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#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
50/*
51 * State of the slab allocator.
52 *
53 * This is used to describe the states of the allocator during bootup.
54 * Allocators use this to gradually bootstrap themselves. Most allocators
55 * have the problem that the structures used for managing slab caches are
56 * allocated from slab caches themselves.
57 */
58enum slab_state {
59 DOWN, /* No slab functionality yet */
60 PARTIAL, /* SLUB: kmem_cache_node available */
61 PARTIAL_NODE, /* SLAB: kmalloc size for node struct available */
62 UP, /* Slab caches usable but not all extras yet */
63 FULL /* Everything is working */
64};
65
66extern enum slab_state slab_state;
67
68/* The slab cache mutex protects the management structures during changes */
69extern struct mutex slab_mutex;
70
71/* The list of all slab caches on the system */
72extern struct list_head slab_caches;
73
74/* The slab cache that manages slab cache information */
75extern struct kmem_cache *kmem_cache;
76
77/* A table of kmalloc cache names and sizes */
78extern const struct kmalloc_info_struct {
79 const char *name[NR_KMALLOC_TYPES];
80 unsigned int size;
81} kmalloc_info[];
82
83#ifndef CONFIG_SLOB
84/* Kmalloc array related functions */
85void setup_kmalloc_cache_index_table(void);
86void create_kmalloc_caches(slab_flags_t);
87
88/* Find the kmalloc slab corresponding for a certain size */
89struct kmem_cache *kmalloc_slab(size_t, gfp_t);
90#endif
91
92gfp_t kmalloc_fix_flags(gfp_t flags);
93
94/* Functions provided by the slab allocators */
95int __kmem_cache_create(struct kmem_cache *, slab_flags_t flags);
96
97struct kmem_cache *create_kmalloc_cache(const char *name, unsigned int size,
98 slab_flags_t flags, unsigned int useroffset,
99 unsigned int usersize);
100extern void create_boot_cache(struct kmem_cache *, const char *name,
101 unsigned int size, slab_flags_t flags,
102 unsigned int useroffset, unsigned int usersize);
103
104int slab_unmergeable(struct kmem_cache *s);
105struct kmem_cache *find_mergeable(unsigned size, unsigned align,
106 slab_flags_t flags, const char *name, void (*ctor)(void *));
107#ifndef CONFIG_SLOB
108struct kmem_cache *
109__kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
110 slab_flags_t flags, void (*ctor)(void *));
111
112slab_flags_t kmem_cache_flags(unsigned int object_size,
113 slab_flags_t flags, const char *name);
114#else
115static inline struct kmem_cache *
116__kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
117 slab_flags_t flags, void (*ctor)(void *))
118{ return NULL; }
119
120static inline slab_flags_t kmem_cache_flags(unsigned int object_size,
121 slab_flags_t flags, const char *name)
122{
123 return flags;
124}
125#endif
126
127
128/* Legal flag mask for kmem_cache_create(), for various configurations */
129#define SLAB_CORE_FLAGS (SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA | \
130 SLAB_CACHE_DMA32 | SLAB_PANIC | \
131 SLAB_TYPESAFE_BY_RCU | SLAB_DEBUG_OBJECTS )
132
133#if defined(CONFIG_DEBUG_SLAB)
134#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
135#elif defined(CONFIG_SLUB_DEBUG)
136#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
137 SLAB_TRACE | SLAB_CONSISTENCY_CHECKS)
138#else
139#define SLAB_DEBUG_FLAGS (0)
140#endif
141
142#if defined(CONFIG_SLAB)
143#define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \
144 SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \
145 SLAB_ACCOUNT)
146#elif defined(CONFIG_SLUB)
147#define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
148 SLAB_TEMPORARY | SLAB_ACCOUNT)
149#else
150#define SLAB_CACHE_FLAGS (0)
151#endif
152
153/* Common flags available with current configuration */
154#define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
155
156/* Common flags permitted for kmem_cache_create */
157#define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \
158 SLAB_RED_ZONE | \
159 SLAB_POISON | \
160 SLAB_STORE_USER | \
161 SLAB_TRACE | \
162 SLAB_CONSISTENCY_CHECKS | \
163 SLAB_MEM_SPREAD | \
164 SLAB_NOLEAKTRACE | \
165 SLAB_RECLAIM_ACCOUNT | \
166 SLAB_TEMPORARY | \
167 SLAB_ACCOUNT)
168
169bool __kmem_cache_empty(struct kmem_cache *);
170int __kmem_cache_shutdown(struct kmem_cache *);
171void __kmem_cache_release(struct kmem_cache *);
172int __kmem_cache_shrink(struct kmem_cache *);
173void slab_kmem_cache_release(struct kmem_cache *);
174
175struct seq_file;
176struct file;
177
178struct slabinfo {
179 unsigned long active_objs;
180 unsigned long num_objs;
181 unsigned long active_slabs;
182 unsigned long num_slabs;
183 unsigned long shared_avail;
184 unsigned int limit;
185 unsigned int batchcount;
186 unsigned int shared;
187 unsigned int objects_per_slab;
188 unsigned int cache_order;
189};
190
191void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo);
192void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s);
193ssize_t slabinfo_write(struct file *file, const char __user *buffer,
194 size_t count, loff_t *ppos);
195
196/*
197 * Generic implementation of bulk operations
198 * These are useful for situations in which the allocator cannot
199 * perform optimizations. In that case segments of the object listed
200 * may be allocated or freed using these operations.
