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 */
9 struct 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))
74 SLAB_MATCH(flags, __page_flags);
75 #ifndef CONFIG_SLOB
76 SLAB_MATCH(compound_head, slab_cache); /* Ensure bit 0 is clear */
77 #else
78 SLAB_MATCH(compound_head, slab_list); /* Ensure bit 0 is clear */
79 #endif
80 SLAB_MATCH(_refcount, __page_refcount);
81 #ifdef CONFIG_MEMCG
82 SLAB_MATCH(memcg_data, memcg_data);
83 #endif
84 #undef SLAB_MATCH
85 static_assert(sizeof(struct slab) <= sizeof(struct page));
86 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && defined(CONFIG_SLUB)
87 static_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 */
slab_test_pfmemalloc(const struct slab * slab)148 static inline bool slab_test_pfmemalloc(const struct slab *slab)
149 {
150 return folio_test_active((struct folio *)slab_folio(slab));
151 }
152
slab_set_pfmemalloc(struct slab * slab)153 static inline void slab_set_pfmemalloc(struct slab *slab)
154 {
155 folio_set_active(slab_folio(slab));
156 }
157
slab_clear_pfmemalloc(struct slab * slab)158 static inline void slab_clear_pfmemalloc(struct slab *slab)
159 {
160 folio_clear_active(slab_folio(slab));
161 }
162
__slab_clear_pfmemalloc(struct slab * slab)163 static inline void __slab_clear_pfmemalloc(struct slab *slab)
164 {
165 __folio_clear_active(slab_folio(slab));
166 }
167
slab_address(const struct slab * slab)168 static inline void *slab_address(const struct slab *slab)
169 {
170 return folio_address(slab_folio(slab));
171 }
172
slab_nid(const struct slab * slab)173 static inline int slab_nid(const struct slab *slab)
174 {
175 return folio_nid(slab_folio(slab));
176 }
177
slab_pgdat(const struct slab * slab)178 static inline pg_data_t *slab_pgdat(const struct slab *slab)
179 {
180 return folio_pgdat(slab_folio(slab));
181 }
182
virt_to_slab(const void * addr)183 static 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
slab_order(const struct slab * slab)193 static inline int slab_order(const struct slab *slab)
194 {
195 return folio_order((struct folio *)slab_folio(slab));
196 }
197
slab_size(const struct slab * slab)198 static 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 */
215 struct 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 */
252 enum 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
260 extern enum slab_state slab_state;
261
262 /* The slab cache mutex protects the management structures during changes */
263 extern struct mutex slab_mutex;
264
265 /* The list of all slab caches on the system */
266 extern struct list_head slab_caches;
267
268 /* The slab cache that manages slab cache information */
269 extern struct kmem_cache *kmem_cache;
270
271 /* A table of kmalloc cache names and sizes */
272 extern 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 */
279 void setup_kmalloc_cache_index_table(void);
280 void create_kmalloc_caches(slab_flags_t);
281
282 /* Find the kmalloc slab corresponding for a certain size */
283 struct kmem_cache *kmalloc_slab(size_t, gfp_t);
284
285 void *__kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags,
286 int node, size_t orig_size,
287 unsigned long caller);
288 void __kmem_cache_free(struct kmem_cache *s, void *x, unsigned long caller);
289 #endif
290
291 gfp_t kmalloc_fix_flags(gfp_t flags);
292
293 /* Functions provided by the slab allocators */
294 int __kmem_cache_create(struct kmem_cache *, slab_flags_t flags);
295
296 struct kmem_cache *create_kmalloc_cache(const char *name, unsigned int size,
297 slab_flags_t flags, unsigned int useroffset,
298 unsigned int usersize);
299 extern 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
303 int slab_unmergeable(struct kmem_cache *s);
304 struct kmem_cache *find_mergeable(unsigned size, unsigned align,
305 slab_flags_t flags, const char *name, void (*ctor)(void *));
306 #ifndef CONFIG_SLOB
307 struct kmem_cache *
308 __kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
309 slab_flags_t flags, void (*ctor)(void *));
310
311 slab_flags_t kmem_cache_flags(unsigned int object_size,
312 slab_flags_t flags, const char *name);
