mm/slab: kmalloc: pass requests larger than order-1 page to page allocator
[linux-2.6-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>
88f2ef73 234#include <linux/list_lru.h>
07f361b2 235
97d06609
CL
236/*
237 * State of the slab allocator.
238 *
239 * This is used to describe the states of the allocator during bootup.
240 * Allocators use this to gradually bootstrap themselves. Most allocators
241 * have the problem that the structures used for managing slab caches are
242 * allocated from slab caches themselves.
243 */
244enum slab_state {
245 DOWN, /* No slab functionality yet */
246 PARTIAL, /* SLUB: kmem_cache_node available */
ce8eb6c4 247 PARTIAL_NODE, /* SLAB: kmalloc size for node struct available */
97d06609
CL
248 UP, /* Slab caches usable but not all extras yet */
249 FULL /* Everything is working */
250};
251
252extern enum slab_state slab_state;
253
18004c5d
CL
254/* The slab cache mutex protects the management structures during changes */
255extern struct mutex slab_mutex;
9b030cb8
CL
256
257/* The list of all slab caches on the system */
18004c5d
CL
258extern struct list_head slab_caches;
259
9b030cb8
CL
260/* The slab cache that manages slab cache information */
261extern struct kmem_cache *kmem_cache;
262
af3b5f87
VB
263/* A table of kmalloc cache names and sizes */
264extern const struct kmalloc_info_struct {
cb5d9fb3 265 const char *name[NR_KMALLOC_TYPES];
55de8b9c 266 unsigned int size;
af3b5f87
VB
267} kmalloc_info[];
268
f97d5f63
CL
269#ifndef CONFIG_SLOB
270/* Kmalloc array related functions */
34cc6990 271void setup_kmalloc_cache_index_table(void);
d50112ed 272void create_kmalloc_caches(slab_flags_t);
2c59dd65
CL
273
274/* Find the kmalloc slab corresponding for a certain size */
275struct kmem_cache *kmalloc_slab(size_t, gfp_t);
f97d5f63
CL
276#endif
277
bf37d791
HY
278void *kmalloc_large_node_notrace(size_t size, gfp_t flags, int node);
279
44405099 280gfp_t kmalloc_fix_flags(gfp_t flags);
f97d5f63 281
9b030cb8 282/* Functions provided by the slab allocators */
d50112ed 283int __kmem_cache_create(struct kmem_cache *, slab_flags_t flags);
97d06609 284
55de8b9c
AD
285struct kmem_cache *create_kmalloc_cache(const char *name, unsigned int size,
286 slab_flags_t flags, unsigned int useroffset,
287 unsigned int usersize);
45530c44 288extern void create_boot_cache(struct kmem_cache *, const char *name,
361d575e
AD
289 unsigned int size, slab_flags_t flags,
290 unsigned int useroffset, unsigned int usersize);
45530c44 291
423c929c 292int slab_unmergeable(struct kmem_cache *s);
f4957d5b 293struct kmem_cache *find_mergeable(unsigned size, unsigned align,
d50112ed 294 slab_flags_t flags, const char *name, void (*ctor)(void *));
12220dea 295#ifndef CONFIG_SLOB
2633d7a0 296struct kmem_cache *
f4957d5b 297__kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
d50112ed 298 slab_flags_t flags, void (*ctor)(void *));
423c929c 299
0293d1fd 300slab_flags_t kmem_cache_flags(unsigned int object_size,
37540008 301 slab_flags_t flags, const char *name);
cbb79694 302#else
2633d7a0 303static inline struct kmem_cache *
f4957d5b 304__kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
d50112ed 305 slab_flags_t flags, void (*ctor)(void *))
cbb79694 306{ return NULL; }
423c929c 307
0293d1fd 308static inline slab_flags_t kmem_cache_flags(unsigned int object_size,
37540008 309 slab_flags_t flags, const char *name)
423c929c
JK
310{
311 return flags;
312}
cbb79694
CL
313#endif
314
315
d8843922 316/* Legal flag mask for kmem_cache_create(), for various configurations */
6d6ea1e9
NB
