1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef MM_SLAB_H
3 #define MM_SLAB_H
4 /*
5 * Internal slab definitions
6 */
7 void __init kmem_cache_init(void);
8
9 #ifdef CONFIG_64BIT
10 # ifdef system_has_cmpxchg128
11 # define system_has_freelist_aba() system_has_cmpxchg128()
12 # define try_cmpxchg_freelist try_cmpxchg128
13 # endif
14 #define this_cpu_try_cmpxchg_freelist this_cpu_try_cmpxchg128
15 typedef u128 freelist_full_t;
16 #else /* CONFIG_64BIT */
17 # ifdef system_has_cmpxchg64
18 # define system_has_freelist_aba() system_has_cmpxchg64()
19 # define try_cmpxchg_freelist try_cmpxchg64
20 # endif
21 #define this_cpu_try_cmpxchg_freelist this_cpu_try_cmpxchg64
22 typedef u64 freelist_full_t;
23 #endif /* CONFIG_64BIT */
24
25 #if defined(system_has_freelist_aba) && !defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
26 #undef system_has_freelist_aba
27 #endif
28
29 /*
30 * Freelist pointer and counter to cmpxchg together, avoids the typical ABA
31 * problems with cmpxchg of just a pointer.
32 */
33 typedef union {
34 struct {
35 void *freelist;
36 unsigned long counter;
37 };
38 freelist_full_t full;
39 } freelist_aba_t;
40
41 /* Reuses the bits in struct page */
42 struct slab {
43 unsigned long __page_flags;
44
45 #if defined(CONFIG_SLAB)
46
47 struct kmem_cache *slab_cache;
48 union {
49 struct {
50 struct list_head slab_list;
51 void *freelist; /* array of free object indexes */
52 void *s_mem; /* first object */
53 };
54 struct rcu_head rcu_head;
55 };
56 unsigned int active;
57
58 #elif defined(CONFIG_SLUB)
59
60 struct kmem_cache *slab_cache;
61 union {
62 struct {
63 union {
64 struct list_head slab_list;
65 #ifdef CONFIG_SLUB_CPU_PARTIAL
66 struct {
67 struct slab *next;
68 int slabs; /* Nr of slabs left */
69 };
70 #endif
71 };
72 /* Double-word boundary */
73 union {
74 struct {
75 void *freelist; /* first free object */
76 union {
77 unsigned long counters;
78 struct {
79 unsigned inuse:16;
80 unsigned objects:15;
81 unsigned frozen:1;
82 };
83 };
84 };
85 #ifdef system_has_freelist_aba
86 freelist_aba_t freelist_counter;
87 #endif
88 };
89 };
90 struct rcu_head rcu_head;
91 };
92 unsigned int __unused;
93
94 #else
95 #error "Unexpected slab allocator configured"
96 #endif
97
98 atomic_t __page_refcount;
99 #ifdef CONFIG_MEMCG
100 unsigned long memcg_data;
101 #endif
102 };
103
104 #define SLAB_MATCH(pg, sl) \
105 static_assert(offsetof(struct page, pg) == offsetof(struct slab, sl))
106 SLAB_MATCH(flags, __page_flags);
107 SLAB_MATCH(compound_head, slab_cache); /* Ensure bit 0 is clear */
108 SLAB_MATCH(_refcount, __page_refcount);
109 #ifdef CONFIG_MEMCG
110 SLAB_MATCH(memcg_data, memcg_data);
111 #endif
112 #undef SLAB_MATCH
113 static_assert(sizeof(struct slab) <= sizeof(struct page));
114 #if defined(system_has_freelist_aba) && defined(CONFIG_SLUB)
115 static_assert(IS_ALIGNED(offsetof(struct slab, freelist), sizeof(freelist_aba_t)));
116 #endif
117
118 /**
119 * folio_slab - Converts from folio to slab.
120 * @folio: The folio.
