mm: list_lru: transpose the array of per-node per-memcg lru lists
[linux-block.git] / mm / slab.h
CommitLineData
b2441318 1/* SPDX-License-Identifier: GPL-2.0 */
97d06609
CL
2#ifndef MM_SLAB_H
3#define MM_SLAB_H
4/*
5 * Internal slab definitions
6 */
7
d122019b
MWO
8/* Reuses the bits in struct page */
9struct slab {
10 unsigned long __page_flags;
401fb12c
VB
11
12#if defined(CONFIG_SLAB)
13
d122019b
MWO
14 union {
15 struct list_head slab_list;
401fb12c
VB
16 struct rcu_head rcu_head;
17 };
18 struct kmem_cache *slab_cache;
19 void *freelist; /* array of free object indexes */
20 void *s_mem; /* first object */
21 unsigned int active;
22
23#elif defined(CONFIG_SLUB)
24
25 union {
26 struct list_head slab_list;
27 struct rcu_head rcu_head;
9c01e9af 28#ifdef CONFIG_SLUB_CPU_PARTIAL
401fb12c 29 struct {
d122019b 30 struct slab *next;
d122019b 31 int slabs; /* Nr of slabs left */
d122019b 32 };
9c01e9af 33#endif
d122019b 34 };
401fb12c 35 struct kmem_cache *slab_cache;
d122019b
MWO
36 /* Double-word boundary */
37 void *freelist; /* first free object */
38 union {
401fb12c
VB
39 unsigned long counters;
40 struct {
d122019b
MWO
41 unsigned inuse:16;
42 unsigned objects:15;
43 unsigned frozen:1;
44 };
45 };
401fb12c
VB
46 unsigned int __unused;
47
48#elif defined(CONFIG_SLOB)
49
50 struct list_head slab_list;
51 void *__unused_1;
52 void *freelist; /* first free block */
b01af5c0
HY
53 long units;
54 unsigned int __unused_2;
401fb12c
VB
55
56#else
57#error "Unexpected slab allocator configured"
58#endif
d122019b 59
d122019b
MWO
60 atomic_t __page_refcount;
61#ifdef CONFIG_MEMCG
62 unsigned long memcg_data;
63#endif
64};
65
66#define SLAB_MATCH(pg, sl) \
67 static_assert(offsetof(struct page, pg) == offsetof(struct slab, sl))
68SLAB_MATCH(flags, __page_flags);
69SLAB_MATCH(compound_head, slab_list); /* Ensure bit 0 is clear */
401fb12c 70#ifndef CONFIG_SLOB
d122019b 71SLAB_MATCH(rcu_head, rcu_head);
401fb12c 72#endif
d122019b
MWO
73SLAB_MATCH(_refcount, __page_refcount);
74#ifdef CONFIG_MEMCG
75SLAB_MATCH(memcg_data, memcg_data);
76#endif
77#undef SLAB_MATCH
78static_assert(sizeof(struct slab) <= sizeof(struct page));
79
80/**
81 * folio_slab - Converts from folio to slab.
82 * @folio: The folio.
83 *
84 * Currently struct slab is a different representation of a folio where
85 * folio_test_slab() is true.
86 *
87 * Return: The slab which contains this folio.
88 */
89#define folio_slab(folio) (_Generic((folio), \
90 const struct folio *: (const struct slab *)(folio), \
91 struct folio *: (struct slab *)(folio)))
92
93/**
94 * slab_folio - The folio allocated for a slab
95 * @slab: The slab.
96 *
97 * Slabs are allocated as folios that contain the individual objects and are
98 * using some fields in the first struct page of the folio - those fields are
99 * now accessed by struct slab. It is occasionally necessary to convert back to
100 * a folio in order to communicate with the rest of the mm. Please use this
101 * helper function instead of casting yourself, as the implementation may change
102 * in the future.
103 */
104#define slab_folio(s) (_Generic((s), \
105 const struct slab *: (const struct folio *)s, \
106 struct slab *: (struct folio *)s))
107
108/**
109 * page_slab - Converts from first struct page to slab.
110 * @p: The first (either head of compound or single) page of slab.
111 *
112 * A temporary wrapper to convert struct page to struct slab in situations where
113 * we know the page is the compound head, or single order-0 page.