201 */
202void __kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
203int __kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
204
205static inline enum node_stat_item cache_vmstat_idx(struct kmem_cache *s)
206{
207 return (s->flags & SLAB_RECLAIM_ACCOUNT) ?
208 NR_SLAB_RECLAIMABLE_B : NR_SLAB_UNRECLAIMABLE_B;
209}
210
211#ifdef CONFIG_SLUB_DEBUG
212#ifdef CONFIG_SLUB_DEBUG_ON
213DECLARE_STATIC_KEY_TRUE(slub_debug_enabled);
214#else
215DECLARE_STATIC_KEY_FALSE(slub_debug_enabled);
216#endif
217extern void print_tracking(struct kmem_cache *s, void *object);
218long validate_slab_cache(struct kmem_cache *s);
219static inline bool __slub_debug_enabled(void)
220{
221 return static_branch_unlikely(&slub_debug_enabled);
222}
223#else
224static inline void print_tracking(struct kmem_cache *s, void *object)
225{
226}
227static inline bool __slub_debug_enabled(void)
228{
229 return false;
230}
231#endif
232
233/*
234 * Returns true if any of the specified slub_debug flags is enabled for the
235 * cache. Use only for flags parsed by setup_slub_debug() as it also enables
236 * the static key.
237 */
238static inline bool kmem_cache_debug_flags(struct kmem_cache *s, slab_flags_t flags)
239{
240 if (IS_ENABLED(CONFIG_SLUB_DEBUG))
241 VM_WARN_ON_ONCE(!(flags & SLAB_DEBUG_FLAGS));
242 if (__slub_debug_enabled())
243 return s->flags & flags;
244 return false;
245}
246
247#ifdef CONFIG_MEMCG_KMEM
248int memcg_alloc_page_obj_cgroups(struct page *page, struct kmem_cache *s,
249 gfp_t gfp, bool new_page);
250void mod_objcg_state(struct obj_cgroup *objcg, struct pglist_data *pgdat,
251 enum node_stat_item idx, int nr);
252
253static inline void memcg_free_page_obj_cgroups(struct page *page)
254{
255 kfree(page_objcgs(page));
256 page->memcg_data = 0;
257}
258
259static inline size_t obj_full_size(struct kmem_cache *s)
260{
261 /*
262 * For each accounted object there is an extra space which is used
263 * to store obj_cgroup membership. Charge it too.
264 */
265 return s->size + sizeof(struct obj_cgroup *);
266}
267
268/*
269 * Returns false if the allocation should fail.