121 *
122 * Currently struct slab is a different representation of a folio where
123 * folio_test_slab() is true.
124 *
125 * Return: The slab which contains this folio.
126 */
127 #define folio_slab(folio) (_Generic((folio), \
128 const struct folio *: (const struct slab *)(folio), \
129 struct folio *: (struct slab *)(folio)))
130
131 /**
132 * slab_folio - The folio allocated for a slab
133 * @slab: The slab.
134 *
135 * Slabs are allocated as folios that contain the individual objects and are
136 * using some fields in the first struct page of the folio - those fields are
137 * now accessed by struct slab. It is occasionally necessary to convert back to
138 * a folio in order to communicate with the rest of the mm. Please use this
139 * helper function instead of casting yourself, as the implementation may change
140 * in the future.
141 */
142 #define slab_folio(s) (_Generic((s), \
143 const struct slab *: (const struct folio *)s, \
144 struct slab *: (struct folio *)s))
145
146 /**
147 * page_slab - Converts from first struct page to slab.
148 * @p: The first (either head of compound or single) page of slab.
149 *
150 * A temporary wrapper to convert struct page to struct slab in situations where
151 * we know the page is the compound head, or single order-0 page.
152 *
153 * Long-term ideally everything would work with struct slab directly or go
154 * through folio to struct slab.
155 *
156 * Return: The slab which contains this page
157 */
158 #define page_slab(p) (_Generic((p), \
159 const struct page *: (const struct slab *)(p), \
160 struct page *: (struct slab *)(p)))
161
162 /**
163 * slab_page - The first struct page allocated for a slab
164 * @slab: The slab.
165 *
166 * A convenience wrapper for converting slab to the first struct page of the
167 * underlying folio, to communicate with code not yet converted to folio or
168 * struct slab.
169 */
170 #define slab_page(s) folio_page(slab_folio(s), 0)
171
172 /*
173 * If network-based swap is enabled, sl*b must keep track of whether pages
174 * were allocated from pfmemalloc reserves.
175 */
slab_test_pfmemalloc(const struct slab * slab)176 static inline bool slab_test_pfmemalloc(const struct slab *slab)
177 {
178 return folio_test_active((struct folio *)slab_folio(slab));
179 }
180
slab_set_pfmemalloc(struct slab * slab)181 static inline void slab_set_pfmemalloc(struct slab *slab)
182 {
183 folio_set_active(slab_folio(slab));
184 }
185
slab_clear_pfmemalloc(struct slab * slab)186 static inline void slab_clear_pfmemalloc(struct slab *slab)
187 {
188 folio_clear_active(slab_folio(slab));
189 }
190
__slab_clear_pfmemalloc(struct slab * slab)191 static inline void __slab_clear_pfmemalloc(struct slab *slab)
192 {
193 __folio_clear_active(slab_folio(slab));
194 }
195
slab_address(const struct slab * slab)196 static inline void *slab_address(const struct slab *slab)
197 {
198 return folio_address(slab_folio(slab));
199 }
200
slab_nid(const struct slab * slab)201 static inline int slab_nid(const struct slab *slab)
202 {
203 return folio_nid(slab_folio(slab));
204 }
205
slab_pgdat(const struct slab * slab)206 static inline pg_data_t *slab_pgdat(const struct slab *slab)
207 {
208 return folio_pgdat(slab_folio(slab));
209 }
210
virt_to_slab(const void * addr)211 static inline struct slab *virt_to_slab(const void *addr)
212 {
213 struct folio *folio = virt_to_folio(addr);
214
215 if (!folio_test_slab(folio))
216 return NULL;
217
218 return folio_slab(folio);
219 }
220
slab_order(const struct slab * slab)221 static inline int slab_order(const struct slab *slab)
222 {
223 return folio_order((struct folio *)slab_folio(slab));
224 }
225
slab_size(const struct slab * slab)226 static inline size_t slab_size(const struct slab *slab)
227 {
228 return PAGE_SIZE << slab_order(slab);
229 }
230
231 #ifdef CONFIG_SLAB
232 #include <linux/slab_def.h>
233 #endif
234
235 #ifdef CONFIG_SLUB
236 #include <linux/slub_def.h>
237 #endif
238
239 #include <linux/memcontrol.h>
240 #include <linux/fault-inject.h>
241 #include <linux/kasan.h>
242 #include <linux/kmemleak.h>
243 #include <linux/random.h>
244 #include <linux/sched/mm.h>
245 #include <linux/list_lru.h>
246
247 /*
248 * State of the slab allocator.
