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 union {
15 struct list_head slab_list;
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;
28 #ifdef CONFIG_SLUB_CPU_PARTIAL
29 struct {
30 struct slab *next;
31 int slabs; /* Nr of slabs left */
32 };
33 #endif
34 };
35 struct kmem_cache *slab_cache;
36 /* Double-word boundary */
37 void *freelist; /* first free object */
38 union {
39 unsigned long counters;
40 struct {
41 unsigned inuse:16;
42 unsigned objects:15;
43 unsigned frozen:1;
44 };
45 };
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 */
53 long units;
54 unsigned int __unused_2;
55
56 #else
57 #error "Unexpected slab allocator configured"
58 #endif
59
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))
68 SLAB_MATCH(flags, __page_flags);
69 SLAB_MATCH(compound_head, slab_list); /* Ensure bit 0 is clear */
70 #ifndef CONFIG_SLOB
71 SLAB_MATCH(rcu_head, rcu_head);
72 #endif
73 SLAB_MATCH(_refcount, __page_refcount);
74 #ifdef CONFIG_MEMCG
75 SLAB_MATCH(memcg_data, memcg_data);
76 #endif
77 #undef SLAB_MATCH
78 static_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 */
slab_test_pfmemalloc(const struct slab * slab)138 static inline bool slab_test_pfmemalloc(const struct slab *slab)
139 {
140 return folio_test_active((struct folio *)slab_folio(slab));
141 }
142
slab_set_pfmemalloc(struct slab * slab)143 static inline void slab_set_pfmemalloc(struct slab *slab)
144 {
145 folio_set_active(slab_folio(slab));
146 }
147
slab_clear_pfmemalloc(struct slab * slab)148 static inline void slab_clear_pfmemalloc(struct slab *slab)
149 {
150 folio_clear_active(slab_folio(slab));
151 }
152
__slab_clear_pfmemalloc(struct slab * slab)153 static inline void __slab_clear_pfmemalloc(struct slab *slab)
154 {
155 __folio_clear_active(slab_folio(slab));
156 }
157
slab_address(const struct slab * slab)158 static inline void *slab_address(const struct slab *slab)
159 {
160 return folio_address(slab_folio(slab));
161 }
162
slab_nid(const struct slab * slab)163 static inline int slab_nid(const struct slab *slab)
164 {
165 return folio_nid(slab_folio(slab));
166 }
167
slab_pgdat(const struct slab * slab)168 static inline pg_data_t *slab_pgdat(const struct slab *slab)
169 {
170 return folio_pgdat(slab_folio(slab));
171 }
172
virt_to_slab(const void * addr)173 static 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
slab_order(const struct slab * slab)183 static inline int slab_order(const struct slab *slab)
184 {
185 return folio_order((struct folio *)slab_folio(slab));
186 }
187
slab_size(const struct slab * slab)188 static inline size_t slab_size(const struct slab *slab)
189 {
190 return PAGE_SIZE << slab_order(slab);
191 }
192
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 */
205 struct 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 */
209 slab_flags_t flags; /* Active flags on the slab */
210 unsigned int useroffset;/* Usercopy region offset */
211 unsigned int usersize; /* Usercopy region size */
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>
229 #include <linux/fault-inject.h>
230 #include <linux/kasan.h>
231 #include <linux/kmemleak.h>
232 #include <linux/random.h>
233 #include <linux/sched/mm.h>
234 #include <linux/list_lru.h>
235
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 */
244 enum slab_state {
245 DOWN, /* No slab functionality yet */
246 PARTIAL, /* SLUB: kmem_cache_node available */
247 PARTIAL_NODE, /* SLAB: kmalloc size for node struct available */