313 #else
314 static inline struct kmem_cache *
__kmem_cache_alias(const char * name,unsigned int size,unsigned int align,slab_flags_t flags,void (* ctor)(void *))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
kmem_cache_flags(unsigned int object_size,slab_flags_t flags,const char * name)319 static 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
is_kmalloc_cache(struct kmem_cache * s)326 static inline bool is_kmalloc_cache(struct kmem_cache *s)
327 {
328 #ifndef CONFIG_SLOB
329 return (s->flags & SLAB_KMALLOC);
330 #else
331 return false;
332 #endif
333 }
334
335 /* Legal flag mask for kmem_cache_create(), for various configurations */
336 #define SLAB_CORE_FLAGS (SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA | \
337 SLAB_CACHE_DMA32 | SLAB_PANIC | \
338 SLAB_TYPESAFE_BY_RCU | SLAB_DEBUG_OBJECTS )
339
340 #if defined(CONFIG_DEBUG_SLAB)
341 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
342 #elif defined(CONFIG_SLUB_DEBUG)
343 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
344 SLAB_TRACE | SLAB_CONSISTENCY_CHECKS)
345 #else
346 #define SLAB_DEBUG_FLAGS (0)
347 #endif
348
349 #if defined(CONFIG_SLAB)
350 #define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \
351 SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \
352 SLAB_ACCOUNT)
353 #elif defined(CONFIG_SLUB)
354 #define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
355 SLAB_TEMPORARY | SLAB_ACCOUNT | \
356 SLAB_NO_USER_FLAGS | SLAB_KMALLOC)
357 #else
358 #define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE)
359 #endif
360
361 /* Common flags available with current configuration */
362 #define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
363
364 /* Common flags permitted for kmem_cache_create */
365 #define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \
366 SLAB_RED_ZONE | \
367 SLAB_POISON | \
368 SLAB_STORE_USER | \
369 SLAB_TRACE | \
370 SLAB_CONSISTENCY_CHECKS | \
371 SLAB_MEM_SPREAD | \
372 SLAB_NOLEAKTRACE | \
373 SLAB_RECLAIM_ACCOUNT | \
374 SLAB_TEMPORARY | \
375 SLAB_ACCOUNT | \
376 SLAB_KMALLOC | \
377 SLAB_NO_USER_FLAGS)
378
379 bool __kmem_cache_empty(struct kmem_cache *);
380 int __kmem_cache_shutdown(struct kmem_cache *);
381 void __kmem_cache_release(struct kmem_cache *);
382 int __kmem_cache_shrink(struct kmem_cache *);
383 void slab_kmem_cache_release(struct kmem_cache *);
384
385 struct seq_file;
386 struct file;
387
388 struct slabinfo {
389 unsigned long active_objs;
390 unsigned long num_objs;
391 unsigned long active_slabs;
392 unsigned long num_slabs;
393 unsigned long shared_avail;
394 unsigned int limit;
395 unsigned int batchcount;
396 unsigned int shared;
397 unsigned int objects_per_slab;
398 unsigned int cache_order;
399 };
400
401 void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo);
402 void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s);
403 ssize_t slabinfo_write(struct file *file, const char __user *buffer,
404 size_t count, loff_t *ppos);
405
cache_vmstat_idx(struct kmem_cache * s)406 static inline enum node_stat_item cache_vmstat_idx(struct kmem_cache *s)
407 {
408 return (s->flags & SLAB_RECLAIM_ACCOUNT) ?
409 NR_SLAB_RECLAIMABLE_B : NR_SLAB_UNRECLAIMABLE_B;
410 }
411
412 #ifdef CONFIG_SLUB_DEBUG
413 #ifdef CONFIG_SLUB_DEBUG_ON
414 DECLARE_STATIC_KEY_TRUE(slub_debug_enabled);
415 #else
416 DECLARE_STATIC_KEY_FALSE(slub_debug_enabled);
417 #endif
418 extern void print_tracking(struct kmem_cache *s, void *object);
419 long validate_slab_cache(struct kmem_cache *s);
__slub_debug_enabled(void)420 static inline bool __slub_debug_enabled(void)
421 {
422 return static_branch_unlikely(&slub_debug_enabled);
423 }
424 #else
print_tracking(struct kmem_cache * s,void * object)425 static inline void print_tracking(struct kmem_cache *s, void *object)
426 {
427 }
__slub_debug_enabled(void)428 static inline bool __slub_debug_enabled(void)
429 {
430 return false;
431 }
432 #endif
433
434 /*
435 * Returns true if any of the specified slub_debug flags is enabled for the
436 * cache. Use only for flags parsed by setup_slub_debug() as it also enables
437 * the static key.