249 *
250 * This is used to describe the states of the allocator during bootup.
251 * Allocators use this to gradually bootstrap themselves. Most allocators
252 * have the problem that the structures used for managing slab caches are
253 * allocated from slab caches themselves.
254 */
255 enum slab_state {
256 DOWN, /* No slab functionality yet */
257 PARTIAL, /* SLUB: kmem_cache_node available */
258 PARTIAL_NODE, /* SLAB: kmalloc size for node struct available */
259 UP, /* Slab caches usable but not all extras yet */
260 FULL /* Everything is working */
261 };
262
263 extern enum slab_state slab_state;
264
265 /* The slab cache mutex protects the management structures during changes */
266 extern struct mutex slab_mutex;
267
268 /* The list of all slab caches on the system */
269 extern struct list_head slab_caches;
270
271 /* The slab cache that manages slab cache information */
272 extern struct kmem_cache *kmem_cache;
273
274 /* A table of kmalloc cache names and sizes */
275 extern const struct kmalloc_info_struct {
276 const char *name[NR_KMALLOC_TYPES];
277 unsigned int size;
278 } kmalloc_info[];
279
280 /* Kmalloc array related functions */
281 void setup_kmalloc_cache_index_table(void);
282 void create_kmalloc_caches(slab_flags_t);
283
284 /* Find the kmalloc slab corresponding for a certain size */
285 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags, unsigned long caller);
286
287 void *__kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags,
288 int node, size_t orig_size,
289 unsigned long caller);
290 void __kmem_cache_free(struct kmem_cache *s, void *x, unsigned long caller);
291
292 gfp_t kmalloc_fix_flags(gfp_t flags);
293
294 /* Functions provided by the slab allocators */
295 int __kmem_cache_create(struct kmem_cache *, slab_flags_t flags);
296
297 void __init new_kmalloc_cache(int idx, enum kmalloc_cache_type type,
298 slab_flags_t flags);
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 struct kmem_cache *
307 __kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
308 slab_flags_t flags, void (*ctor)(void *));
309
310 slab_flags_t kmem_cache_flags(unsigned int object_size,
311 slab_flags_t flags, const char *name);
312
is_kmalloc_cache(struct kmem_cache * s)313 static inline bool is_kmalloc_cache(struct kmem_cache *s)
314 {
315 return (s->flags & SLAB_KMALLOC);
316 }
317
318 /* Legal flag mask for kmem_cache_create(), for various configurations */
319 #define SLAB_CORE_FLAGS (SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA | \
320 SLAB_CACHE_DMA32 | SLAB_PANIC | \
321 SLAB_TYPESAFE_BY_RCU | SLAB_DEBUG_OBJECTS )
322
323 #if defined(CONFIG_DEBUG_SLAB)
324 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
325 #elif defined(CONFIG_SLUB_DEBUG)
326 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
327 SLAB_TRACE | SLAB_CONSISTENCY_CHECKS)
328 #else
329 #define SLAB_DEBUG_FLAGS (0)
330 #endif
331
332 #if defined(CONFIG_SLAB)
333 #define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \
334 SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \
335 SLAB_ACCOUNT | SLAB_NO_MERGE)
336 #elif defined(CONFIG_SLUB)
337 #define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
338 SLAB_TEMPORARY | SLAB_ACCOUNT | \
339 SLAB_NO_USER_FLAGS | SLAB_KMALLOC | SLAB_NO_MERGE)
340 #else
341 #define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE)
342 #endif
343
344 /* Common flags available with current configuration */