248 UP, /* Slab caches usable but not all extras yet */
249 FULL /* Everything is working */
250 };
251
252 extern enum slab_state slab_state;
253
254 /* The slab cache mutex protects the management structures during changes */
255 extern struct mutex slab_mutex;
256
257 /* The list of all slab caches on the system */
258 extern struct list_head slab_caches;
259
260 /* The slab cache that manages slab cache information */
261 extern struct kmem_cache *kmem_cache;
262
263 /* A table of kmalloc cache names and sizes */
264 extern const struct kmalloc_info_struct {
265 const char *name[NR_KMALLOC_TYPES];
266 unsigned int size;
267 } kmalloc_info[];
268
269 #ifndef CONFIG_SLOB
270 /* Kmalloc array related functions */
271 void setup_kmalloc_cache_index_table(void);
272 void create_kmalloc_caches(slab_flags_t);
273
274 /* Find the kmalloc slab corresponding for a certain size */
275 struct kmem_cache *kmalloc_slab(size_t, gfp_t);
276 #endif
277
278 gfp_t kmalloc_fix_flags(gfp_t flags);
279
280 /* Functions provided by the slab allocators */
281 int __kmem_cache_create(struct kmem_cache *, slab_flags_t flags);
282
283 struct kmem_cache *create_kmalloc_cache(const char *name, unsigned int size,
284 slab_flags_t flags, unsigned int useroffset,
285 unsigned int usersize);
286 extern void create_boot_cache(struct kmem_cache *, const char *name,
287 unsigned int size, slab_flags_t flags,
288 unsigned int useroffset, unsigned int usersize);
289
290 int slab_unmergeable(struct kmem_cache *s);
291 struct kmem_cache *find_mergeable(unsigned size, unsigned align,
292 slab_flags_t flags, const char *name, void (*ctor)(void *));
293 #ifndef CONFIG_SLOB
294 struct kmem_cache *
295 __kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
296 slab_flags_t flags, void (*ctor)(void *));
297
298 slab_flags_t kmem_cache_flags(unsigned int object_size,
299 slab_flags_t flags, const char *name);
300 #else
301 static inline struct kmem_cache *
__kmem_cache_alias(const char * name,unsigned int size,unsigned int align,slab_flags_t flags,void (* ctor)(void *))302 __kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
303 slab_flags_t flags, void (*ctor)(void *))
304 { return NULL; }
305
kmem_cache_flags(unsigned int object_size,slab_flags_t flags,const char * name)306 static inline slab_flags_t kmem_cache_flags(unsigned int object_size,
307 slab_flags_t flags, const char *name)
308 {
309 return flags;
310 }
311 #endif
312
313
314 /* Legal flag mask for kmem_cache_create(), for various configurations */
315 #define SLAB_CORE_FLAGS (SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA | \
316 SLAB_CACHE_DMA32 | SLAB_PANIC | \
317 SLAB_TYPESAFE_BY_RCU | SLAB_DEBUG_OBJECTS )
318
319 #if defined(CONFIG_DEBUG_SLAB)
320 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
321 #elif defined(CONFIG_SLUB_DEBUG)
322 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
323 SLAB_TRACE | SLAB_CONSISTENCY_CHECKS)
324 #else
325 #define SLAB_DEBUG_FLAGS (0)
326 #endif
327
328 #if defined(CONFIG_SLAB)
329 #define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \
330 SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \
331 SLAB_ACCOUNT)
332 #elif defined(CONFIG_SLUB)
333 #define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
334 SLAB_TEMPORARY | SLAB_ACCOUNT | SLAB_NO_USER_FLAGS)
335 #else
336 #define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE)
337 #endif
338
339 /* Common flags available with current configuration */