438 */
kmem_cache_debug_flags(struct kmem_cache * s,slab_flags_t flags)439 static inline bool kmem_cache_debug_flags(struct kmem_cache *s, slab_flags_t flags)
440 {
441 if (IS_ENABLED(CONFIG_SLUB_DEBUG))
442 VM_WARN_ON_ONCE(!(flags & SLAB_DEBUG_FLAGS));
443 if (__slub_debug_enabled())
444 return s->flags & flags;
445 return false;
446 }
447
448 #ifdef CONFIG_MEMCG_KMEM
449 /*
450 * slab_objcgs - get the object cgroups vector associated with a slab
451 * @slab: a pointer to the slab struct
452 *
453 * Returns a pointer to the object cgroups vector associated with the slab,
454 * or NULL if no such vector has been associated yet.
455 */
slab_objcgs(struct slab * slab)456 static inline struct obj_cgroup **slab_objcgs(struct slab *slab)
457 {
458 unsigned long memcg_data = READ_ONCE(slab->memcg_data);
459
460 VM_BUG_ON_PAGE(memcg_data && !(memcg_data & MEMCG_DATA_OBJCGS),
461 slab_page(slab));
462 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_KMEM, slab_page(slab));
463
464 return (struct obj_cgroup **)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
465 }
466
467 int memcg_alloc_slab_cgroups(struct slab *slab, struct kmem_cache *s,
468 gfp_t gfp, bool new_slab);
469 void mod_objcg_state(struct obj_cgroup *objcg, struct pglist_data *pgdat,
470 enum node_stat_item idx, int nr);
471
memcg_free_slab_cgroups(struct slab * slab)472 static inline void memcg_free_slab_cgroups(struct slab *slab)
473 {
474 kfree(slab_objcgs(slab));
475 slab->memcg_data = 0;
476 }
477
obj_full_size(struct kmem_cache * s)478 static inline size_t obj_full_size(struct kmem_cache *s)
479 {
480 /*
481 * For each accounted object there is an extra space which is used
482 * to store obj_cgroup membership. Charge it too.
483 */
484 return s->size + sizeof(struct obj_cgroup *);
485 }
486
487 /*
488 * Returns false if the allocation should fail.
489 */
memcg_slab_pre_alloc_hook(struct kmem_cache * s,struct list_lru * lru,struct obj_cgroup ** objcgp,size_t objects,gfp_t flags)490 static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s,
491 struct list_lru *lru,
492 struct obj_cgroup **objcgp,
493 size_t objects, gfp_t flags)
494 {
495 struct obj_cgroup *objcg;
496
497 if (!memcg_kmem_online())
498 return true;
499
500 if (!(flags & __GFP_ACCOUNT) && !(s->flags & SLAB_ACCOUNT))
501 return true;
502
503 objcg = get_obj_cgroup_from_current();
504 if (!objcg)
505 return true;
506
507 if (lru) {
508 int ret;
509 struct mem_cgroup *memcg;
510
511 memcg = get_mem_cgroup_from_objcg(objcg);
512 ret = memcg_list_lru_alloc(memcg, lru, flags);
513 css_put(&memcg->css);
514
515 if (ret)
516 goto out;
517 }
518
519 if (obj_cgroup_charge(objcg, flags, objects * obj_full_size(s)))
520 goto out;
521
522 *objcgp = objcg;
523 return true;
524 out:
525 obj_cgroup_put(objcg);
526 return false;
527 }
528
memcg_slab_post_alloc_hook(struct kmem_cache * s,struct obj_cgroup * objcg,gfp_t flags,size_t size,void ** p)529 static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
530 struct obj_cgroup *objcg,
531 gfp_t flags, size_t size,
532 void **p)
533 {
534 struct slab *slab;
535 unsigned long off;
536 size_t i;
537
538 if (!memcg_kmem_online() || !objcg)
539 return;
540
541 for (i = 0; i < size; i++) {
542 if (likely(p[i])) {
543 slab = virt_to_slab(p[i]);
544
545 if (!slab_objcgs(slab) &&
546 memcg_alloc_slab_cgroups(slab, s, flags,
547 false)) {
548 obj_cgroup_uncharge(objcg, obj_full_size(s));
549 continue;
550 }
551
552 off = obj_to_index(s, slab, p[i]);
553 obj_cgroup_get(objcg);
554 slab_objcgs(slab)[off] = objcg;
555 mod_objcg_state(objcg, slab_pgdat(slab),
556 cache_vmstat_idx(s), obj_full_size(s));
557 } else {
558 obj_cgroup_uncharge(objcg, obj_full_size(s));
559 }
560 }
561 obj_cgroup_put(objcg);
562 }
563
memcg_slab_free_hook(struct kmem_cache * s,struct slab * slab,void ** p,int objects)564 static inline void memcg_slab_free_hook(struct kmem_cache *s, struct slab *slab,
565 void **p, int objects)
566 {
567 struct obj_cgroup **objcgs;
568 int i;
569
570 if (!memcg_kmem_online())