345 #define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
346
347 /* Common flags permitted for kmem_cache_create */
348 #define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \
349 SLAB_RED_ZONE | \
350 SLAB_POISON | \
351 SLAB_STORE_USER | \
352 SLAB_TRACE | \
353 SLAB_CONSISTENCY_CHECKS | \
354 SLAB_MEM_SPREAD | \
355 SLAB_NOLEAKTRACE | \
356 SLAB_RECLAIM_ACCOUNT | \
357 SLAB_TEMPORARY | \
358 SLAB_ACCOUNT | \
359 SLAB_KMALLOC | \
360 SLAB_NO_MERGE | \
361 SLAB_NO_USER_FLAGS)
362
363 bool __kmem_cache_empty(struct kmem_cache *);
364 int __kmem_cache_shutdown(struct kmem_cache *);
365 void __kmem_cache_release(struct kmem_cache *);
366 int __kmem_cache_shrink(struct kmem_cache *);
367 void slab_kmem_cache_release(struct kmem_cache *);
368
369 struct seq_file;
370 struct file;
371
372 struct slabinfo {
373 unsigned long active_objs;
374 unsigned long num_objs;
375 unsigned long active_slabs;
376 unsigned long num_slabs;
377 unsigned long shared_avail;
378 unsigned int limit;
379 unsigned int batchcount;
380 unsigned int shared;
381 unsigned int objects_per_slab;
382 unsigned int cache_order;
383 };
384
385 void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo);
386 void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s);
387 ssize_t slabinfo_write(struct file *file, const char __user *buffer,
388 size_t count, loff_t *ppos);
389
cache_vmstat_idx(struct kmem_cache * s)390 static inline enum node_stat_item cache_vmstat_idx(struct kmem_cache *s)
391 {
392 return (s->flags & SLAB_RECLAIM_ACCOUNT) ?
393 NR_SLAB_RECLAIMABLE_B : NR_SLAB_UNRECLAIMABLE_B;
394 }
395
396 #ifdef CONFIG_SLUB_DEBUG
397 #ifdef CONFIG_SLUB_DEBUG_ON
398 DECLARE_STATIC_KEY_TRUE(slub_debug_enabled);
399 #else
400 DECLARE_STATIC_KEY_FALSE(slub_debug_enabled);
401 #endif
402 extern void print_tracking(struct kmem_cache *s, void *object);
403 long validate_slab_cache(struct kmem_cache *s);
__slub_debug_enabled(void)404 static inline bool __slub_debug_enabled(void)
405 {
406 return static_branch_unlikely(&slub_debug_enabled);
407 }
408 #else
print_tracking(struct kmem_cache * s,void * object)409 static inline void print_tracking(struct kmem_cache *s, void *object)
410 {
411 }
__slub_debug_enabled(void)412 static inline bool __slub_debug_enabled(void)
413 {
414 return false;
415 }
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 */
kmem_cache_debug_flags(struct kmem_cache * s,slab_flags_t flags)423 static inline bool kmem_cache_debug_flags(struct kmem_cache *s, slab_flags_t flags)
424 {
425 if (IS_ENABLED(CONFIG_SLUB_DEBUG))
426 VM_WARN_ON_ONCE(!(flags & SLAB_DEBUG_FLAGS));
427 if (__slub_debug_enabled())
428 return s->flags & flags;
429 return false;
430 }
431
432 #ifdef CONFIG_MEMCG_KMEM
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 */
slab_objcgs(struct slab * slab)440 static 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
451 int memcg_alloc_slab_cgroups(struct slab *slab, struct kmem_cache *s,
452 gfp_t gfp, bool new_slab);
453 void mod_objcg_state(struct obj_cgroup *objcg, struct pglist_data *pgdat,
454 enum node_stat_item idx, int nr);
455
memcg_free_slab_cgroups(struct slab * slab)456 static inline void memcg_free_slab_cgroups(struct slab *slab)
457 {
458 kfree(slab_objcgs(slab));
459 slab->memcg_data = 0;
460 }
461
obj_full_size(struct kmem_cache * s)462 static 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
471 /*
472 * Returns false if the allocation should fail.