270 */
271static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s,
272 struct obj_cgroup **objcgp,
273 size_t objects, gfp_t flags)
274{
275 struct obj_cgroup *objcg;
276
277 if (!memcg_kmem_enabled())
278 return true;
279
280 if (!(flags & __GFP_ACCOUNT) && !(s->flags & SLAB_ACCOUNT))
281 return true;
282
283 objcg = get_obj_cgroup_from_current();
284 if (!objcg)
285 return true;
286
287 if (obj_cgroup_charge(objcg, flags, objects * obj_full_size(s))) {
288 obj_cgroup_put(objcg);
289 return false;
290 }
291
292 *objcgp = objcg;
293 return true;
294}
295
296static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
297 struct obj_cgroup *objcg,
298 gfp_t flags, size_t size,
299 void **p)
300{
301 struct page *page;
302 unsigned long off;
303 size_t i;
304
305 if (!memcg_kmem_enabled() || !objcg)
306 return;
307
308 for (i = 0; i < size; i++) {
309 if (likely(p[i])) {
310 page = virt_to_head_page(p[i]);
311
312 if (!page_objcgs(page) &&
313 memcg_alloc_page_obj_cgroups(page, s, flags,
314 false)) {
315 obj_cgroup_uncharge(objcg, obj_full_size(s));
316 continue;
317 }
318
319 off = obj_to_index(s, page, p[i]);
320 obj_cgroup_get(objcg);
321 page_objcgs(page)[off] = objcg;
322 mod_objcg_state(objcg, page_pgdat(page),
323 cache_vmstat_idx(s), obj_full_size(s));
324 } else {
325 obj_cgroup_uncharge(objcg, obj_full_size(s));
326 }
327 }
328 obj_cgroup_put(objcg);
329}
330
331static inline void memcg_slab_free_hook(struct kmem_cache *s_orig,
332 void **p, int objects)
333{
334 struct kmem_cache *s;
335 struct obj_cgroup **objcgs;
336 struct obj_cgroup *objcg;
337 struct page *page;
338 unsigned int off;
339 int i;
340
341 if (!memcg_kmem_enabled())
342 return;
343
344 for (i = 0; i < objects; i++) {
345 if (unlikely(!p[i]))
346 continue;
347
348 page = virt_to_head_page(p[i]);
349 objcgs = page_objcgs_check(page);
350 if (!objcgs)
351 continue;
352
353 if (!s_orig)
354 s = page->slab_cache;
355 else
356 s = s_orig;
357
358 off = obj_to_index(s, page, p[i]);
359 objcg = objcgs[off];
360 if (!objcg)
361 continue;
362
363 objcgs[off] = NULL;
364 obj_cgroup_uncharge(objcg, obj_full_size(s));
365 mod_objcg_state(objcg, page_pgdat(page), cache_vmstat_idx(s),
366 -obj_full_size(s));
367 obj_cgroup_put(objcg);
368 }
369}
370
371#else /* CONFIG_MEMCG_KMEM */
372static inline struct mem_cgroup *memcg_from_slab_obj(void *ptr)
373{
374 return NULL;
375}
376
377static inline int memcg_alloc_page_obj_cgroups(struct page *page,
378 struct kmem_cache *s, gfp_t gfp,
379 bool new_page)
380{
381 return 0;
382}
383
384static inline void memcg_free_page_obj_cgroups(struct page *page)
385{
386}
387
388static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s,
389 struct obj_cgroup **objcgp,
390 size_t objects, gfp_t flags)
391{
392 return true;
393}
394
395static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
396 struct obj_cgroup *objcg,
397 gfp_t flags, size_t size,
398 void **p)
399{
400}
401
402static inline void memcg_slab_free_hook(struct kmem_cache *s,
403 void **p, int objects)
404{
405}
406#endif /* CONFIG_MEMCG_KMEM */
407
408static inline struct kmem_cache *virt_to_cache(const void *obj)
409{
410 struct page *page;
411
412 page = virt_to_head_page(obj);
413 if (WARN_ONCE(!PageSlab(page), "%s: Object is not a Slab page!\n",
414 __func__))
415 return NULL;
416 return page->slab_cache;
417}
418
419static __always_inline void account_slab_page(struct page *page, int order,
420 struct kmem_cache *s,
421 gfp_t gfp)
422{
423 if (memcg_kmem_enabled() && (s->flags & SLAB_ACCOUNT))
424 memcg_alloc_page_obj_cgroups(page, s, gfp, true);
425
426 mod_node_page_state(page_pgdat(page), cache_vmstat_idx(s),
427 PAGE_SIZE << order);
428}
429
430static __always_inline void unaccount_slab_page(struct page *page, int order,
431 struct kmem_cache *s)
432{
433 if (memcg_kmem_enabled())
434 memcg_free_page_obj_cgroups(page);
435
436 mod_node_page_state(page_pgdat(page), cache_vmstat_idx(s),
437 -(PAGE_SIZE << order));
438}
439
440static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x)
441{
442 struct kmem_cache *cachep;
443
444 if (!IS_ENABLED(CONFIG_SLAB_FREELIST_HARDENED) &&
445 !kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS))
446 return s;
447
448 cachep = virt_to_cache(x);
449 if (WARN(cachep && cachep != s,
450 "%s: Wrong slab cache. %s but object is from %s\n",
451 __func__, s->name, cachep->name))
452 print_tracking(cachep, x);
453 return cachep;
454}
455
456static inline size_t slab_ksize(const struct kmem_cache *s)
457{
458#ifndef CONFIG_SLUB
459 return s->object_size;
460
461#else /* CONFIG_SLUB */
462# ifdef CONFIG_SLUB_DEBUG
463 /*
464 * Debugging requires use of the padding between object
465 * and whatever may come after it.