340 #define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
341
342 /* Common flags permitted for kmem_cache_create */
343 #define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \
344 SLAB_RED_ZONE | \
345 SLAB_POISON | \
346 SLAB_STORE_USER | \
347 SLAB_TRACE | \
348 SLAB_CONSISTENCY_CHECKS | \
349 SLAB_MEM_SPREAD | \
350 SLAB_NOLEAKTRACE | \
351 SLAB_RECLAIM_ACCOUNT | \
352 SLAB_TEMPORARY | \
353 SLAB_ACCOUNT | \
354 SLAB_NO_USER_FLAGS)
355
356 bool __kmem_cache_empty(struct kmem_cache *);
357 int __kmem_cache_shutdown(struct kmem_cache *);
358 void __kmem_cache_release(struct kmem_cache *);
359 int __kmem_cache_shrink(struct kmem_cache *);
360 void slab_kmem_cache_release(struct kmem_cache *);
361
362 struct seq_file;
363 struct file;
364
365 struct slabinfo {
366 unsigned long active_objs;
367 unsigned long num_objs;
368 unsigned long active_slabs;
369 unsigned long num_slabs;
370 unsigned long shared_avail;
371 unsigned int limit;
372 unsigned int batchcount;
373 unsigned int shared;
374 unsigned int objects_per_slab;
375 unsigned int cache_order;
376 };
377
378 void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo);
379 void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s);
380 ssize_t slabinfo_write(struct file *file, const char __user *buffer,
381 size_t count, loff_t *ppos);
382
383 /*
384 * Generic implementation of bulk operations
385 * These are useful for situations in which the allocator cannot
386 * perform optimizations. In that case segments of the object listed
387 * may be allocated or freed using these operations.
388 */
389 void __kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
390 int __kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
391
cache_vmstat_idx(struct kmem_cache * s)392 static inline enum node_stat_item cache_vmstat_idx(struct kmem_cache *s)
393 {
394 return (s->flags & SLAB_RECLAIM_ACCOUNT) ?
395 NR_SLAB_RECLAIMABLE_B : NR_SLAB_UNRECLAIMABLE_B;
396 }
397
398 #ifdef CONFIG_SLUB_DEBUG
399 #ifdef CONFIG_SLUB_DEBUG_ON
400 DECLARE_STATIC_KEY_TRUE(slub_debug_enabled);
401 #else
402 DECLARE_STATIC_KEY_FALSE(slub_debug_enabled);
403 #endif
404 extern void print_tracking(struct kmem_cache *s, void *object);
405 long validate_slab_cache(struct kmem_cache *s);
__slub_debug_enabled(void)406 static inline bool __slub_debug_enabled(void)
407 {
408 return static_branch_unlikely(&slub_debug_enabled);
409 }
410 #else
print_tracking(struct kmem_cache * s,void * object)411 static inline void print_tracking(struct kmem_cache *s, void *object)
412 {
413 }
__slub_debug_enabled(void)414 static inline bool __slub_debug_enabled(void)
415 {
416 return false;
417 }
418 #endif
419
420 /*
421 * Returns true if any of the specified slub_debug flags is enabled for the
422 * cache. Use only for flags parsed by setup_slub_debug() as it also enables
423 * the static key.
424 */
kmem_cache_debug_flags(struct kmem_cache * s,slab_flags_t flags)425 static inline bool kmem_cache_debug_flags(struct kmem_cache *s, slab_flags_t flags)
426 {
427 if (IS_ENABLED(CONFIG_SLUB_DEBUG))
428 VM_WARN_ON_ONCE(!(flags & SLAB_DEBUG_FLAGS));
429 if (__slub_debug_enabled())
430 return s->flags & flags;
431 return false;
432 }
433
434 #ifdef CONFIG_MEMCG_KMEM
435 /*
436 * slab_objcgs - get the object cgroups vector associated with a slab
437 * @slab: a pointer to the slab struct
438 *
439 * Returns a pointer to the object cgroups vector associated with the slab,
440 * or NULL if no such vector has been associated yet.