473 */
memcg_slab_pre_alloc_hook(struct kmem_cache * s,struct list_lru * lru,struct obj_cgroup ** objcgp,size_t objects,gfp_t flags)474 static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s,
475 struct list_lru *lru,
476 struct obj_cgroup **objcgp,
477 size_t objects, gfp_t flags)
478 {
479 struct obj_cgroup *objcg;
480
481 if (!memcg_kmem_online())
482 return true;
483
484 if (!(flags & __GFP_ACCOUNT) && !(s->flags & SLAB_ACCOUNT))
485 return true;
486
487 objcg = get_obj_cgroup_from_current();
488 if (!objcg)
489 return true;
490
491 if (lru) {
492 int ret;
493 struct mem_cgroup *memcg;
494
495 memcg = get_mem_cgroup_from_objcg(objcg);
496 ret = memcg_list_lru_alloc(memcg, lru, flags);
497 css_put(&memcg->css);
498
499 if (ret)
500 goto out;
501 }
502
503 if (obj_cgroup_charge(objcg, flags, objects * obj_full_size(s)))
504 goto out;
505
506 *objcgp = objcg;
507 return true;
508 out:
509 obj_cgroup_put(objcg);
510 return false;
511 }
512
memcg_slab_post_alloc_hook(struct kmem_cache * s,struct obj_cgroup * objcg,gfp_t flags,size_t size,void ** p)513 static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
514 struct obj_cgroup *objcg,
515 gfp_t flags, size_t size,
516 void **p)
517 {
518 struct slab *slab;
519 unsigned long off;
520 size_t i;
521
522 if (!memcg_kmem_online() || !objcg)
523 return;
524
525 for (i = 0; i < size; i++) {
526 if (likely(p[i])) {
527 slab = virt_to_slab(p[i]);
528
529 if (!slab_objcgs(slab) &&
530 memcg_alloc_slab_cgroups(slab, s, flags,
531 false)) {
532 obj_cgroup_uncharge(objcg, obj_full_size(s));
533 continue;
534 }
535
536 off = obj_to_index(s, slab, p[i]);
537 obj_cgroup_get(objcg);
538 slab_objcgs(slab)[off] = objcg;
539 mod_objcg_state(objcg, slab_pgdat(slab),
540 cache_vmstat_idx(s), obj_full_size(s));
541 } else {
542 obj_cgroup_uncharge(objcg, obj_full_size(s));
543 }
544 }
545 obj_cgroup_put(objcg);
546 }
547
memcg_slab_free_hook(struct kmem_cache * s,struct slab * slab,void ** p,int objects)548 static inline void memcg_slab_free_hook(struct kmem_cache *s, struct slab *slab,
549 void **p, int objects)
550 {
551 struct obj_cgroup **objcgs;
552 int i;
553
554 if (!memcg_kmem_online())
555 return;
556
557 objcgs = slab_objcgs(slab);
558 if (!objcgs)
559 return;
560
561 for (i = 0; i < objects; i++) {
562 struct obj_cgroup *objcg;
563 unsigned int off;
564
565 off = obj_to_index(s, slab, p[i]);
566 objcg = objcgs[off];
567 if (!objcg)
568 continue;
569
570 objcgs[off] = NULL;
571 obj_cgroup_uncharge(objcg, obj_full_size(s));
572 mod_objcg_state(objcg, slab_pgdat(slab), cache_vmstat_idx(s),
573 -obj_full_size(s));
574 obj_cgroup_put(objcg);
575 }
576 }
577
578 #else /* CONFIG_MEMCG_KMEM */
slab_objcgs(struct slab * slab)579 static inline struct obj_cgroup **slab_objcgs(struct slab *slab)
580 {
581 return NULL;
582 }
583
memcg_from_slab_obj(void * ptr)584 static inline struct mem_cgroup *memcg_from_slab_obj(void *ptr)
585 {
586 return NULL;
587 }
588