466 */
467 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
468 return s->object_size;
469# endif
470 if (s->flags & SLAB_KASAN)
471 return s->object_size;
472 /*
473 * If we have the need to store the freelist pointer
474 * back there or track user information then we can
475 * only use the space before that information.
476 */
477 if (s->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_STORE_USER))
478 return s->inuse;
479 /*
480 * Else we can use all the padding etc for the allocation
481 */
482 return s->size;
483#endif
484}
485
486static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s,
487 struct obj_cgroup **objcgp,
488 size_t size, gfp_t flags)
489{
490 flags &= gfp_allowed_mask;
491
492 might_alloc(flags);
493
494 if (should_failslab(s, flags))
495 return NULL;
496
497 if (!memcg_slab_pre_alloc_hook(s, objcgp, size, flags))
498 return NULL;
499
500 return s;
501}
502
503static inline void slab_post_alloc_hook(struct kmem_cache *s,
504 struct obj_cgroup *objcg, gfp_t flags,
505 size_t size, void **p, bool init)
506{
507 size_t i;
508
509 flags &= gfp_allowed_mask;
510
511 /*
512 * As memory initialization might be integrated into KASAN,
513 * kasan_slab_alloc and initialization memset must be
514 * kept together to avoid discrepancies in behavior.
515 *
516 * As p[i] might get tagged, memset and kmemleak hook come after KASAN.
517 */
518 for (i = 0; i < size; i++) {
519 p[i] = kasan_slab_alloc(s, p[i], flags, init);
520 if (p[i] && init && !kasan_has_integrated_init())
521 memset(p[i], 0, s->object_size);
522 kmemleak_alloc_recursive(p[i], s->object_size, 1,
523 s->flags, flags);
524 }
525
526 memcg_slab_post_alloc_hook(s, objcg, flags, size, p);
527}
528
529#ifndef CONFIG_SLOB
530/*
531 * The slab lists for all objects.
532 */
533struct kmem_cache_node {
534 spinlock_t list_lock;
535
536#ifdef CONFIG_SLAB
537 struct list_head slabs_partial; /* partial list first, better asm code */
538 struct list_head slabs_full;
539 struct list_head slabs_free;
540 unsigned long total_slabs; /* length of all slab lists */
541 unsigned long free_slabs; /* length of free slab list only */
542 unsigned long free_objects;
543 unsigned int free_limit;
544 unsigned int colour_next; /* Per-node cache coloring */
545 struct array_cache *shared; /* shared per node */
546 struct alien_cache **alien; /* on other nodes */
547 unsigned long next_reap; /* updated without locking */
548 int free_touched; /* updated without locking */
549#endif
550
551#ifdef CONFIG_SLUB
552 unsigned long nr_partial;
553 struct list_head partial;
554#ifdef CONFIG_SLUB_DEBUG
555 atomic_long_t nr_slabs;
556 atomic_long_t total_objects;
557 struct list_head full;
558#endif
559#endif
560
561};
562
563static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
564{
565 return s->node[node];
566}
567
568/*
569 * Iterator over all nodes. The body will be executed for each node that has
570 * a kmem_cache_node structure allocated (which is true for all online nodes)
571 */
572#define for_each_kmem_cache_node(__s, __node, __n) \
573 for (__node = 0; __node < nr_node_ids; __node++) \
574 if ((__n = get_node(__s, __node)))
575
576#endif
577
578void *slab_start(struct seq_file *m, loff_t *pos);
579void *slab_next(struct seq_file *m, void *p, loff_t *pos);
580void slab_stop(struct seq_file *m, void *p);
581int memcg_slab_show(struct seq_file *m, void *p);
582
583#if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
584void dump_unreclaimable_slab(void);
585#else
586static inline void dump_unreclaimable_slab(void)
587{
588}
589#endif
590
591void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr);
592
593#ifdef CONFIG_SLAB_FREELIST_RANDOM
594int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
595 gfp_t gfp);
596void cache_random_seq_destroy(struct kmem_cache *cachep);
597#else