441 */
slab_objcgs(struct slab * slab)442 static inline struct obj_cgroup **slab_objcgs(struct slab *slab)
443 {
444 unsigned long memcg_data = READ_ONCE(slab->memcg_data);
445
446 VM_BUG_ON_PAGE(memcg_data && !(memcg_data & MEMCG_DATA_OBJCGS),
447 slab_page(slab));
448 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_KMEM, slab_page(slab));
449
450 return (struct obj_cgroup **)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
451 }
452
453 int memcg_alloc_slab_cgroups(struct slab *slab, struct kmem_cache *s,
454 gfp_t gfp, bool new_slab);
455 void mod_objcg_state(struct obj_cgroup *objcg, struct pglist_data *pgdat,
456 enum node_stat_item idx, int nr);
457
memcg_free_slab_cgroups(struct slab * slab)458 static inline void memcg_free_slab_cgroups(struct slab *slab)
459 {
460 kfree(slab_objcgs(slab));
461 slab->memcg_data = 0;
462 }
463
obj_full_size(struct kmem_cache * s)464 static inline size_t obj_full_size(struct kmem_cache *s)
465 {
466 /*
467 * For each accounted object there is an extra space which is used
468 * to store obj_cgroup membership. Charge it too.
469 */
470 return s->size + sizeof(struct obj_cgroup *);
471 }
472
473 /*
474 * Returns false if the allocation should fail.
475 */
memcg_slab_pre_alloc_hook(struct kmem_cache * s,struct list_lru * lru,struct obj_cgroup ** objcgp,size_t objects,gfp_t flags)476 static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s,
477 struct list_lru *lru,
478 struct obj_cgroup **objcgp,
479 size_t objects, gfp_t flags)
480 {
481 struct obj_cgroup *objcg;
482
483 if (!memcg_kmem_enabled())
484 return true;
485
486 if (!(flags & __GFP_ACCOUNT) && !(s->flags & SLAB_ACCOUNT))
487 return true;
488
489 objcg = get_obj_cgroup_from_current();
490 if (!objcg)
491 return true;
492
493 if (lru) {
494 int ret;
495 struct mem_cgroup *memcg;
496
497 memcg = get_mem_cgroup_from_objcg(objcg);
498 ret = memcg_list_lru_alloc(memcg, lru, flags);
499 css_put(&memcg->css);
500
501 if (ret)
502 goto out;
503 }
504
505 if (obj_cgroup_charge(objcg, flags, objects * obj_full_size(s)))
506 goto out;
507
508 *objcgp = objcg;
509 return true;
510 out:
511 obj_cgroup_put(objcg);
512 return false;
513 }
514
memcg_slab_post_alloc_hook(struct kmem_cache * s,struct obj_cgroup * objcg,gfp_t flags,size_t size,void ** p)515 static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
516 struct obj_cgroup *objcg,
517 gfp_t flags, size_t size,
518 void **p)
519 {
520 struct slab *slab;
521 unsigned long off;
522 size_t i;
523
524 if (!memcg_kmem_enabled() || !objcg)
525 return;
526
527 for (i = 0; i < size; i++) {
528 if (likely(p[i])) {
529 slab = virt_to_slab(p[i]);
530
531 if (!slab_objcgs(slab) &&
532 memcg_alloc_slab_cgroups(slab, s, flags,
533 false)) {
534 obj_cgroup_uncharge(objcg, obj_full_size(s));
535 continue;
536 }
537
538 off = obj_to_index(s, slab, p[i]);
539 obj_cgroup_get(objcg);
540 slab_objcgs(slab)[off] = objcg;
541 mod_objcg_state(objcg, slab_pgdat(slab),
542 cache_vmstat_idx(s), obj_full_size(s));
543 } else {
544 obj_cgroup_uncharge(objcg, obj_full_size(s));
545 }
546 }
547 obj_cgroup_put(objcg);
548 }
549
memcg_slab_free_hook(struct kmem_cache * s_orig,void ** p,int objects)550 static inline void memcg_slab_free_hook(struct kmem_cache *s_orig,
551 void **p, int objects)
552 {
553 struct kmem_cache *s;
554 struct obj_cgroup **objcgs;
555 struct obj_cgroup *objcg;
556 struct slab *slab;
557 unsigned int off;
558 int i;
559
560 if (!memcg_kmem_enabled())
561 return;
562
563 for (i = 0; i < objects; i++) {
564 if (unlikely(!p[i]))
565 continue;
566
567 slab = virt_to_slab(p[i]);
568 /* we could be given a kmalloc_large() object, skip those */
569 if (!slab)
570 continue;
571
572 objcgs = slab_objcgs(slab);
573 if (!objcgs)
574 continue;
575
576 if (!s_orig)
577 s = slab->slab_cache;
578 else
579 s = s_orig;
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,void ** p,int objects)631 static inline void memcg_slab_free_hook(struct kmem_cache *s,
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_enabled() && (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_enabled())
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 #endif /* CONFIG_SLOB */
685
slab_ksize(const struct kmem_cache * s)686 static inline size_t slab_ksize(const struct kmem_cache *s)
687 {
688 #ifndef CONFIG_SLUB
689 return s->object_size;
690
691 #else /* CONFIG_SLUB */
692 # ifdef CONFIG_SLUB_DEBUG
693 /*
694 * Debugging requires use of the padding between object
695 * and whatever may come after it.