114 *
115 * Long-term ideally everything would work with struct slab directly or go
116 * through folio to struct slab.
117 *
118 * Return: The slab which contains this page
119 */
120#define page_slab(p) (_Generic((p), \
121 const struct page *: (const struct slab *)(p), \
122 struct page *: (struct slab *)(p)))
123
124/**
125 * slab_page - The first struct page allocated for a slab
126 * @slab: The slab.
127 *
128 * A convenience wrapper for converting slab to the first struct page of the
129 * underlying folio, to communicate with code not yet converted to folio or
130 * struct slab.
131 */
132#define slab_page(s) folio_page(slab_folio(s), 0)
133
134/*
135 * If network-based swap is enabled, sl*b must keep track of whether pages
136 * were allocated from pfmemalloc reserves.
137 */
138static inline bool slab_test_pfmemalloc(const struct slab *slab)
139{
140 return folio_test_active((struct folio *)slab_folio(slab));
141}
142
143static inline void slab_set_pfmemalloc(struct slab *slab)
144{
145 folio_set_active(slab_folio(slab));
146}
147
148static inline void slab_clear_pfmemalloc(struct slab *slab)
149{
150 folio_clear_active(slab_folio(slab));
151}
152
153static inline void __slab_clear_pfmemalloc(struct slab *slab)
154{
155 __folio_clear_active(slab_folio(slab));
156}
157
158static inline void *slab_address(const struct slab *slab)
159{
160 return folio_address(slab_folio(slab));
161}
162
163static inline int slab_nid(const struct slab *slab)
164{
165 return folio_nid(slab_folio(slab));
166}
167
168static inline pg_data_t *slab_pgdat(const struct slab *slab)
169{
170 return folio_pgdat(slab_folio(slab));
171}
172
173static inline struct slab *virt_to_slab(const void *addr)
174{
175 struct folio *folio = virt_to_folio(addr);
176
177 if (!folio_test_slab(folio))
178 return NULL;
179
180 return folio_slab(folio);
181}
182
183static inline int slab_order(const struct slab *slab)
184{
185 return folio_order((struct folio *)slab_folio(slab));
186}
187
188static inline size_t slab_size(const struct slab *slab)
189{
190 return PAGE_SIZE << slab_order(slab);
191}
192
07f361b2
JK
193#ifdef CONFIG_SLOB
194/*
195 * Common fields provided in kmem_cache by all slab allocators
196 * This struct is either used directly by the allocator (SLOB)
197 * or the allocator must include definitions for all fields
198 * provided in kmem_cache_common in their definition of kmem_cache.
199 *
200 * Once we can do anonymous structs (C11 standard) we could put a
201 * anonymous struct definition in these allocators so that the
202 * separate allocations in the kmem_cache structure of SLAB and
203 * SLUB is no longer needed.
204 */
205struct kmem_cache {
206 unsigned int object_size;/* The original size of the object */
207 unsigned int size; /* The aligned/padded/added on size */
208 unsigned int align; /* Alignment as calculated */
d50112ed 209 slab_flags_t flags; /* Active flags on the slab */
7bbdb81e
AD
210 unsigned int useroffset;/* Usercopy region offset */
211 unsigned int usersize; /* Usercopy region size */
07f361b2
JK
212 const char *name; /* Slab name for sysfs */
213 int refcount; /* Use counter */
214 void (*ctor)(void *); /* Called on object slot creation */
215 struct list_head list; /* List of all slab caches on the system */
216};
217
218#endif /* CONFIG_SLOB */
219
220#ifdef CONFIG_SLAB
221#include <linux/slab_def.h>
222#endif
223
224#ifdef CONFIG_SLUB
225#include <linux/slub_def.h>
226#endif
227
228#include <linux/memcontrol.h>
11c7aec2 229#include <linux/fault-inject.h>
11c7aec2
JDB
230#include <linux/kasan.h>
231#include <linux/kmemleak.h>
7c00fce9 232#include <linux/random.h>
d92a8cfc 233#include <linux/sched/mm.h>
07f361b2 234
97d06609
CL
235/*
236 * State of the slab allocator.
237 *
238 * This is used to describe the states of the allocator during bootup.
239 * Allocators use this to gradually bootstrap themselves. Most allocators
240 * have the problem that the structures used for managing slab caches are
241 * allocated from slab caches themselves.