598static inline int cache_random_seq_create(struct kmem_cache *cachep,
599 unsigned int count, gfp_t gfp)
600{
601 return 0;
602}
603static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { }
604#endif /* CONFIG_SLAB_FREELIST_RANDOM */
605
606static inline bool slab_want_init_on_alloc(gfp_t flags, struct kmem_cache *c)
607{
608 if (static_branch_maybe(CONFIG_INIT_ON_ALLOC_DEFAULT_ON,
609 &init_on_alloc)) {
610 if (c->ctor)
611 return false;
612 if (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON))
613 return flags & __GFP_ZERO;
614 return true;
615 }
616 return flags & __GFP_ZERO;
617}
618
619static inline bool slab_want_init_on_free(struct kmem_cache *c)
620{
621 if (static_branch_maybe(CONFIG_INIT_ON_FREE_DEFAULT_ON,
622 &init_on_free))
623 return !(c->ctor ||
624 (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)));
625 return false;
626}
627
628#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG)
629void debugfs_slab_release(struct kmem_cache *);
630#else
631static inline void debugfs_slab_release(struct kmem_cache *s) { }
632#endif
633
634#ifdef CONFIG_PRINTK
635#define KS_ADDRS_COUNT 16
636struct kmem_obj_info {
637 void *kp_ptr;
638 struct page *kp_page;
639 void *kp_objp;
640 unsigned long kp_data_offset;
641 struct kmem_cache *kp_slab_cache;
642 void *kp_ret;
643 void *kp_stack[KS_ADDRS_COUNT];
644 void *kp_free_stack[KS_ADDRS_COUNT];
645};
646void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct page *page);
647#endif
648
649#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#include <linux/random.h>
46
47/*
48 * State of the slab allocator.
49 *
50 * This is used to describe the states of the allocator during bootup.
51 * Allocators use this to gradually bootstrap themselves. Most allocators
52 * have the problem that the structures used for managing slab caches are
53 * allocated from slab caches themselves.
54 */
55enum slab_state {
56 DOWN, /* No slab functionality yet */
57 PARTIAL, /* SLUB: kmem_cache_node available */
58 PARTIAL_NODE, /* SLAB: kmalloc size for node struct available */
59 UP, /* Slab caches usable but not all extras yet */
60 FULL /* Everything is working */
61};
62
63extern enum slab_state slab_state;
64
65/* The slab cache mutex protects the management structures during changes */
66extern struct mutex slab_mutex;
67
68/* The list of all slab caches on the system */
69extern struct list_head slab_caches;
70
71/* The slab cache that manages slab cache information */
72extern struct kmem_cache *kmem_cache;
73
74unsigned long calculate_alignment(unsigned long flags,
75 unsigned long align, unsigned long size);
76
77#ifndef CONFIG_SLOB
78/* Kmalloc array related functions */
79void setup_kmalloc_cache_index_table(void);
80void create_kmalloc_caches(unsigned long);
81
82/* Find the kmalloc slab corresponding for a certain size */
83struct kmem_cache *kmalloc_slab(size_t, gfp_t);
84#endif
85
86
87/* Functions provided by the slab allocators */
88extern int __kmem_cache_create(struct kmem_cache *, unsigned long flags);
89
90extern struct kmem_cache *create_kmalloc_cache(const char *name, size_t size,
91 unsigned long flags);
92extern void create_boot_cache(struct kmem_cache *, const char *name,
93 size_t size, unsigned long flags);
94
95int slab_unmergeable(struct kmem_cache *s);
96struct kmem_cache *find_mergeable(size_t size, size_t align,
97 unsigned long flags, const char *name, void (*ctor)(void *));
98#ifndef CONFIG_SLOB
99struct kmem_cache *
100__kmem_cache_alias(const char *name, size_t size, size_t align,
101 unsigned long flags, void (*ctor)(void *));
102
103unsigned long kmem_cache_flags(unsigned long object_size,
104 unsigned long flags, const char *name,
105 void (*ctor)(void *));
106#else
107static inline struct kmem_cache *
108__kmem_cache_alias(const char *name, size_t size, size_t align,
109 unsigned long flags, void (*ctor)(void *))