memcg_alloc_slab_cgroups(struct slab * slab,struct kmem_cache * s,gfp_t gfp,bool new_slab)589 static inline int memcg_alloc_slab_cgroups(struct slab *slab,
590 struct kmem_cache *s, gfp_t gfp,
591 bool new_slab)
592 {
593 return 0;
594 }
595
memcg_free_slab_cgroups(struct slab * slab)596 static inline void memcg_free_slab_cgroups(struct slab *slab)
597 {
598 }
599
memcg_slab_pre_alloc_hook(struct kmem_cache * s,struct list_lru * lru,struct obj_cgroup ** objcgp,size_t objects,gfp_t flags)600 static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s,
601 struct list_lru *lru,
602 struct obj_cgroup **objcgp,
603 size_t objects, gfp_t flags)
604 {
605 return true;
606 }
607
memcg_slab_post_alloc_hook(struct kmem_cache * s,struct obj_cgroup * objcg,gfp_t flags,size_t size,void ** p)608 static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
609 struct obj_cgroup *objcg,
610 gfp_t flags, size_t size,
611 void **p)
612 {
613 }
614
memcg_slab_free_hook(struct kmem_cache * s,struct slab * slab,void ** p,int objects)615 static inline void memcg_slab_free_hook(struct kmem_cache *s, struct slab *slab,
616 void **p, int objects)
617 {
618 }
619 #endif /* CONFIG_MEMCG_KMEM */
620
virt_to_cache(const void * obj)621 static inline struct kmem_cache *virt_to_cache(const void *obj)
622 {
623 struct slab *slab;
624
625 slab = virt_to_slab(obj);
626 if (WARN_ONCE(!slab, "%s: Object is not a Slab page!\n",
627 __func__))
628 return NULL;
629 return slab->slab_cache;
630 }
631
account_slab(struct slab * slab,int order,struct kmem_cache * s,gfp_t gfp)632 static __always_inline void account_slab(struct slab *slab, int order,
633 struct kmem_cache *s, gfp_t gfp)
634 {
635 if (memcg_kmem_online() && (s->flags & SLAB_ACCOUNT))
636 memcg_alloc_slab_cgroups(slab, s, gfp, true);
637
638 mod_node_page_state(slab_pgdat(slab), cache_vmstat_idx(s),
639 PAGE_SIZE << order);
640 }
641
unaccount_slab(struct slab * slab,int order,struct kmem_cache * s)642 static __always_inline void unaccount_slab(struct slab *slab, int order,
643 struct kmem_cache *s)
644 {
645 if (memcg_kmem_online())
646 memcg_free_slab_cgroups(slab);
647
648 mod_node_page_state(slab_pgdat(slab), cache_vmstat_idx(s),
649 -(PAGE_SIZE << order));
650 }
651
cache_from_obj(struct kmem_cache * s,void * x)652 static 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) &&
657 !kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS))
658 return s;
659
660 cachep = virt_to_cache(x);
661 if (WARN(cachep && cachep != s,
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 }
667
668 void free_large_kmalloc(struct folio *folio, void *object);
669
670 size_t __ksize(const void *objp);
671
slab_ksize(const struct kmem_cache * s)672 static inline size_t slab_ksize(const struct kmem_cache *s)
673 {
674 #ifndef CONFIG_SLUB
675 return s->object_size;
676
677 #else /* CONFIG_SLUB */
678 # ifdef CONFIG_SLUB_DEBUG
679 /*
680 * Debugging requires use of the padding between object
681 * and whatever may come after it.