571 return;
572
573 objcgs = slab_objcgs(slab);
574 if (!objcgs)
575 return;
576
577 for (i = 0; i < objects; i++) {
578 struct obj_cgroup *objcg;
579 unsigned int off;
580
581 off = obj_to_index(s, slab, p[i]);
582 objcg = objcgs[off];
583 if (!objcg)
584 continue;
585
586 objcgs[off] = NULL;
587 obj_cgroup_uncharge(objcg, obj_full_size(s));
588 mod_objcg_state(objcg, slab_pgdat(slab), cache_vmstat_idx(s),
589 -obj_full_size(s));
590 obj_cgroup_put(objcg);
591 }
592 }
593
594 #else /* CONFIG_MEMCG_KMEM */
slab_objcgs(struct slab * slab)595 static inline struct obj_cgroup **slab_objcgs(struct slab *slab)
596 {
597 return NULL;
598 }
599
memcg_from_slab_obj(void * ptr)600 static inline struct mem_cgroup *memcg_from_slab_obj(void *ptr)
601 {
602 return NULL;
603 }
604
memcg_alloc_slab_cgroups(struct slab * slab,struct kmem_cache * s,gfp_t gfp,bool new_slab)605 static inline int memcg_alloc_slab_cgroups(struct slab *slab,
606 struct kmem_cache *s, gfp_t gfp,
607 bool new_slab)
608 {
609 return 0;
610 }
611
memcg_free_slab_cgroups(struct slab * slab)612 static inline void memcg_free_slab_cgroups(struct slab *slab)
613 {
614 }
615
memcg_slab_pre_alloc_hook(struct kmem_cache * s,struct list_lru * lru,struct obj_cgroup ** objcgp,size_t objects,gfp_t flags)616 static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s,
617 struct list_lru *lru,
618 struct obj_cgroup **objcgp,
619 size_t objects, gfp_t flags)
620 {
621 return true;
622 }
623
memcg_slab_post_alloc_hook(struct kmem_cache * s,struct obj_cgroup * objcg,gfp_t flags,size_t size,void ** p)624 static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
625 struct obj_cgroup *objcg,
626 gfp_t flags, size_t size,
627 void **p)
628 {
629 }
630
memcg_slab_free_hook(struct kmem_cache * s,struct slab * slab,void ** p,int objects)631 static inline void memcg_slab_free_hook(struct kmem_cache *s, struct slab *slab,
632 void **p, int objects)
633 {
634 }
635 #endif /* CONFIG_MEMCG_KMEM */
636
637 #ifndef CONFIG_SLOB
virt_to_cache(const void * obj)638 static inline struct kmem_cache *virt_to_cache(const void *obj)
639 {
640 struct slab *slab;
641
642 slab = virt_to_slab(obj);
643 if (WARN_ONCE(!slab, "%s: Object is not a Slab page!\n",
644 __func__))
645 return NULL;
646 return slab->slab_cache;
647 }
648
account_slab(struct slab * slab,int order,struct kmem_cache * s,gfp_t gfp)649 static __always_inline void account_slab(struct slab *slab, int order,
650 struct kmem_cache *s, gfp_t gfp)
651 {
652 if (memcg_kmem_online() && (s->flags & SLAB_ACCOUNT))
653 memcg_alloc_slab_cgroups(slab, s, gfp, true);
654
655 mod_node_page_state(slab_pgdat(slab), cache_vmstat_idx(s),
656 PAGE_SIZE << order);
657 }
658
unaccount_slab(struct slab * slab,int order,struct kmem_cache * s)659 static __always_inline void unaccount_slab(struct slab *slab, int order,
660 struct kmem_cache *s)
661 {
662 if (memcg_kmem_online())
663 memcg_free_slab_cgroups(slab);
664
665 mod_node_page_state(slab_pgdat(slab), cache_vmstat_idx(s),
666 -(PAGE_SIZE << order));
667 }
668
cache_from_obj(struct kmem_cache * s,void * x)669 static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x)
670 {
671 struct kmem_cache *cachep;
672
673 if (!IS_ENABLED(CONFIG_SLAB_FREELIST_HARDENED) &&
674 !kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS))
675 return s;
676
677 cachep = virt_to_cache(x);
678 if (WARN(cachep && cachep != s,
679 "%s: Wrong slab cache. %s but object is from %s\n",
680 __func__, s->name, cachep->name))
681 print_tracking(cachep, x);
682 return cachep;
683 }
684
685 void free_large_kmalloc(struct folio *folio, void *object);
686
687 #endif /* CONFIG_SLOB */
688
689 size_t __ksize(const void *objp);
690
slab_ksize(const struct kmem_cache * s)691 static inline size_t slab_ksize(const struct kmem_cache *s)
692 {
693 #ifndef CONFIG_SLUB
694 return s->object_size;
695
696 #else /* CONFIG_SLUB */
697 # ifdef CONFIG_SLUB_DEBUG
698 /*
699 * Debugging requires use of the padding between object
700 * and whatever may come after it.
