {
403 void *object = p[i];
404
405 kmemcheck_slab_alloc(s, flags, object, slab_ksize(s));
406 kmemleak_alloc_recursive(object, s->object_size, 1,
407 s->flags, flags);
408 kasan_slab_alloc(s, object, flags);
409 }
410 memcg_kmem_put_cache(s);
411}
412
413#ifndef CONFIG_SLOB
414/*
415 * The slab lists for all objects.
416 */
417struct kmem_cache_node {
418 spinlock_t list_lock;
419
420#ifdef CONFIG_SLAB
421 struct list_head slabs_partial; /* partial list first, better asm code */
422 struct list_head slabs_full;
423 struct list_head slabs_free;
424 unsigned long free_objects;
425 unsigned int free_limit;
426 unsigned int colour_next; /* Per-node cache coloring */
427 struct array_cache *shared; /* shared per node */
428 struct alien_cache **alien; /* on other nodes */
429 unsigned long next_reap; /* updated without locking */
430 int free_touched; /* updated without locking */
431#endif
432
433#ifdef CONFIG_SLUB
434 unsigned long nr_partial;
435 struct list_head partial;
436#ifdef CONFIG_SLUB_DEBUG
437 atomic_long_t nr_slabs;
438 atomic_long_t total_objects;
439 struct list_head full;
440#endif
441#endif
442
443};
444
445static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
446{
447 return s->node[node];
448}
449
450/*
451 * Iterator over all nodes. The body will be executed for each node that has
452 * a kmem_cache_node structure allocated (which is true for all online nodes)
453 */
454#define for_each_kmem_cache_node(__s, __node, __n) \
455 for (__node = 0; __node < nr_node_ids; __node++) \
456 if ((__n = get_node(__s, __node)))
457
458#endif
459
460void *slab_start(struct seq_file *m, loff_t *pos);
461void *slab_next(struct seq_file *m, void *p, loff_t *pos);
462void slab_stop(struct seq_file *m, void *p);
463int memcg_slab_show(struct seq_file *m, void *p);
464
465#endif /* MM_SLAB_H */