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 #ifdef CONFIG_SLUB
295 /*
296 * Tracking user of a slab.
297 */
298 #define TRACK_ADDRS_COUNT 16
299 struct track {
300 unsigned long addr; /* Called from address */
301 #ifdef CONFIG_STACKDEPOT
302 depot_stack_handle_t handle;
303 #endif
304 int cpu; /* Was running on cpu */
305 int pid; /* Pid context */
306 unsigned long when; /* When did the operation occur */
307 };
308
309 enum track_item { TRACK_ALLOC, TRACK_FREE };
310 #endif
311
312 /* Functions provided by the slab allocators */
313 int __kmem_cache_create(struct kmem_cache *, slab_flags_t flags);
314
315 void __init new_kmalloc_cache(int idx, enum kmalloc_cache_type type,
316 slab_flags_t flags);
317 extern void create_boot_cache(struct kmem_cache *, const char *name,
318 unsigned int size, slab_flags_t flags,
319 unsigned int useroffset, unsigned int usersize);
320
321 int slab_unmergeable(struct kmem_cache *s);
322 struct kmem_cache *find_mergeable(unsigned size, unsigned align,
323 slab_flags_t flags, const char *name, void (*ctor)(void *));
324 struct kmem_cache *
325 __kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
326 slab_flags_t flags, void (*ctor)(void *));
327
328 slab_flags_t kmem_cache_flags(unsigned int object_size,
329 slab_flags_t flags, const char *name);
330
is_kmalloc_cache(struct kmem_cache * s)331 static inline bool is_kmalloc_cache(struct kmem_cache *s)
332 {
333 return (s->flags & SLAB_KMALLOC);
334 }
335
336 /* Legal flag mask for kmem_cache_create(), for various configurations */
337 #define SLAB_CORE_FLAGS (SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA | \
338 SLAB_CACHE_DMA32 | SLAB_PANIC | \
339 SLAB_TYPESAFE_BY_RCU | SLAB_DEBUG_OBJECTS )
340
341 #if defined(CONFIG_DEBUG_SLAB)
342 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
343 #elif defined(CONFIG_SLUB_DEBUG)
344 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
345 SLAB_TRACE | SLAB_CONSISTENCY_CHECKS)
346 #else
347 #define SLAB_DEBUG_FLAGS (0)
348 #endif
349
350 #if defined(CONFIG_SLAB)
351 #define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \
352 SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \
353 SLAB_ACCOUNT | SLAB_NO_MERGE)
354 #elif defined(CONFIG_SLUB)
355 #define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
356 SLAB_TEMPORARY | SLAB_ACCOUNT | \
357 SLAB_NO_USER_FLAGS | SLAB_KMALLOC | SLAB_NO_MERGE)
358 #else
359 #define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE)
360 #endif
361
362 /* Common flags available with current configuration */
363 #define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
364
365 /* Common flags permitted for kmem_cache_create */
366 #define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \
367 SLAB_RED_ZONE | \
368 SLAB_POISON | \
369 SLAB_STORE_USER | \
370 SLAB_TRACE | \
371 SLAB_CONSISTENCY_CHECKS | \
372 SLAB_MEM_SPREAD | \
373 SLAB_NOLEAKTRACE | \
374 SLAB_RECLAIM_ACCOUNT | \
375 SLAB_TEMPORARY | \
376 SLAB_ACCOUNT | \
377 SLAB_KMALLOC | \
378 SLAB_NO_MERGE | \
379 SLAB_NO_USER_FLAGS)
380
381 bool __kmem_cache_empty(struct kmem_cache *);
382 int __kmem_cache_shutdown(struct kmem_cache *);
383 void __kmem_cache_release(struct kmem_cache *);
384 int __kmem_cache_shrink(struct kmem_cache *);
385 void slab_kmem_cache_release(struct kmem_cache *);
386
387 struct seq_file;
388 struct file;
389
390 struct slabinfo {
391 unsigned long active_objs;
392 unsigned long num_objs;
393 unsigned long active_slabs;
394 unsigned long num_slabs;
395 unsigned long shared_avail;
396 unsigned int limit;
397 unsigned int batchcount;
398 unsigned int shared;
399 unsigned int objects_per_slab;
400 unsigned int cache_order;
401 };
402
403 void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo);
404 void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s);
405 ssize_t slabinfo_write(struct file *file, const char __user *buffer,
406 size_t count, loff_t *ppos);
407
cache_vmstat_idx(struct kmem_cache * s)408 static inline enum node_stat_item cache_vmstat_idx(struct kmem_cache *s)
409 {
410 return (s->flags & SLAB_RECLAIM_ACCOUNT) ?