317#define SLAB_CORE_FLAGS (SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA | \
318 SLAB_CACHE_DMA32 | SLAB_PANIC | \
5f0d5a3a 319 SLAB_TYPESAFE_BY_RCU | SLAB_DEBUG_OBJECTS )
d8843922
GC
320
321#if defined(CONFIG_DEBUG_SLAB)
322#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
323#elif defined(CONFIG_SLUB_DEBUG)
324#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
becfda68 325 SLAB_TRACE | SLAB_CONSISTENCY_CHECKS)
d8843922
GC
326#else
327#define SLAB_DEBUG_FLAGS (0)
328#endif
329
330#if defined(CONFIG_SLAB)
331#define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \
230e9fc2 332 SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \
75f296d9 333 SLAB_ACCOUNT)
d8843922
GC
334#elif defined(CONFIG_SLUB)
335#define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
a285909f 336 SLAB_TEMPORARY | SLAB_ACCOUNT | SLAB_NO_USER_FLAGS)
d8843922 337#else
34dbc3aa 338#define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE)
d8843922
GC
339#endif
340
e70954fd 341/* Common flags available with current configuration */
d8843922
GC
342#define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
343
e70954fd
TG
344/* Common flags permitted for kmem_cache_create */
345#define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \
346 SLAB_RED_ZONE | \
347 SLAB_POISON | \
348 SLAB_STORE_USER | \
349 SLAB_TRACE | \
350 SLAB_CONSISTENCY_CHECKS | \
351 SLAB_MEM_SPREAD | \
352 SLAB_NOLEAKTRACE | \
353 SLAB_RECLAIM_ACCOUNT | \
354 SLAB_TEMPORARY | \
a285909f
HY
355 SLAB_ACCOUNT | \
356 SLAB_NO_USER_FLAGS)
e70954fd 357
f9e13c0a 358bool __kmem_cache_empty(struct kmem_cache *);
945cf2b6 359int __kmem_cache_shutdown(struct kmem_cache *);
52b4b950 360void __kmem_cache_release(struct kmem_cache *);
c9fc5864 361int __kmem_cache_shrink(struct kmem_cache *);
41a21285 362void slab_kmem_cache_release(struct kmem_cache *);
945cf2b6 363
b7454ad3
GC
364struct seq_file;
365struct file;
b7454ad3 366
0d7561c6
GC
367struct slabinfo {
368 unsigned long active_objs;
369 unsigned long num_objs;
370 unsigned long active_slabs;
371 unsigned long num_slabs;
372 unsigned long shared_avail;
373 unsigned int limit;
374 unsigned int batchcount;
375 unsigned int shared;
376 unsigned int objects_per_slab;
377 unsigned int cache_order;
378};
379
380void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo);
381void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s);
b7454ad3
GC
382ssize_t slabinfo_write(struct file *file, const char __user *buffer,
383 size_t count, loff_t *ppos);
ba6c496e 384
1a984c4e 385static inline enum node_stat_item cache_vmstat_idx(struct kmem_cache *s)
6cea1d56
RG
386{
387 return (s->flags & SLAB_RECLAIM_ACCOUNT) ?
d42f3245 388 NR_SLAB_RECLAIMABLE_B : NR_SLAB_UNRECLAIMABLE_B;
6cea1d56
RG
389}
390
e42f174e
VB
391#ifdef CONFIG_SLUB_DEBUG
392#ifdef CONFIG_SLUB_DEBUG_ON
393DECLARE_STATIC_KEY_TRUE(slub_debug_enabled);
394#else
395DECLARE_STATIC_KEY_FALSE(slub_debug_enabled);
396#endif
397extern void print_tracking(struct kmem_cache *s, void *object);
1f9f78b1 398long validate_slab_cache(struct kmem_cache *s);
0d4a062a
ME
399static inline bool __slub_debug_enabled(void)
400{
401 return static_branch_unlikely(&slub_debug_enabled);
402}
e42f174e
VB
403#else
404static inline void print_tracking(struct kmem_cache *s, void *object)
405{
406}
0d4a062a
ME
407static inline bool __slub_debug_enabled(void)
408{
409 return false;
410}
e42f174e
VB
411#endif
412
413/*
414 * Returns true if any of the specified slub_debug flags is enabled for the
415 * cache. Use only for flags parsed by setup_slub_debug() as it also enables
416 * the static key.