696 */
697 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
698 return s->object_size;
699 # endif
700 if (s->flags & SLAB_KASAN)
701 return s->object_size;
702 /*
703 * If we have the need to store the freelist pointer
704 * back there or track user information then we can
705 * only use the space before that information.
706 */
707 if (s->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_STORE_USER))
708 return s->inuse;
709 /*
710 * Else we can use all the padding etc for the allocation
711 */
712 return s->size;
713 #endif
714 }
715
slab_pre_alloc_hook(struct kmem_cache * s,struct list_lru * lru,struct obj_cgroup ** objcgp,size_t size,gfp_t flags)716 static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s,
717 struct list_lru *lru,
718 struct obj_cgroup **objcgp,
719 size_t size, gfp_t flags)
720 {
721 flags &= gfp_allowed_mask;
722
723 might_alloc(flags);
724
725 if (should_failslab(s, flags))
726 return NULL;
727
728 if (!memcg_slab_pre_alloc_hook(s, lru, objcgp, size, flags))
729 return NULL;
730
731 return s;
732 }
733
slab_post_alloc_hook(struct kmem_cache * s,struct obj_cgroup * objcg,gfp_t flags,size_t size,void ** p,bool init)734 static inline void slab_post_alloc_hook(struct kmem_cache *s,
735 struct obj_cgroup *objcg, gfp_t flags,
736 size_t size, void **p, bool init)
737 {
738 size_t i;
739
740 flags &= gfp_allowed_mask;
741
742 /*
743 * As memory initialization might be integrated into KASAN,
744 * kasan_slab_alloc and initialization memset must be
745 * kept together to avoid discrepancies in behavior.
746 *
747 * As p[i] might get tagged, memset and kmemleak hook come after KASAN.
748 */
749 for (i = 0; i < size; i++) {
750 p[i] = kasan_slab_alloc(s, p[i], flags, init);
751 if (p[i] && init && !kasan_has_integrated_init())
752 memset(p[i], 0, s->object_size);
753 kmemleak_alloc_recursive(p[i], s->object_size, 1,
754 s->flags, flags);
755 }
756
757 memcg_slab_post_alloc_hook(s, objcg, flags, size, p);
758 }
759
760 #ifndef CONFIG_SLOB
761 /*
762 * The slab lists for all objects.