242 */
243enum slab_state {
244 DOWN, /* No slab functionality yet */
245 PARTIAL, /* SLUB: kmem_cache_node available */
ce8eb6c4 246 PARTIAL_NODE, /* SLAB: kmalloc size for node struct available */
97d06609
CL
247 UP, /* Slab caches usable but not all extras yet */
248 FULL /* Everything is working */
249};
250
251extern enum slab_state slab_state;
252
18004c5d
CL
253/* The slab cache mutex protects the management structures during changes */
254extern struct mutex slab_mutex;
9b030cb8
CL
255
256/* The list of all slab caches on the system */
18004c5d
CL
257extern struct list_head slab_caches;
258
9b030cb8
CL
259/* The slab cache that manages slab cache information */
260extern struct kmem_cache *kmem_cache;
261
af3b5f87
VB
262/* A table of kmalloc cache names and sizes */
263extern const struct kmalloc_info_struct {
cb5d9fb3 264 const char *name[NR_KMALLOC_TYPES];
55de8b9c 265 unsigned int size;
af3b5f87
VB
266} kmalloc_info[];
267
f97d5f63
CL
268#ifndef CONFIG_SLOB
269/* Kmalloc array related functions */
34cc6990 270void setup_kmalloc_cache_index_table(void);
d50112ed 271void create_kmalloc_caches(slab_flags_t);
2c59dd65
CL
272
273/* Find the kmalloc slab corresponding for a certain size */
274struct kmem_cache *kmalloc_slab(size_t, gfp_t);
f97d5f63
CL
275#endif
276
44405099 277gfp_t kmalloc_fix_flags(gfp_t flags);
f97d5f63 278
9b030cb8 279/* Functions provided by the slab allocators */
d50112ed 280int __kmem_cache_create(struct kmem_cache *, slab_flags_t flags);
97d06609 281
55de8b9c
AD
282struct kmem_cache *create_kmalloc_cache(const char *name, unsigned int size,
283 slab_flags_t flags, unsigned int useroffset,
284 unsigned int usersize);
45530c44 285extern void create_boot_cache(struct kmem_cache *, const char *name,
361d575e
AD
286 unsigned int size, slab_flags_t flags,
287 unsigned int useroffset, unsigned int usersize);
45530c44 288
423c929c 289int slab_unmergeable(struct kmem_cache *s);
f4957d5b 290struct kmem_cache *find_mergeable(unsigned size, unsigned align,
d50112ed 291 slab_flags_t flags, const char *name, void (*ctor)(void *));
12220dea 292#ifndef CONFIG_SLOB
2633d7a0 293struct kmem_cache *
f4957d5b 294__kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
d50112ed 295 slab_flags_t flags, void (*ctor)(void *));
423c929c 296
0293d1fd 297slab_flags_t kmem_cache_flags(unsigned int object_size,
37540008 298 slab_flags_t flags, const char *name);
cbb79694 299#else
2633d7a0 300static inline struct kmem_cache *
f4957d5b 301__kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
d50112ed 302 slab_flags_t flags, void (*ctor)(void *))
cbb79694 303{ return NULL; }
423c929c 304
0293d1fd 305static inline slab_flags_t kmem_cache_flags(unsigned int object_size,
37540008 306 slab_flags_t flags, const char *name)
423c929c
JK
307{
308 return flags;
309}
cbb79694
CL
310#endif
311
312
d8843922 313/* Legal flag mask for kmem_cache_create(), for various configurations */
6d6ea1e9
NB
314#define SLAB_CORE_FLAGS (SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA | \
315 SLAB_CACHE_DMA32 | SLAB_PANIC | \
5f0d5a3a 316 SLAB_TYPESAFE_BY_RCU | SLAB_DEBUG_OBJECTS )
d8843922
GC
317
318#if defined(CONFIG_DEBUG_SLAB)
319#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
320#elif defined(CONFIG_SLUB_DEBUG)
321#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
becfda68 322 SLAB_TRACE | SLAB_CONSISTENCY_CHECKS)
d8843922
GC
323#else
324#define SLAB_DEBUG_FLAGS (0)
325#endif
326
327#if defined(CONFIG_SLAB)
328#define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \
230e9fc2 329 SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \
75f296d9 330 SLAB_ACCOUNT)
d8843922
GC
331#elif defined(CONFIG_SLUB)
332#define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
75f296d9 333 SLAB_TEMPORARY | SLAB_ACCOUNT)