417 */
418static inline bool kmem_cache_debug_flags(struct kmem_cache *s, slab_flags_t flags)
419{
0d4a062a
ME
420 if (IS_ENABLED(CONFIG_SLUB_DEBUG))
421 VM_WARN_ON_ONCE(!(flags & SLAB_DEBUG_FLAGS));
422 if (__slub_debug_enabled())
e42f174e 423 return s->flags & flags;
e42f174e
VB
424 return false;
425}
426
84c07d11 427#ifdef CONFIG_MEMCG_KMEM
4b5f8d9a
VB
428/*
429 * slab_objcgs - get the object cgroups vector associated with a slab
430 * @slab: a pointer to the slab struct
431 *
432 * Returns a pointer to the object cgroups vector associated with the slab,
433 * or NULL if no such vector has been associated yet.
434 */
435static inline struct obj_cgroup **slab_objcgs(struct slab *slab)
436{
437 unsigned long memcg_data = READ_ONCE(slab->memcg_data);
438
439 VM_BUG_ON_PAGE(memcg_data && !(memcg_data & MEMCG_DATA_OBJCGS),
440 slab_page(slab));
441 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_KMEM, slab_page(slab));
442
443 return (struct obj_cgroup **)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
444}
445
446int memcg_alloc_slab_cgroups(struct slab *slab, struct kmem_cache *s,
447 gfp_t gfp, bool new_slab);
fdbcb2a6
WL
448void mod_objcg_state(struct obj_cgroup *objcg, struct pglist_data *pgdat,
449 enum node_stat_item idx, int nr);
286e04b8 450
4b5f8d9a 451static inline void memcg_free_slab_cgroups(struct slab *slab)
286e04b8 452{
4b5f8d9a
VB
453 kfree(slab_objcgs(slab));
454 slab->memcg_data = 0;
286e04b8
RG
455}
456
f2fe7b09
RG
457static inline size_t obj_full_size(struct kmem_cache *s)
458{
459 /*
460 * For each accounted object there is an extra space which is used
461 * to store obj_cgroup membership. Charge it too.
462 */
463 return s->size + sizeof(struct obj_cgroup *);
464}
465
becaba65
RG
466/*
467 * Returns false if the allocation should fail.
468 */
469static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s,
88f2ef73 470 struct list_lru *lru,
becaba65
RG
471 struct obj_cgroup **objcgp,
472 size_t objects, gfp_t flags)
f2fe7b09 473{
9855609b
RG
474 struct obj_cgroup *objcg;
475
becaba65
RG
476 if (!memcg_kmem_enabled())
477 return true;
478
479 if (!(flags & __GFP_ACCOUNT) && !(s->flags & SLAB_ACCOUNT))
480 return true;
481
9855609b
RG
482 objcg = get_obj_cgroup_from_current();
483 if (!objcg)
becaba65 484 return true;
9855609b 485
88f2ef73
MS
486 if (lru) {
487 int ret;
488 struct mem_cgroup *memcg;
489
490 memcg = get_mem_cgroup_from_objcg(objcg);
491 ret = memcg_list_lru_alloc(memcg, lru, flags);
492 css_put(&memcg->css);
493
494 if (ret)
495 goto out;
f2fe7b09
RG
496 }
497
88f2ef73
MS
498 if (obj_cgroup_charge(objcg, flags, objects * obj_full_size(s)))
499 goto out;
500
becaba65
RG
501 *objcgp = objcg;
502 return true;
88f2ef73
MS
503out:
504 obj_cgroup_put(objcg);
505 return false;
f2fe7b09
RG
506}
507
964d4bd3
RG
508static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
509 struct obj_cgroup *objcg,
10befea9
RG
510 gfp_t flags, size_t size,
511 void **p)
964d4bd3 512{