110{ return NULL; }
111
112static inline unsigned long kmem_cache_flags(unsigned long object_size,
113 unsigned long flags, const char *name,
114 void (*ctor)(void *))
115{
116 return flags;
117}
118#endif
119
120
121/* Legal flag mask for kmem_cache_create(), for various configurations */
122#define SLAB_CORE_FLAGS (SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA | SLAB_PANIC | \
123 SLAB_DESTROY_BY_RCU | SLAB_DEBUG_OBJECTS )
124
125#if defined(CONFIG_DEBUG_SLAB)
126#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
127#elif defined(CONFIG_SLUB_DEBUG)
128#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
129 SLAB_TRACE | SLAB_CONSISTENCY_CHECKS)
130#else
131#define SLAB_DEBUG_FLAGS (0)
132#endif
133
134#if defined(CONFIG_SLAB)
135#define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \
136 SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \
137 SLAB_NOTRACK | SLAB_ACCOUNT)
138#elif defined(CONFIG_SLUB)
139#define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
140 SLAB_TEMPORARY | SLAB_NOTRACK | SLAB_ACCOUNT)
141#else
142#define SLAB_CACHE_FLAGS (0)
143#endif
144
145/* Common flags available with current configuration */
146#define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
147
148/* Common flags permitted for kmem_cache_create */
149#define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \
150 SLAB_RED_ZONE | \
151 SLAB_POISON | \
152 SLAB_STORE_USER | \
153 SLAB_TRACE | \
154 SLAB_CONSISTENCY_CHECKS | \
155 SLAB_MEM_SPREAD | \
156 SLAB_NOLEAKTRACE | \
157 SLAB_RECLAIM_ACCOUNT | \
158 SLAB_TEMPORARY | \
159 SLAB_NOTRACK | \
160 SLAB_ACCOUNT)
161
162int __kmem_cache_shutdown(struct kmem_cache *);
163void __kmem_cache_release(struct kmem_cache *);
164int __kmem_cache_shrink(struct kmem_cache *);
165void slab_kmem_cache_release(struct kmem_cache *);
166
167struct seq_file;
168struct file;
169
170struct slabinfo {
171 unsigned long active_objs;
172 unsigned long num_objs;
173 unsigned long active_slabs;
174 unsigned long num_slabs;
175 unsigned long shared_avail;
176 unsigned int limit;
177 unsigned int batchcount;
178 unsigned int shared;
179 unsigned int objects_per_slab;
180 unsigned int cache_order;
181};
182
183void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo);
184void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s);
185ssize_t slabinfo_write(struct file *file, const char __user *buffer,
186 size_t count, loff_t *ppos);
187
188/*
189 * Generic implementation of bulk operations
190 * These are useful for situations in which the allocator cannot
191 * perform optimizations. In that case segments of the object listed
192 * may be allocated or freed using these operations.
193 */
194void __kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
195int __kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
196
197#if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
198/*
199 * Iterate over all memcg caches of the given root cache. The caller must hold
200 * slab_mutex.
201 */
202#define for_each_memcg_cache(iter, root) \
203 list_for_each_entry(iter, &(root)->memcg_params.list, \
204 memcg_params.list)
205
206static inline bool is_root_cache(struct kmem_cache *s)
207{
208 return s->memcg_params.is_root_cache;
209}
210
211static inline bool slab_equal_or_root(struct kmem_cache *s,
212 struct kmem_cache *p)
213{
214 return p == s || p == s->memcg_params.root_cache;
215}
216
217/*
218 * We use suffixes to the name in memcg because we can't have caches
219 * created in the system with the same name. But when we print them
220 * locally, better refer to them with the base name
221 */
222static inline const char *cache_name(struct kmem_cache *s)
223{
224 if (!is_root_cache(s))
225 s = s->memcg_params.root_cache;
226 return s->name;
227}
228
229/*
230 * Note, we protect with RCU only the memcg_caches array, not per-memcg caches.