d8843922 334#else
34dbc3aa 335#define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE)
d8843922
GC
336#endif
337
e70954fd 338/* Common flags available with current configuration */
d8843922
GC
339#define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
340
e70954fd
TG
341/* Common flags permitted for kmem_cache_create */
342#define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \
343 SLAB_RED_ZONE | \
344 SLAB_POISON | \
345 SLAB_STORE_USER | \
346 SLAB_TRACE | \
347 SLAB_CONSISTENCY_CHECKS | \
348 SLAB_MEM_SPREAD | \
349 SLAB_NOLEAKTRACE | \
350 SLAB_RECLAIM_ACCOUNT | \
351 SLAB_TEMPORARY | \
e70954fd
TG
352 SLAB_ACCOUNT)
353
f9e13c0a 354bool __kmem_cache_empty(struct kmem_cache *);
945cf2b6 355int __kmem_cache_shutdown(struct kmem_cache *);
52b4b950 356void __kmem_cache_release(struct kmem_cache *);
c9fc5864 357int __kmem_cache_shrink(struct kmem_cache *);
41a21285 358void slab_kmem_cache_release(struct kmem_cache *);
945cf2b6 359
b7454ad3
GC
360struct seq_file;
361struct file;
b7454ad3 362
0d7561c6
GC
363struct slabinfo {
364 unsigned long active_objs;
365 unsigned long num_objs;
366 unsigned long active_slabs;
367 unsigned long num_slabs;
368 unsigned long shared_avail;
369 unsigned int limit;
370 unsigned int batchcount;
371 unsigned int shared;
372 unsigned int objects_per_slab;
373 unsigned int cache_order;
374};
375
376void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo);
377void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s);
b7454ad3
GC
378ssize_t slabinfo_write(struct file *file, const char __user *buffer,
379 size_t count, loff_t *ppos);
ba6c496e 380
484748f0
CL
381/*
382 * Generic implementation of bulk operations
383 * These are useful for situations in which the allocator cannot
9f706d68 384 * perform optimizations. In that case segments of the object listed
484748f0
CL
385 * may be allocated or freed using these operations.
386 */
387void __kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
865762a8 388int __kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
484748f0 389
1a984c4e 390static inline enum node_stat_item cache_vmstat_idx(struct kmem_cache *s)
6cea1d56
RG
391{
392 return (s->flags & SLAB_RECLAIM_ACCOUNT) ?
d42f3245 393 NR_SLAB_RECLAIMABLE_B : NR_SLAB_UNRECLAIMABLE_B;
6cea1d56
RG
394}
395
e42f174e
VB
396#ifdef CONFIG_SLUB_DEBUG
397#ifdef CONFIG_SLUB_DEBUG_ON
398DECLARE_STATIC_KEY_TRUE(slub_debug_enabled);
399#else
400DECLARE_STATIC_KEY_FALSE(slub_debug_enabled);
401#endif
402extern void print_tracking(struct kmem_cache *s, void *object);
1f9f78b1 403long validate_slab_cache(struct kmem_cache *s);
0d4a062a
ME
404static inline bool __slub_debug_enabled(void)
405{
406 return static_branch_unlikely(&slub_debug_enabled);
407}
e42f174e
VB
408#else
409static inline void print_tracking(struct kmem_cache *s, void *object)
410{
411}
0d4a062a
ME
412static inline bool __slub_debug_enabled(void)
413{
414 return false;
415}
e42f174e
VB
416#endif
417
418/*
419 * Returns true if any of the specified slub_debug flags is enabled for the
420 * cache. Use only for flags parsed by setup_slub_debug() as it also enables
421 * the static key.
422 */
423static inline bool kmem_cache_debug_flags(struct kmem_cache *s, slab_flags_t flags)
424{
0d4a062a
ME
425 if (IS_ENABLED(CONFIG_SLUB_DEBUG))
426 VM_WARN_ON_ONCE(!(flags & SLAB_DEBUG_FLAGS));
427 if (__slub_debug_enabled())
e42f174e 428 return s->flags & flags;
e42f174e
VB
429 return false;
430}
431
84c07d11 432#ifdef CONFIG_MEMCG_KMEM
4b5f8d9a
VB
433/*
434 * slab_objcgs - get the object cgroups vector associated with a slab
435 * @slab: a pointer to the slab struct
436 *
437 * Returns a pointer to the object cgroups vector associated with the slab,
438 * or NULL if no such vector has been associated yet.