411 NR_SLAB_RECLAIMABLE_B : NR_SLAB_UNRECLAIMABLE_B;
412 }
413
414 #ifdef CONFIG_SLUB_DEBUG
415 #ifdef CONFIG_SLUB_DEBUG_ON
416 DECLARE_STATIC_KEY_TRUE(slub_debug_enabled);
417 #else
418 DECLARE_STATIC_KEY_FALSE(slub_debug_enabled);
419 #endif
420 extern void print_tracking(struct kmem_cache *s, void *object);
421 long validate_slab_cache(struct kmem_cache *s);
422 extern unsigned long get_each_object_track(struct kmem_cache *s,
423 struct slab *slab, enum track_item alloc,
424 int (*fn)(const struct kmem_cache *, const void *,
425 const struct track *, void *), void *private);
__slub_debug_enabled(void)426 static inline bool __slub_debug_enabled(void)
427 {
428 return static_branch_unlikely(&slub_debug_enabled);
429 }
430 #else
print_tracking(struct kmem_cache * s,void * object)431 static inline void print_tracking(struct kmem_cache *s, void *object)
432 {
433 }
434 #ifdef CONFIG_SLUB
get_each_object_track(struct kmem_cache * s,struct page * page,enum track_item alloc,int (* fn)(const struct kmem_cache *,const void *,const struct track *,void *),void * private)435 static inline unsigned long get_each_object_track(struct kmem_cache *s,
436 struct page *page, enum track_item alloc,
437 int (*fn)(const struct kmem_cache *, const void *,
438 const struct track *, void *), void *private)
439 {
440 return 0;
441 }
442 #endif
__slub_debug_enabled(void)443 static inline bool __slub_debug_enabled(void)
444 {
445 return false;
446 }
447 #endif
448
449 /*
450 * Returns true if any of the specified slub_debug flags is enabled for the
451 * cache. Use only for flags parsed by setup_slub_debug() as it also enables
452 * the static key.
453 */
kmem_cache_debug_flags(struct kmem_cache * s,slab_flags_t flags)454 static inline bool kmem_cache_debug_flags(struct kmem_cache *s, slab_flags_t flags)
455 {
456 if (IS_ENABLED(CONFIG_SLUB_DEBUG))
457 VM_WARN_ON_ONCE(!(flags & SLAB_DEBUG_FLAGS));
458 if (__slub_debug_enabled())
459 return s->flags & flags;
460 return false;
461 }
462
463 #ifdef CONFIG_MEMCG_KMEM
464 /*
465 * slab_objcgs - get the object cgroups vector associated with a slab
466 * @slab: a pointer to the slab struct
467 *
468 * Returns a pointer to the object cgroups vector associated with the slab,
469 * or NULL if no such vector has been associated yet.
470 */
slab_objcgs(struct slab * slab)471 static inline struct obj_cgroup **slab_objcgs(struct slab *slab)
472 {
473 unsigned long memcg_data = READ_ONCE(slab->memcg_data);
474
475 VM_BUG_ON_PAGE(memcg_data && !(memcg_data & MEMCG_DATA_OBJCGS),
476 slab_page(slab));
477 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_KMEM, slab_page(slab));
478
479 return (struct obj_cgroup **)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
480 }
481
482 int memcg_alloc_slab_cgroups(struct slab *slab, struct kmem_cache *s,
483 gfp_t gfp, bool new_slab);
484 void mod_objcg_state(struct obj_cgroup *objcg, struct pglist_data *pgdat,
485 enum node_stat_item idx, int nr);
486
memcg_free_slab_cgroups(struct slab * slab)487 static inline void memcg_free_slab_cgroups(struct slab *slab)
488 {
489 kfree(slab_objcgs(slab));
490 slab->memcg_data = 0;
491 }
492
obj_full_size(struct kmem_cache * s)493 static inline size_t obj_full_size(struct kmem_cache *s)
494 {
495 /*
496 * For each accounted object there is an extra space which is used
497 * to store obj_cgroup membership. Charge it too.
498 */
499 return s->size + sizeof(struct obj_cgroup *);
500 }
501
502 /*
503 * Returns false if the allocation should fail.