701 */
702 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
703 return s->object_size;
704 # endif
705 if (s->flags & SLAB_KASAN)
706 return s->object_size;
707 /*
708 * If we have the need to store the freelist pointer
709 * back there or track user information then we can
710 * only use the space before that information.
711 */
712 if (s->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_STORE_USER))
713 return s->inuse;
714 /*
715 * Else we can use all the padding etc for the allocation
716 */
717 return s->size;
718 #endif
719 }
720
slab_pre_alloc_hook(struct kmem_cache * s,struct list_lru * lru,struct obj_cgroup ** objcgp,size_t size,gfp_t flags)721 static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s,
722 struct list_lru *lru,
723 struct obj_cgroup **objcgp,
724 size_t size, gfp_t flags)
725 {
726 flags &= gfp_allowed_mask;
727
728 might_alloc(flags);
729
730 if (should_failslab(s, flags))
731 return NULL;
732
733 if (!memcg_slab_pre_alloc_hook(s, lru, objcgp, size, flags))
734 return NULL;
735
736 return s;
737 }
738
slab_post_alloc_hook(struct kmem_cache * s,struct obj_cgroup * objcg,gfp_t flags,size_t size,void ** p,bool init,unsigned int orig_size)739 static inline void slab_post_alloc_hook(struct kmem_cache *s,
740 struct obj_cgroup *objcg, gfp_t flags,
741 size_t size, void **p, bool init,
742 unsigned int orig_size)
743 {
744 unsigned int zero_size = s->object_size;
745 size_t i;
746
747 flags &= gfp_allowed_mask;
748
749 /*
750 * For kmalloc object, the allocated memory size(object_size) is likely
751 * larger than the requested size(orig_size). If redzone check is
752 * enabled for the extra space, don't zero it, as it will be redzoned
753 * soon. The redzone operation for this extra space could be seen as a
754 * replacement of current poisoning under certain debug option, and
755 * won't break other sanity checks.
756 */
757 if (kmem_cache_debug_flags(s, SLAB_STORE_USER | SLAB_RED_ZONE) &&
758 (s->flags & SLAB_KMALLOC))
759 zero_size = orig_size;
760
761 /*
762 * As memory initialization might be integrated into KASAN,
763 * kasan_slab_alloc and initialization memset must be
764 * kept together to avoid discrepancies in behavior.
765 *
766 * As p[i] might get tagged, memset and kmemleak hook come after KASAN.
767 */
768 for (i = 0; i < size; i++) {
769 p[i] = kasan_slab_alloc(s, p[i], flags, init);
770 if (p[i] && init && !kasan_has_integrated_init())
771 memset(p[i], 0, zero_size);
772 kmemleak_alloc_recursive(p[i], s->object_size, 1,
773 s->flags, flags);
774 kmsan_slab_alloc(s, p[i], flags);
775 }
776
777 memcg_slab_post_alloc_hook(s, objcg, flags, size, p);
778 }
779
780 #ifndef CONFIG_SLOB
781 /*
782 * The slab lists for all objects.