439 */
440static inline struct obj_cgroup **slab_objcgs(struct slab *slab)
441{
442 unsigned long memcg_data = READ_ONCE(slab->memcg_data);
443
444 VM_BUG_ON_PAGE(memcg_data && !(memcg_data & MEMCG_DATA_OBJCGS),
445 slab_page(slab));
446 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_KMEM, slab_page(slab));
447
448 return (struct obj_cgroup **)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
449}
450
451int memcg_alloc_slab_cgroups(struct slab *slab, struct kmem_cache *s,
452 gfp_t gfp, bool new_slab);
fdbcb2a6
WL
453void mod_objcg_state(struct obj_cgroup *objcg, struct pglist_data *pgdat,
454 enum node_stat_item idx, int nr);
286e04b8 455
4b5f8d9a 456static inline void memcg_free_slab_cgroups(struct slab *slab)
286e04b8 457{
4b5f8d9a
VB
458 kfree(slab_objcgs(slab));
459 slab->memcg_data = 0;
286e04b8
RG
460}
461
f2fe7b09
RG
462static inline size_t obj_full_size(struct kmem_cache *s)
463{
464 /*
465 * For each accounted object there is an extra space which is used
466 * to store obj_cgroup membership. Charge it too.
467 */
468 return s->size + sizeof(struct obj_cgroup *);
469}
470
becaba65
RG
471/*
472 * Returns false if the allocation should fail.
473 */
474static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s,
475 struct obj_cgroup **objcgp,
476 size_t objects, gfp_t flags)
f2fe7b09 477{
9855609b
RG
478 struct obj_cgroup *objcg;
479
becaba65
RG
480 if (!memcg_kmem_enabled())
481 return true;
482
483 if (!(flags & __GFP_ACCOUNT) && !(s->flags & SLAB_ACCOUNT))
484 return true;
485
9855609b
RG
486 objcg = get_obj_cgroup_from_current();
487 if (!objcg)
becaba65 488 return true;
9855609b
RG
489
490 if (obj_cgroup_charge(objcg, flags, objects * obj_full_size(s))) {
491 obj_cgroup_put(objcg);
becaba65 492 return false;
f2fe7b09
RG
493 }
494
becaba65
RG
495 *objcgp = objcg;
496 return true;
f2fe7b09
RG
497}
498
964d4bd3
RG
499static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
500 struct obj_cgroup *objcg,
10befea9
RG
501 gfp_t flags, size_t size,
502 void **p)
964d4bd3 503{
4b5f8d9a 504 struct slab *slab;
964d4bd3
RG
505 unsigned long off;
506 size_t i;
507
becaba65 508 if (!memcg_kmem_enabled() || !objcg)
10befea9
RG
509 return;
510
964d4bd3
RG
511 for (i = 0; i < size; i++) {
512 if (likely(p[i])) {
4b5f8d9a 513 slab = virt_to_slab(p[i]);
10befea9 514
4b5f8d9a
VB
515 if (!slab_objcgs(slab) &&
516 memcg_alloc_slab_cgroups(slab, s, flags,
2e9bd483 517 false)) {
10befea9
RG
518 obj_cgroup_uncharge(objcg, obj_full_size(s));
519 continue;
520 }
521
4b5f8d9a 522 off = obj_to_index(s, slab, p[i]);
964d4bd3 523 obj_cgroup_get(objcg);
4b5f8d9a
VB
524 slab_objcgs(slab)[off] = objcg;
525 mod_objcg_state(objcg, slab_pgdat(slab),
f2fe7b09
RG
526 cache_vmstat_idx(s), obj_full_size(s));
527 } else {
528 obj_cgroup_uncharge(objcg, obj_full_size(s));
964d4bd3
RG
529 }
530 }
531 obj_cgroup_put(objcg);
964d4bd3
RG
532}
533
d1b2cf6c
BR
534static inline void memcg_slab_free_hook(struct kmem_cache *s_orig,
535 void **p, int objects)
964d4bd3 536{
d1b2cf6c 537 struct kmem_cache *s;
270c6a71 538 struct obj_cgroup **objcgs;
964d4bd3 539 struct obj_cgroup *objcg;