504 */
memcg_slab_pre_alloc_hook(struct kmem_cache * s,struct list_lru * lru,struct obj_cgroup ** objcgp,size_t objects,gfp_t flags)505 static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s,
506 struct list_lru *lru,
507 struct obj_cgroup **objcgp,
508 size_t objects, gfp_t flags)
509 {
510 struct obj_cgroup *objcg;
511
512 if (!memcg_kmem_online())
513 return true;
514
515 if (!(flags & __GFP_ACCOUNT) && !(s->flags & SLAB_ACCOUNT))
516 return true;
517
518 objcg = get_obj_cgroup_from_current();
519 if (!objcg)
520 return true;
521
522 if (lru) {
523 int ret;
524 struct mem_cgroup *memcg;
525
526 memcg = get_mem_cgroup_from_objcg(objcg);
527 ret = memcg_list_lru_alloc(memcg, lru, flags);
528 css_put(&memcg->css);
529
530 if (ret)
531 goto out;
532 }
533
534 if (obj_cgroup_charge(objcg, flags, objects * obj_full_size(s)))
535 goto out;
536
537 *objcgp = objcg;
538 return true;
539 out:
540 obj_cgroup_put(objcg);
541 return false;
542 }
543
memcg_slab_post_alloc_hook(struct kmem_cache * s,struct obj_cgroup * objcg,gfp_t flags,size_t size,void ** p)544 static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
545 struct obj_cgroup *objcg,
546 gfp_t flags, size_t size,
547 void **p)
548 {
549 struct slab *slab;
550 unsigned long off;
551 size_t i;
552
553 if (!memcg_kmem_online() || !objcg)
554 return;
555
556 for (i = 0; i < size; i++) {
557 if (likely(p[i])) {
558 slab = virt_to_slab(p[i]);
559
560 if (!slab_objcgs(slab) &&
561 memcg_alloc_slab_cgroups(slab, s, flags,
562 false)) {
563 obj_cgroup_uncharge(objcg, obj_full_size(s));
564 continue;
565 }
566
567 off = obj_to_index(s, slab, p[i]);
568 obj_cgroup_get(objcg);
569 slab_objcgs(slab)[off] = objcg;
570 mod_objcg_state(objcg, slab_pgdat(slab),
571 cache_vmstat_idx(s), obj_full_size(s));
572 } else {
573 obj_cgroup_uncharge(objcg, obj_full_size(s));
574 }
575 }
576 obj_cgroup_put(objcg);
577 }
578
memcg_slab_free_hook(struct kmem_cache * s,struct slab * slab,void ** p,int objects)579 static inline void memcg_slab_free_hook(struct kmem_cache *s, struct slab *slab,
580 void **p, int objects)
581 {
582 struct obj_cgroup **objcgs;
583 int i;
584
585 if (!memcg_kmem_online())
586 return;
587
588 objcgs = slab_objcgs(slab);
589 if (!objcgs)
590 return;
591
592 for (i = 0; i < objects; i++) {
593 struct obj_cgroup *objcg;
594 unsigned int off;
595
596 off = obj_to_index(s, slab, p[i]);
597 objcg = objcgs[off];
598 if (!objcg)
599 continue;
600
601 objcgs[off] = NULL;
602 obj_cgroup_uncharge(objcg, obj_full_size(s));
603 mod_objcg_state(objcg, slab_pgdat(slab), cache_vmstat_idx(s),
604 -obj_full_size(s));
605 obj_cgroup_put(objcg);
606 }
607 }
608
609 #else /* CONFIG_MEMCG_KMEM */
slab_objcgs(struct slab * slab)610 static inline struct obj_cgroup **slab_objcgs(struct slab *slab)
611 {
612 return NULL;
613 }
614
memcg_from_slab_obj(void * ptr)615 static inline struct mem_cgroup *memcg_from_slab_obj(void *ptr)
616 {
617 return NULL;
618 }
619
memcg_alloc_slab_cgroups(struct slab * slab,struct kmem_cache * s,gfp_t gfp,bool new_slab)620 static inline int memcg_alloc_slab_cgroups(struct slab *slab,
621 struct kmem_cache *s, gfp_t gfp,
622 bool new_slab)
623 {
624 return 0;
625 }
626
memcg_free_slab_cgroups(struct slab * slab)627 static inline void memcg_free_slab_cgroups(struct slab *slab)
628 {
629 }
630
memcg_slab_pre_alloc_hook(struct kmem_cache * s,struct list_lru * lru,struct obj_cgroup ** objcgp,size_t objects,gfp_t flags)631 static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s,
632 struct list_lru *lru,
633 struct obj_cgroup **objcgp,
634 size_t objects, gfp_t flags)
635 {
636 return true;
637 }
638
memcg_slab_post_alloc_hook(struct kmem_cache * s,struct obj_cgroup * objcg,gfp_t flags,size_t size,void ** p)639 static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
640 struct obj_cgroup *objcg,
641 gfp_t flags, size_t size,
642 void **p)
643 {
644 }
645
memcg_slab_free_hook(struct kmem_cache * s,struct slab * slab,void ** p,int objects)646 static inline void memcg_slab_free_hook(struct kmem_cache *s, struct slab *slab,
647 void **p, int objects)
648 {
649 }
650 #endif /* CONFIG_MEMCG_KMEM */
651
virt_to_cache(const void * obj)652 static inline struct kmem_cache *virt_to_cache(const void *obj)
653 {
654 struct slab *slab;
655
656 slab = virt_to_slab(obj);
657 if (WARN_ONCE(!slab, "%s: Object is not a Slab page!\n",
658 __func__))