763 */
764 struct kmem_cache_node {
765 spinlock_t list_lock;
766
767 #ifdef CONFIG_SLAB
768 struct list_head slabs_partial; /* partial list first, better asm code */
769 struct list_head slabs_full;
770 struct list_head slabs_free;
771 unsigned long total_slabs; /* length of all slab lists */
772 unsigned long free_slabs; /* length of free slab list only */
773 unsigned long free_objects;
774 unsigned int free_limit;
775 unsigned int colour_next; /* Per-node cache coloring */
776 struct array_cache *shared; /* shared per node */
777 struct alien_cache **alien; /* on other nodes */
778 unsigned long next_reap; /* updated without locking */
779 int free_touched; /* updated without locking */
780 #endif
781
782 #ifdef CONFIG_SLUB
783 unsigned long nr_partial;
784 struct list_head partial;
785 #ifdef CONFIG_SLUB_DEBUG
786 atomic_long_t nr_slabs;
787 atomic_long_t total_objects;
788 struct list_head full;
789 #endif
790 #endif
791
792 };
793
get_node(struct kmem_cache * s,int node)794 static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
795 {
796 return s->node[node];
797 }
798
799 /*
800 * Iterator over all nodes. The body will be executed for each node that has
801 * a kmem_cache_node structure allocated (which is true for all online nodes)
802 */
803 #define for_each_kmem_cache_node(__s, __node, __n) \
804 for (__node = 0; __node < nr_node_ids; __node++) \
805 if ((__n = get_node(__s, __node)))
806
807 #endif
808
809 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
810 void dump_unreclaimable_slab(void);
811 #else
dump_unreclaimable_slab(void)812 static inline void dump_unreclaimable_slab(void)
813 {
814 }
815 #endif
816
817 void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr);
818
819 #ifdef CONFIG_SLAB_FREELIST_RANDOM
820 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
821 gfp_t gfp);
822 void cache_random_seq_destroy(struct kmem_cache *cachep);
823 #else
cache_random_seq_create(struct kmem_cache * cachep,unsigned int count,gfp_t gfp)824 static inline int cache_random_seq_create(struct kmem_cache *cachep,
825 unsigned int count, gfp_t gfp)
826 {
827 return 0;
828 }
cache_random_seq_destroy(struct kmem_cache * cachep)829 static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { }
830 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
831
slab_want_init_on_alloc(gfp_t flags,struct kmem_cache * c)832 static inline bool slab_want_init_on_alloc(gfp_t flags, struct kmem_cache *c)
833 {
834 if (static_branch_maybe(CONFIG_INIT_ON_ALLOC_DEFAULT_ON,
835 &init_on_alloc)) {
836 if (c->ctor)
837 return false;
838 if (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON))
839 return flags & __GFP_ZERO;
840 return true;
841 }
842 return flags & __GFP_ZERO;
843 }
844
slab_want_init_on_free(struct kmem_cache * c)845 static inline bool slab_want_init_on_free(struct kmem_cache *c)
846 {
847 if (static_branch_maybe(CONFIG_INIT_ON_FREE_DEFAULT_ON,
848 &init_on_free))
849 return !(c->ctor ||
850 (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)));
851 return false;
852 }
853
854 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG)
855 void debugfs_slab_release(struct kmem_cache *);
856 #else
debugfs_slab_release(struct kmem_cache * s)857 static inline void debugfs_slab_release(struct kmem_cache *s) { }
858 #endif
859
860 #ifdef CONFIG_PRINTK
861 #define KS_ADDRS_COUNT 16
862 struct kmem_obj_info {
863 void *kp_ptr;
864 struct slab *kp_slab;
865 void *kp_objp;
866 unsigned long kp_data_offset;
867 struct kmem_cache *kp_slab_cache;
868 void *kp_ret;
869 void *kp_stack[KS_ADDRS_COUNT];
870 void *kp_free_stack[KS_ADDRS_COUNT];
871 };
872 void __kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab);
873 #endif
874
875 #ifdef CONFIG_HAVE_HARDENED_USERCOPY_ALLOCATOR
876 void __check_heap_object(const void *ptr, unsigned long n,
877 const struct slab *slab, bool to_user);
878 #else
879 static inline
__check_heap_object(const void * ptr,unsigned long n,const struct slab * slab,bool to_user)880 void __check_heap_object(const void *ptr, unsigned long n,
881 const struct slab *slab, bool to_user)
882 {
883 }
884 #endif
885
886 #endif /* MM_SLAB_H */
887