4b5f8d9a 540 struct slab *slab;
964d4bd3 541 unsigned int off;
d1b2cf6c 542 int i;
964d4bd3 543
10befea9
RG
544 if (!memcg_kmem_enabled())
545 return;
546
d1b2cf6c
BR
547 for (i = 0; i < objects; i++) {
548 if (unlikely(!p[i]))
549 continue;
964d4bd3 550
4b5f8d9a
VB
551 slab = virt_to_slab(p[i]);
552 /* we could be given a kmalloc_large() object, skip those */
553 if (!slab)
554 continue;
555
556 objcgs = slab_objcgs(slab);
270c6a71 557 if (!objcgs)
d1b2cf6c 558 continue;
f2fe7b09 559
d1b2cf6c 560 if (!s_orig)
4b5f8d9a 561 s = slab->slab_cache;
d1b2cf6c
BR
562 else
563 s = s_orig;
10befea9 564
4b5f8d9a 565 off = obj_to_index(s, slab, p[i]);
270c6a71 566 objcg = objcgs[off];
d1b2cf6c
BR
567 if (!objcg)
568 continue;
f2fe7b09 569
270c6a71 570 objcgs[off] = NULL;
d1b2cf6c 571 obj_cgroup_uncharge(objcg, obj_full_size(s));
4b5f8d9a 572 mod_objcg_state(objcg, slab_pgdat(slab), cache_vmstat_idx(s),
d1b2cf6c
BR
573 -obj_full_size(s));
574 obj_cgroup_put(objcg);
575 }
964d4bd3
RG
576}
577
84c07d11 578#else /* CONFIG_MEMCG_KMEM */
4b5f8d9a
VB
579static inline struct obj_cgroup **slab_objcgs(struct slab *slab)
580{
581 return NULL;
582}
583
9855609b 584static inline struct mem_cgroup *memcg_from_slab_obj(void *ptr)
4d96ba35
RG
585{
586 return NULL;
587}
588
4b5f8d9a 589static inline int memcg_alloc_slab_cgroups(struct slab *slab,
2e9bd483 590 struct kmem_cache *s, gfp_t gfp,
4b5f8d9a 591 bool new_slab)
286e04b8
RG
592{
593 return 0;
594}
595
4b5f8d9a 596static inline void memcg_free_slab_cgroups(struct slab *slab)
286e04b8
RG
597{
598}
599
becaba65
RG
600static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s,
601 struct obj_cgroup **objcgp,
602 size_t objects, gfp_t flags)
f2fe7b09 603{
becaba65 604 return true;
f2fe7b09
RG
605}
606
964d4bd3
RG
607static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
608 struct obj_cgroup *objcg,
10befea9
RG
609 gfp_t flags, size_t size,
610 void **p)
964d4bd3
RG
611{
612}
613
d1b2cf6c
BR
614static inline void memcg_slab_free_hook(struct kmem_cache *s,
615 void **p, int objects)
964d4bd3
RG
616{
617}
84c07d11 618#endif /* CONFIG_MEMCG_KMEM */
b9ce5ef4 619
401fb12c 620#ifndef CONFIG_SLOB
a64b5378
KC
621static inline struct kmem_cache *virt_to_cache(const void *obj)
622{
82c1775d 623 struct slab *slab;
a64b5378 624
82c1775d
MWO
625 slab = virt_to_slab(obj);
626 if (WARN_ONCE(!slab, "%s: Object is not a Slab page!\n",
a64b5378
KC
627 __func__))
628 return NULL;
82c1775d 629 return slab->slab_cache;
a64b5378
KC
630}
631
b918653b
MWO
632static __always_inline void account_slab(struct slab *slab, int order,
633 struct kmem_cache *s, gfp_t gfp)
6cea1d56 634{
2e9bd483 635 if (memcg_kmem_enabled() && (s->flags & SLAB_ACCOUNT))
4b5f8d9a 636 memcg_alloc_slab_cgroups(slab, s, gfp, true);
2e9bd483 637
b918653b 638 mod_node_page_state(slab_pgdat(slab), cache_vmstat_idx(s),
f2fe7b09 639 PAGE_SIZE << order);
6cea1d56
RG
640}
641
b918653b
MWO
642static __always_inline void unaccount_slab(struct slab *slab, int order,
643 struct kmem_cache *s)
6cea1d56 644{
10befea9 645 if (memcg_kmem_enabled())