783 */
784 struct kmem_cache_node {
785 #ifdef CONFIG_SLAB
786 raw_spinlock_t list_lock;
787 struct list_head slabs_partial; /* partial list first, better asm code */
788 struct list_head slabs_full;
789 struct list_head slabs_free;
790 unsigned long total_slabs; /* length of all slab lists */
791 unsigned long free_slabs; /* length of free slab list only */
792 unsigned long free_objects;
793 unsigned int free_limit;
794 unsigned int colour_next; /* Per-node cache coloring */
795 struct array_cache *shared; /* shared per node */
796 struct alien_cache **alien; /* on other nodes */
797 unsigned long next_reap; /* updated without locking */
798 int free_touched; /* updated without locking */
799 #endif
800
801 #ifdef CONFIG_SLUB
802 spinlock_t list_lock;
803 unsigned long nr_partial;
804 struct list_head partial;
805 #ifdef CONFIG_SLUB_DEBUG
806 atomic_long_t nr_slabs;
807 atomic_long_t total_objects;
808 struct list_head full;
809 #endif
810 #endif
811
812 };
813
get_node(struct kmem_cache * s,int node)814 static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
815 {
816 return s->node[node];
817 }
818
819 /*
820 * Iterator over all nodes. The body will be executed for each node that has
821 * a kmem_cache_node structure allocated (which is true for all online nodes)
822 */
823 #define for_each_kmem_cache_node(__s, __node, __n) \
824 for (__node = 0; __node < nr_node_ids; __node++) \
825 if ((__n = get_node(__s, __node)))
826
827 #endif
828
829 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
830 void dump_unreclaimable_slab(void);
831 #else
dump_unreclaimable_slab(void)832 static inline void dump_unreclaimable_slab(void)
833 {
834 }
835 #endif
836
837 void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr);
838
839 #ifdef CONFIG_SLAB_FREELIST_RANDOM
840 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
841 gfp_t gfp);
842 void cache_random_seq_destroy(struct kmem_cache *cachep);
843 #else
cache_random_seq_create(struct kmem_cache * cachep,unsigned int count,gfp_t gfp)844 static inline int cache_random_seq_create(struct kmem_cache *cachep,
845 unsigned int count, gfp_t gfp)
846 {
847 return 0;
848 }
cache_random_seq_destroy(struct kmem_cache * cachep)849 static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { }
850 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
851
slab_want_init_on_alloc(gfp_t flags,struct kmem_cache * c)852 static inline bool slab_want_init_on_alloc(gfp_t flags, struct kmem_cache *c)
853 {
854 if (static_branch_maybe(CONFIG_INIT_ON_ALLOC_DEFAULT_ON,
855 &init_on_alloc)) {
856 if (c->ctor)
857 return false;
858 if (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON))
859 return flags & __GFP_ZERO;
860 return true;
861 }
862 return flags & __GFP_ZERO;
863 }
864
slab_want_init_on_free(struct kmem_cache * c)865 static inline bool slab_want_init_on_free(struct kmem_cache *c)
866 {
867 if (static_branch_maybe(CONFIG_INIT_ON_FREE_DEFAULT_ON,
868 &init_on_free))
869 return !(c->ctor ||
870 (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)));
871 return false;
872 }
873
874 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG)
875 void debugfs_slab_release(struct kmem_cache *);
876 #else
debugfs_slab_release(struct kmem_cache * s)877 static inline void debugfs_slab_release(struct kmem_cache *s) { }
878 #endif
879
880 #ifdef CONFIG_PRINTK
881 #define KS_ADDRS_COUNT 16
882 struct kmem_obj_info {
883 void *kp_ptr;
884 struct slab *kp_slab;
885 void *kp_objp;
886 unsigned long kp_data_offset;
887 struct kmem_cache *kp_slab_cache;
888 void *kp_ret;
889 void *kp_stack[KS_ADDRS_COUNT];
890 void *kp_free_stack[KS_ADDRS_COUNT];
891 };
892 void __kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab);
893 #endif
894
895 #ifdef CONFIG_HAVE_HARDENED_USERCOPY_ALLOCATOR
896 void __check_heap_object(const void *ptr, unsigned long n,
897 const struct slab *slab, bool to_user);
898 #else
899 static inline
__check_heap_object(const void * ptr,unsigned long n,const struct slab * slab,bool to_user)900 void __check_heap_object(const void *ptr, unsigned long n,
901 const struct slab *slab, bool to_user)
902 {
903 }
904 #endif
905
906 #ifdef CONFIG_SLUB_DEBUG
907 void skip_orig_size_check(struct kmem_cache *s, const void *object);
908 #endif
909
910 #endif /* MM_SLAB_H */
911