682 */
683 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
684 return s->object_size;
685 # endif
686 if (s->flags & SLAB_KASAN)
687 return s->object_size;
688 /*
689 * If we have the need to store the freelist pointer
690 * back there or track user information then we can
691 * only use the space before that information.
692 */
693 if (s->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_STORE_USER))
694 return s->inuse;
695 /*
696 * Else we can use all the padding etc for the allocation
697 */
698 return s->size;
699 #endif
700 }
701
slab_pre_alloc_hook(struct kmem_cache * s,struct list_lru * lru,struct obj_cgroup ** objcgp,size_t size,gfp_t flags)702 static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s,
703 struct list_lru *lru,
704 struct obj_cgroup **objcgp,
705 size_t size, gfp_t flags)
706 {
707 flags &= gfp_allowed_mask;
708
709 might_alloc(flags);
710
711 if (should_failslab(s, flags))
712 return NULL;
713
714 if (!memcg_slab_pre_alloc_hook(s, lru, objcgp, size, flags))
715 return NULL;
716
717 return s;
718 }
719
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)720 static inline void slab_post_alloc_hook(struct kmem_cache *s,
721 struct obj_cgroup *objcg, gfp_t flags,
722 size_t size, void **p, bool init,
723 unsigned int orig_size)
724 {
725 unsigned int zero_size = s->object_size;
726 bool kasan_init = init;
727 size_t i;
728
729 flags &= gfp_allowed_mask;
730
731 /*
732 * For kmalloc object, the allocated memory size(object_size) is likely
733 * larger than the requested size(orig_size). If redzone check is
734 * enabled for the extra space, don't zero it, as it will be redzoned
735 * soon. The redzone operation for this extra space could be seen as a
736 * replacement of current poisoning under certain debug option, and
737 * won't break other sanity checks.
738 */
739 if (kmem_cache_debug_flags(s, SLAB_STORE_USER | SLAB_RED_ZONE) &&
740 (s->flags & SLAB_KMALLOC))
741 zero_size = orig_size;
742
743 /*
744 * When slub_debug is enabled, avoid memory initialization integrated
745 * into KASAN and instead zero out the memory via the memset below with
746 * the proper size. Otherwise, KASAN might overwrite SLUB redzones and
747 * cause false-positive reports. This does not lead to a performance
748 * penalty on production builds, as slub_debug is not intended to be
749 * enabled there.
750 */
751 if (__slub_debug_enabled())
752 kasan_init = false;
753
754 /*
755 * As memory initialization might be integrated into KASAN,
756 * kasan_slab_alloc and initialization memset must be
757 * kept together to avoid discrepancies in behavior.
758 *
759 * As p[i] might get tagged, memset and kmemleak hook come after KASAN.
760 */
761 for (i = 0; i < size; i++) {
762 p[i] = kasan_slab_alloc(s, p[i], flags, kasan_init);
763 if (p[i] && init && (!kasan_init || !kasan_has_integrated_init()))
764 memset(p[i], 0, zero_size);
765 kmemleak_alloc_recursive(p[i], s->object_size, 1,
766 s->flags, flags);
767 kmsan_slab_alloc(s, p[i], flags);
768 }
769
770 memcg_slab_post_alloc_hook(s, objcg, flags, size, p);
771 }
772
773 /*
774 * The slab lists for all objects.