231 * That said the caller must assure the memcg's cache won't go away by either
232 * taking a css reference to the owner cgroup, or holding the slab_mutex.
233 */
234static inline struct kmem_cache *
235cache_from_memcg_idx(struct kmem_cache *s, int idx)
236{
237 struct kmem_cache *cachep;
238 struct memcg_cache_array *arr;
239
240 rcu_read_lock();
241 arr = rcu_dereference(s->memcg_params.memcg_caches);
242
243 /*
244 * Make sure we will access the up-to-date value. The code updating
245 * memcg_caches issues a write barrier to match this (see
246 * memcg_create_kmem_cache()).
247 */
248 cachep = lockless_dereference(arr->entries[idx]);
249 rcu_read_unlock();
250
251 return cachep;
252}
253
254static inline struct kmem_cache *memcg_root_cache(struct kmem_cache *s)
255{
256 if (is_root_cache(s))
257 return s;
258 return s->memcg_params.root_cache;
259}
260
261static __always_inline int memcg_charge_slab(struct page *page,
262 gfp_t gfp, int order,
263 struct kmem_cache *s)
264{
265 int ret;
266
267 if (!memcg_kmem_enabled())
268 return 0;
269 if (is_root_cache(s))
270 return 0;
271
272 ret = memcg_kmem_charge_memcg(page, gfp, order, s->memcg_params.memcg);
273 if (ret)
274 return ret;
275
276 memcg_kmem_update_page_stat(page,
277 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
278 MEMCG_SLAB_RECLAIMABLE : MEMCG_SLAB_UNRECLAIMABLE,
279 1 << order);
280 return 0;
281}
282
283static __always_inline void memcg_uncharge_slab(struct page *page, int order,
284 struct kmem_cache *s)
285{
286 if (!memcg_kmem_enabled())
287 return;
288
289 memcg_kmem_update_page_stat(page,
290 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
291 MEMCG_SLAB_RECLAIMABLE : MEMCG_SLAB_UNRECLAIMABLE,
292 -(1 << order));
293 memcg_kmem_uncharge(page, order);
294}
295
296extern void slab_init_memcg_params(struct kmem_cache *);
297
298#else /* CONFIG_MEMCG && !CONFIG_SLOB */
299
300#define for_each_memcg_cache(iter, root) \
301 for ((void)(iter), (void)(root); 0; )
302
303static inline bool is_root_cache(struct kmem_cache *s)
304{
305 return true;
306}
307
308static inline bool slab_equal_or_root(struct kmem_cache *s,
309 struct kmem_cache *p)
310{
311 return true;
312}
313
314static inline const char *cache_name(struct kmem_cache *s)
315{
316 return s->name;
317}
318
319static inline struct kmem_cache *
320cache_from_memcg_idx(struct kmem_cache *s, int idx)
321{
322 return NULL;
323}
324
325static inline struct kmem_cache *memcg_root_cache(struct kmem_cache *s)
326{
327 return s;
328}
329
330static inline int memcg_charge_slab(struct page *page, gfp_t gfp, int order,
331 struct kmem_cache *s)
332{
333 return 0;
334}
335
336static inline void memcg_uncharge_slab(struct page *page, int order,
337 struct kmem_cache *s)
338{
339}
340
341static inline void slab_init_memcg_params(struct kmem_cache *s)
342{
343}
344#endif /* CONFIG_MEMCG && !CONFIG_SLOB */
345
346static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x)
347{
348 struct kmem_cache *cachep;
349 struct page *page;
350
351 /*
352 * When kmemcg is not being used, both assignments should return the
353 * same value. but we don't want to pay the assignment price in that
354 * case. If it is not compiled in, the compiler should be smart enough
355 * to not do even the assignment. In that case, slab_equal_or_root
356 * will also be a constant.