4b5f8d9a 513 struct slab *slab;
964d4bd3
RG
514 unsigned long off;
515 size_t i;
516
becaba65 517 if (!memcg_kmem_enabled() || !objcg)
10befea9
RG
518 return;
519
964d4bd3
RG
520 for (i = 0; i < size; i++) {
521 if (likely(p[i])) {
4b5f8d9a 522 slab = virt_to_slab(p[i]);
10befea9 523
4b5f8d9a
VB
524 if (!slab_objcgs(slab) &&
525 memcg_alloc_slab_cgroups(slab, s, flags,
2e9bd483 526 false)) {
10befea9
RG
527 obj_cgroup_uncharge(objcg, obj_full_size(s));
528 continue;
529 }
530
4b5f8d9a 531 off = obj_to_index(s, slab, p[i]);
964d4bd3 532 obj_cgroup_get(objcg);
4b5f8d9a
VB
533 slab_objcgs(slab)[off] = objcg;
534 mod_objcg_state(objcg, slab_pgdat(slab),
f2fe7b09
RG
535 cache_vmstat_idx(s), obj_full_size(s));
536 } else {
537 obj_cgroup_uncharge(objcg, obj_full_size(s));
964d4bd3
RG
538 }
539 }
540 obj_cgroup_put(objcg);
964d4bd3
RG
541}
542
b77d5b1b 543static inline void memcg_slab_free_hook(struct kmem_cache *s, struct slab *slab,
d1b2cf6c 544 void **p, int objects)
964d4bd3 545{
270c6a71 546 struct obj_cgroup **objcgs;
d1b2cf6c 547 int i;
964d4bd3 548
10befea9
RG
549 if (!memcg_kmem_enabled())
550 return;
551
b77d5b1b
MS
552 objcgs = slab_objcgs(slab);
553 if (!objcgs)
554 return;
f2fe7b09 555
b77d5b1b
MS
556 for (i = 0; i < objects; i++) {
557 struct obj_cgroup *objcg;
558 unsigned int off;
10befea9 559
4b5f8d9a 560 off = obj_to_index(s, slab, p[i]);
270c6a71 561 objcg = objcgs[off];
d1b2cf6c
BR
562 if (!objcg)
563 continue;
f2fe7b09 564
270c6a71 565 objcgs[off] = NULL;
d1b2cf6c 566 obj_cgroup_uncharge(objcg, obj_full_size(s));
4b5f8d9a 567 mod_objcg_state(objcg, slab_pgdat(slab), cache_vmstat_idx(s),
d1b2cf6c
BR
568 -obj_full_size(s));
569 obj_cgroup_put(objcg);
570 }
964d4bd3
RG
571}
572
84c07d11 573#else /* CONFIG_MEMCG_KMEM */
4b5f8d9a
VB
574static inline struct obj_cgroup **slab_objcgs(struct slab *slab)
575{
576 return NULL;
577}
578
9855609b 579static inline struct mem_cgroup *memcg_from_slab_obj(void *ptr)
4d96ba35
RG
580{
581 return NULL;
582}
583
4b5f8d9a 584static inline int memcg_alloc_slab_cgroups(struct slab *slab,
2e9bd483 585 struct kmem_cache *s, gfp_t gfp,
4b5f8d9a 586 bool new_slab)
286e04b8
RG
587{
588 return 0;
589}
590
4b5f8d9a 591static inline void memcg_free_slab_cgroups(struct slab *slab)
286e04b8
RG
592{
593}
594
becaba65 595static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s,
88f2ef73 596 struct list_lru *lru,
becaba65
RG
597 struct obj_cgroup **objcgp,
598 size_t objects, gfp_t flags)
f2fe7b09 599{
becaba65 600 return true;
f2fe7b09
RG
601}
602
964d4bd3
RG
603static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
604 struct obj_cgroup *objcg,
10befea9
RG
605 gfp_t flags, size_t size,
606 void **p)
964d4bd3
RG
607{
608}
609
b77d5b1b 610static inline void memcg_slab_free_hook(struct kmem_cache *s, struct slab *slab,
d1b2cf6c 611 void **p, int objects)
964d4bd3
RG
612{
613}
84c07d11 614#endif /* CONFIG_MEMCG_KMEM */
b9ce5ef4 615
401fb12c 616#ifndef CONFIG_SLOB
a64b5378
KC