775 */
776 struct kmem_cache_node {
777 #ifdef CONFIG_SLAB
778 raw_spinlock_t list_lock;
779 struct list_head slabs_partial; /* partial list first, better asm code */
780 struct list_head slabs_full;
781 struct list_head slabs_free;
782 unsigned long total_slabs; /* length of all slab lists */
783 unsigned long free_slabs; /* length of free slab list only */
784 unsigned long free_objects;
785 unsigned int free_limit;
786 unsigned int colour_next; /* Per-node cache coloring */
787 struct array_cache *shared; /* shared per node */
788 struct alien_cache **alien; /* on other nodes */
789 unsigned long next_reap; /* updated without locking */
790 int free_touched; /* updated without locking */
791 #endif
792
793 #ifdef CONFIG_SLUB
794 spinlock_t list_lock;
795 unsigned long nr_partial;
796 struct list_head partial;
797 #ifdef CONFIG_SLUB_DEBUG
798 atomic_long_t nr_slabs;
799 atomic_long_t total_objects;
800 struct list_head full;
801 #endif
802 #endif
803
804 };
805
get_node(struct kmem_cache * s,int node)806 static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
807 {
808 return s->node[node];
809 }
810
811 /*
812 * Iterator over all nodes. The body will be executed for each node that has
813 * a kmem_cache_node structure allocated (which is true for all online nodes)
814 */
815 #define for_each_kmem_cache_node(__s, __node, __n) \
816 for (__node = 0; __node < nr_node_ids; __node++) \
817 if ((__n = get_node(__s, __node)))
818
819
820 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
821 void dump_unreclaimable_slab(void);
822 #else
dump_unreclaimable_slab(void)823 static inline void dump_unreclaimable_slab(void)
824 {
825 }
826 #endif
827
828 void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr);
829
830 #ifdef CONFIG_SLAB_FREELIST_RANDOM
831 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
832 gfp_t gfp);
833 void cache_random_seq_destroy(struct kmem_cache *cachep);
834 #else
cache_random_seq_create(struct kmem_cache * cachep,unsigned int count,gfp_t gfp)835 static inline int cache_random_seq_create(struct kmem_cache *cachep,
836 unsigned int count, gfp_t gfp)
837 {
838 return 0;
839 }
cache_random_seq_destroy(struct kmem_cache * cachep)840 static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { }
841 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
842
slab_want_init_on_alloc(gfp_t flags,struct kmem_cache * c)843 static inline bool slab_want_init_on_alloc(gfp_t flags, struct kmem_cache *c)
844 {
845 if (static_branch_maybe(CONFIG_INIT_ON_ALLOC_DEFAULT_ON,
846 &init_on_alloc)) {
847 if (c->ctor)
848 return false;
849 if (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON))
850 return flags & __GFP_ZERO;
851 return true;
852 }
853 return flags & __GFP_ZERO;
854 }
855
slab_want_init_on_free(struct kmem_cache * c)856 static inline bool slab_want_init_on_free(struct kmem_cache *c)
857 {
858 if (static_branch_maybe(CONFIG_INIT_ON_FREE_DEFAULT_ON,
859 &init_on_free))
860 return !(c->ctor ||
861 (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)));
862 return false;
863 }
864
865 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG)
866 void debugfs_slab_release(struct kmem_cache *);
867 #else
debugfs_slab_release(struct kmem_cache * s)868 static inline void debugfs_slab_release(struct kmem_cache *s) { }
869 #endif
870
871 #ifdef CONFIG_PRINTK
872 #define KS_ADDRS_COUNT 16
873 struct kmem_obj_info {
874 void *kp_ptr;
875 struct slab *kp_slab;
876 void *kp_objp;
877 unsigned long kp_data_offset;
878 struct kmem_cache *kp_slab_cache;
879 void *kp_ret;
880 void *kp_stack[KS_ADDRS_COUNT];
881 void *kp_free_stack[KS_ADDRS_COUNT];
882 };
883 void __kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab);
884 #endif
885
886 void __check_heap_object(const void *ptr, unsigned long n,
887 const struct slab *slab, bool to_user);
888
889 #ifdef CONFIG_SLUB_DEBUG
890 void skip_orig_size_check(struct kmem_cache *s, const void *object);
891 #endif
892
893 #endif /* MM_SLAB_H */
894