4b5f8d9a 646 memcg_free_slab_cgroups(slab);
9855609b 647
b918653b 648 mod_node_page_state(slab_pgdat(slab), cache_vmstat_idx(s),
f2fe7b09 649 -(PAGE_SIZE << order));
6cea1d56
RG
650}
651
e42f174e
VB
652static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x)
653{
654 struct kmem_cache *cachep;
655
656 if (!IS_ENABLED(CONFIG_SLAB_FREELIST_HARDENED) &&
e42f174e
VB
657 !kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS))
658 return s;
659
660 cachep = virt_to_cache(x);
10befea9 661 if (WARN(cachep && cachep != s,
e42f174e
VB
662 "%s: Wrong slab cache. %s but object is from %s\n",
663 __func__, s->name, cachep->name))
664 print_tracking(cachep, x);
665 return cachep;
666}
401fb12c 667#endif /* CONFIG_SLOB */
e42f174e 668
11c7aec2
JDB
669static inline size_t slab_ksize(const struct kmem_cache *s)
670{
671#ifndef CONFIG_SLUB
672 return s->object_size;
673
674#else /* CONFIG_SLUB */
675# ifdef CONFIG_SLUB_DEBUG
676 /*
677 * Debugging requires use of the padding between object
678 * and whatever may come after it.
679 */
680 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
681 return s->object_size;
682# endif
80a9201a
AP
683 if (s->flags & SLAB_KASAN)
684 return s->object_size;
11c7aec2
JDB
685 /*
686 * If we have the need to store the freelist pointer
687 * back there or track user information then we can
688 * only use the space before that information.
689 */
5f0d5a3a 690 if (s->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_STORE_USER))
11c7aec2
JDB
691 return s->inuse;
692 /*
693 * Else we can use all the padding etc for the allocation
694 */
695 return s->size;
696#endif
697}
698
699static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s,
964d4bd3
RG
700 struct obj_cgroup **objcgp,
701 size_t size, gfp_t flags)
11c7aec2
JDB
702{
703 flags &= gfp_allowed_mask;
d92a8cfc 704
95d6c701 705 might_alloc(flags);
11c7aec2 706
fab9963a 707 if (should_failslab(s, flags))
11c7aec2
JDB
708 return NULL;
709
becaba65
RG
710 if (!memcg_slab_pre_alloc_hook(s, objcgp, size, flags))
711 return NULL;
45264778
VD
712
713 return s;
11c7aec2
JDB
714}
715
964d4bd3 716static inline void slab_post_alloc_hook(struct kmem_cache *s,
da844b78
AK
717 struct obj_cgroup *objcg, gfp_t flags,
718 size_t size, void **p, bool init)
11c7aec2
JDB
719{
720 size_t i;
721
722 flags &= gfp_allowed_mask;
da844b78
AK
723
724 /*
725 * As memory initialization might be integrated into KASAN,
726 * kasan_slab_alloc and initialization memset must be
727 * kept together to avoid discrepancies in behavior.
728 *
729 * As p[i] might get tagged, memset and kmemleak hook come after KASAN.
730 */
11c7aec2 731 for (i = 0; i < size; i++) {
da844b78
AK
732 p[i] = kasan_slab_alloc(s, p[i], flags, init);
733 if (p[i] && init && !kasan_has_integrated_init())
734 memset(p[i], 0, s->object_size);
53128245 735 kmemleak_alloc_recursive(p[i], s->object_size, 1,
11c7aec2 736 s->flags, flags);
11c7aec2 737 }
45264778 738
becaba65 739 memcg_slab_post_alloc_hook(s, objcg, flags, size, p);
11c7aec2
JDB
740}
741
44c5356f 742#ifndef CONFIG_SLOB
ca34956b
CL
743/*
744 * The slab lists for all objects.