617static inline struct kmem_cache *virt_to_cache(const void *obj)
618{
82c1775d 619 struct slab *slab;
a64b5378 620
82c1775d
MWO
621 slab = virt_to_slab(obj);
622 if (WARN_ONCE(!slab, "%s: Object is not a Slab page!\n",
a64b5378
KC
623 __func__))
624 return NULL;
82c1775d 625 return slab->slab_cache;
a64b5378
KC
626}
627
b918653b
MWO
628static __always_inline void account_slab(struct slab *slab, int order,
629 struct kmem_cache *s, gfp_t gfp)
6cea1d56 630{
2e9bd483 631 if (memcg_kmem_enabled() && (s->flags & SLAB_ACCOUNT))
4b5f8d9a 632 memcg_alloc_slab_cgroups(slab, s, gfp, true);
2e9bd483 633
b918653b 634 mod_node_page_state(slab_pgdat(slab), cache_vmstat_idx(s),
f2fe7b09 635 PAGE_SIZE << order);
6cea1d56
RG
636}
637
b918653b
MWO
638static __always_inline void unaccount_slab(struct slab *slab, int order,
639 struct kmem_cache *s)
6cea1d56 640{
10befea9 641 if (memcg_kmem_enabled())
4b5f8d9a 642 memcg_free_slab_cgroups(slab);
9855609b 643
b918653b 644 mod_node_page_state(slab_pgdat(slab), cache_vmstat_idx(s),
f2fe7b09 645 -(PAGE_SIZE << order));
6cea1d56
RG
646}
647
e42f174e
VB
648static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x)
649{
650 struct kmem_cache *cachep;
651
652 if (!IS_ENABLED(CONFIG_SLAB_FREELIST_HARDENED) &&
e42f174e
VB
653 !kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS))
654 return s;
655
656 cachep = virt_to_cache(x);
10befea9 657 if (WARN(cachep && cachep != s,
e42f174e
VB
658 "%s: Wrong slab cache. %s but object is from %s\n",
659 __func__, s->name, cachep->name))
660 print_tracking(cachep, x);
661 return cachep;
662}
d6a71648
HY
663
664void free_large_kmalloc(struct folio *folio, void *object);
665
401fb12c 666#endif /* CONFIG_SLOB */
e42f174e 667
11c7aec2
JDB
668static inline size_t slab_ksize(const struct kmem_cache *s)
669{
670#ifndef CONFIG_SLUB
671 return s->object_size;
672
673#else /* CONFIG_SLUB */
674# ifdef CONFIG_SLUB_DEBUG
675 /*
676 * Debugging requires use of the padding between object
677 * and whatever may come after it.
678 */
679 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
680 return s->object_size;
681# endif
80a9201a
AP
682 if (s->flags & SLAB_KASAN)
683 return s->object_size;
11c7aec2
JDB
684 /*
685 * If we have the need to store the freelist pointer
686 * back there or track user information then we can
687 * only use the space before that information.
688 */
5f0d5a3a 689 if (s->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_STORE_USER))
11c7aec2
JDB
690 return s->inuse;
691 /*
692 * Else we can use all the padding etc for the allocation
693 */
694 return s->size;
695#endif
696}
697
698static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s,
88f2ef73 699 struct list_lru *lru,
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
88f2ef73 710 if (!memcg_slab_pre_alloc_hook(s, lru, objcgp, size, flags))
becaba65 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};
2dfe63e6 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 */