357 */
358 if (!memcg_kmem_enabled() &&
359 !unlikely(s->flags & SLAB_CONSISTENCY_CHECKS))
360 return s;
361
362 page = virt_to_head_page(x);
363 cachep = page->slab_cache;
364 if (slab_equal_or_root(cachep, s))
365 return cachep;
366
367 pr_err("%s: Wrong slab cache. %s but object is from %s\n",
368 __func__, s->name, cachep->name);
369 WARN_ON_ONCE(1);
370 return s;
371}
372
373static inline size_t slab_ksize(const struct kmem_cache *s)
374{
375#ifndef CONFIG_SLUB
376 return s->object_size;
377
378#else /* CONFIG_SLUB */
379# ifdef CONFIG_SLUB_DEBUG
380 /*
381 * Debugging requires use of the padding between object
382 * and whatever may come after it.
383 */
384 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
385 return s->object_size;
386# endif
387 if (s->flags & SLAB_KASAN)
388 return s->object_size;
389 /*
390 * If we have the need to store the freelist pointer
391 * back there or track user information then we can
392 * only use the space before that information.
393 */
394 if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER))
395 return s->inuse;
396 /*
397 * Else we can use all the padding etc for the allocation
398 */
399 return s->size;
400#endif
401}
402
403static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s,
404 gfp_t flags)
405{
406 flags &= gfp_allowed_mask;
407 lockdep_trace_alloc(flags);
408 might_sleep_if(gfpflags_allow_blocking(flags));
409
410 if (should_failslab(s, flags))
411 return NULL;
412
413 if (memcg_kmem_enabled() &&
414 ((flags & __GFP_ACCOUNT) || (s->flags & SLAB_ACCOUNT)))
415 return memcg_kmem_get_cache(s);
416
417 return s;
418}
419
420static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags,
421 size_t size, void **p)
422{
423 size_t i;
424
425 flags &= gfp_allowed_mask;
426 for (i = 0; i < size; i++) {
427 void *object = p[i];
428
429 kmemcheck_slab_alloc(s, flags, object, slab_ksize(s));
430 kmemleak_alloc_recursive(object, s->object_size, 1,
431 s->flags, flags);
432 kasan_slab_alloc(s, object, flags);
433 }
434
435 if (memcg_kmem_enabled())
436 memcg_kmem_put_cache(s);
437}
438
439#ifndef CONFIG_SLOB
440/*
441 * The slab lists for all objects.
442 */
443struct kmem_cache_node {
444 spinlock_t list_lock;
445
446#ifdef CONFIG_SLAB
447 struct list_head slabs_partial; /* partial list first, better asm code */
448 struct list_head slabs_full;
449 struct list_head slabs_free;
450 unsigned long total_slabs; /* length of all slab lists */
451 unsigned long free_slabs; /* length of free slab list only */
452 unsigned long free_objects;
453 unsigned int free_limit;
454 unsigned int colour_next; /* Per-node cache coloring */
455 struct array_cache *shared; /* shared per node */
456 struct alien_cache **alien; /* on other nodes */
457 unsigned long next_reap; /* updated without locking */
458 int free_touched; /* updated without locking */
459#endif
460
461#ifdef CONFIG_SLUB
462 unsigned long nr_partial;
463 struct list_head partial;
464#ifdef CONFIG_SLUB_DEBUG
465 atomic_long_t nr_slabs;
466 atomic_long_t total_objects;
467 struct list_head full;
468#endif
469#endif
470
471};
472
473static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
474{
475 return s->node[node];
476}
477
478/*
479 * Iterator over all nodes. The body will be executed for each node that has
480 * a kmem_cache_node structure allocated (which is true for all online nodes)
481 */
482#define for_each_kmem_cache_node(__s, __node, __n) \
483 for (__node = 0; __node < nr_node_ids; __node++) \
484 if ((__n = get_node(__s, __node)))
485
486#endif
487
488void *slab_start(struct seq_file *m, loff_t *pos);
489void *slab_next(struct seq_file *m, void *p, loff_t *pos);
490void slab_stop(struct seq_file *m, void *p);
491int memcg_slab_show(struct seq_file *m, void *p);
492
493void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr);
494
495#ifdef CONFIG_SLAB_FREELIST_RANDOM
496int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
497 gfp_t gfp);
498void cache_random_seq_destroy(struct kmem_cache *cachep);
499#else
500static inline int cache_random_seq_create(struct kmem_cache *cachep,
501 unsigned int count, gfp_t gfp)
502{
503 return 0;
504}
505static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { }
506#endif /* CONFIG_SLAB_FREELIST_RANDOM */
507
508#endif /* MM_SLAB_H */