745 */
746struct kmem_cache_node {
747 spinlock_t list_lock;
748
749#ifdef CONFIG_SLAB
750 struct list_head slabs_partial; /* partial list first, better asm code */
751 struct list_head slabs_full;
752 struct list_head slabs_free;
bf00bd34
DR
753 unsigned long total_slabs; /* length of all slab lists */
754 unsigned long free_slabs; /* length of free slab list only */
ca34956b
CL
755 unsigned long free_objects;
756 unsigned int free_limit;
757 unsigned int colour_next; /* Per-node cache coloring */
758 struct array_cache *shared; /* shared per node */
c8522a3a 759 struct alien_cache **alien; /* on other nodes */
ca34956b
CL
760 unsigned long next_reap; /* updated without locking */
761 int free_touched; /* updated without locking */
762#endif
763
764#ifdef CONFIG_SLUB
765 unsigned long nr_partial;
766 struct list_head partial;
767#ifdef CONFIG_SLUB_DEBUG
768 atomic_long_t nr_slabs;
769 atomic_long_t total_objects;
770 struct list_head full;
771#endif
772#endif
773
774};
e25839f6 775
44c5356f
CL
776static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
777{
778 return s->node[node];
779}
780
781/*
782 * Iterator over all nodes. The body will be executed for each node that has
783 * a kmem_cache_node structure allocated (which is true for all online nodes)
784 */
785#define for_each_kmem_cache_node(__s, __node, __n) \
9163582c
MP
786 for (__node = 0; __node < nr_node_ids; __node++) \
787 if ((__n = get_node(__s, __node)))
44c5356f
CL
788
789#endif
790
852d8be0
YS
791#if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
792void dump_unreclaimable_slab(void);
793#else
794static inline void dump_unreclaimable_slab(void)
795{
796}
797#endif
798
55834c59
AP
799void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr);
800
7c00fce9
TG
801#ifdef CONFIG_SLAB_FREELIST_RANDOM
802int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
803 gfp_t gfp);
804void cache_random_seq_destroy(struct kmem_cache *cachep);
805#else
806static inline int cache_random_seq_create(struct kmem_cache *cachep,
807 unsigned int count, gfp_t gfp)
808{
809 return 0;
810}
811static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { }
812#endif /* CONFIG_SLAB_FREELIST_RANDOM */
813
6471384a
AP
814static inline bool slab_want_init_on_alloc(gfp_t flags, struct kmem_cache *c)
815{
51cba1eb
KC
816 if (static_branch_maybe(CONFIG_INIT_ON_ALLOC_DEFAULT_ON,
817 &init_on_alloc)) {
6471384a
AP
818 if (c->ctor)
819 return false;
820 if (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON))
821 return flags & __GFP_ZERO;
822 return true;
823 }
824 return flags & __GFP_ZERO;
825}
826
827static inline bool slab_want_init_on_free(struct kmem_cache *c)
828{
51cba1eb
KC
829 if (static_branch_maybe(CONFIG_INIT_ON_FREE_DEFAULT_ON,
830 &init_on_free))
6471384a
AP
831 return !(c->ctor ||
832 (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)));
833 return false;
834}
835
64dd6849
FM
836#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG)
837void debugfs_slab_release(struct kmem_cache *);
838#else
839static inline void debugfs_slab_release(struct kmem_cache *s) { }
840#endif
841
5bb1bb35 842#ifdef CONFIG_PRINTK
8e7f37f2
PM
843#define KS_ADDRS_COUNT 16
844struct kmem_obj_info {
845 void *kp_ptr;
7213230a 846 struct slab *kp_slab;
8e7f37f2
PM
847 void *kp_objp;
848 unsigned long kp_data_offset;
849 struct kmem_cache *kp_slab_cache;
850 void *kp_ret;
851 void *kp_stack[KS_ADDRS_COUNT];
e548eaa1 852 void *kp_free_stack[KS_ADDRS_COUNT];
8e7f37f2 853};
7213230a 854void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab);
5bb1bb35 855#endif
8e7f37f2 856
0b3eb091
MWO
857#ifdef CONFIG_HAVE_HARDENED_USERCOPY_ALLOCATOR
858void __check_heap_object(const void *ptr, unsigned long n,
859 const struct slab *slab, bool to_user);
860#else
861static inline
862void __check_heap_object(const void *ptr, unsigned long n,
863 const struct slab *slab, bool to_user)
864{
865}
866#endif
867
5240ab40 868#endif /* MM_SLAB_H */