659 return NULL;
660 return slab->slab_cache;
661 }
662
account_slab(struct slab * slab,int order,struct kmem_cache * s,gfp_t gfp)663 static __always_inline void account_slab(struct slab *slab, int order,
664 struct kmem_cache *s, gfp_t gfp)
665 {
666 if (memcg_kmem_online() && (s->flags & SLAB_ACCOUNT))
667 memcg_alloc_slab_cgroups(slab, s, gfp, true);
668
669 mod_node_page_state(slab_pgdat(slab), cache_vmstat_idx(s),
670 PAGE_SIZE << order);
671 }
672
unaccount_slab(struct slab * slab,int order,struct kmem_cache * s)673 static __always_inline void unaccount_slab(struct slab *slab, int order,
674 struct kmem_cache *s)
675 {
676 if (memcg_kmem_online())
677 memcg_free_slab_cgroups(slab);
678
679 mod_node_page_state(slab_pgdat(slab), cache_vmstat_idx(s),
680 -(PAGE_SIZE << order));
681 }
682
cache_from_obj(struct kmem_cache * s,void * x)683 static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x)
684 {
685 struct kmem_cache *cachep;
686
687 if (!IS_ENABLED(CONFIG_SLAB_FREELIST_HARDENED) &&
688 !kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS))
689 return s;
690
691 cachep = virt_to_cache(x);
692 if (WARN(cachep && cachep != s,
693 "%s: Wrong slab cache. %s but object is from %s\n",
694 __func__, s->name, cachep->name))
695 print_tracking(cachep, x);
696 return cachep;
697 }
698
699 void free_large_kmalloc(struct folio *folio, void *object);
700
701 size_t __ksize(const void *objp);
702
slab_ksize(const struct kmem_cache * s)703 static inline size_t slab_ksize(const struct kmem_cache *s)
704 {
705 #ifndef CONFIG_SLUB
706 return s->object_size;
707
708 #else /* CONFIG_SLUB */
709 # ifdef CONFIG_SLUB_DEBUG
710 /*
711 * Debugging requires use of the padding between object
712 * and whatever may come after it.
713 */
714 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
715 return s->object_size;
716 # endif
717 if (s->flags & SLAB_KASAN)
718 return s->object_size;
719 /*
720 * If we have the need to store the freelist pointer
721 * back there or track user information then we can
722 * only use the space before that information.
723 */
724 if (s->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_STORE_USER))
725 return s->inuse;
726 /*
727 * Else we can use all the padding etc for the allocation
728 */
729 return s->size;
730 #endif
731 }
732
slab_pre_alloc_hook(struct kmem_cache * s,struct list_lru * lru,struct obj_cgroup ** objcgp,size_t size,gfp_t flags)733 static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s,
734 struct list_lru *lru,
735 struct obj_cgroup **objcgp,
736 size_t size, gfp_t flags)
737 {
738 flags &= gfp_allowed_mask;
739
740 might_alloc(flags);
741
742 if (should_failslab(s, flags))
743 return NULL;
744
745 if (!memcg_slab_pre_alloc_hook(s, lru, objcgp, size, flags))
746 return NULL;
747
748 return s;
749 }
750
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)751 static inline void slab_post_alloc_hook(struct kmem_cache *s,
752 struct obj_cgroup *objcg, gfp_t flags,
753 size_t size, void **p, bool init,
754 unsigned int orig_size)
755 {
756 unsigned int zero_size = s->object_size;
757 bool kasan_init = init;
758 size_t i;
759
760 flags &= gfp_allowed_mask;
761
762 /*
763 * For kmalloc object, the allocated memory size(object_size) is likely
764 * larger than the requested size(orig_size). If redzone check is
765 * enabled for the extra space, don't zero it, as it will be redzoned
766 * soon. The redzone operation for this extra space could be seen as a
767 * replacement of current poisoning under certain debug option, and
768 * won't break other sanity checks.
769 */
770 if (kmem_cache_debug_flags(s, SLAB_STORE_USER | SLAB_RED_ZONE) &&
771 (s->flags & SLAB_KMALLOC))
772 zero_size = orig_size;
773
774 /*
775 * When slub_debug is enabled, avoid memory initialization integrated
776 * into KASAN and instead zero out the memory via the memset below with
777 * the proper size. Otherwise, KASAN might overwrite SLUB redzones and
778 * cause false-positive reports. This does not lead to a performance
779 * penalty on production builds, as slub_debug is not intended to be
780 * enabled there.
781 */
782 if (__slub_debug_enabled())
783 kasan_init = false;
784
785 /*
786 * As memory initialization might be integrated into KASAN,
787 * kasan_slab_alloc and initialization memset must be
788 * kept together to avoid discrepancies in behavior.
789 *
790 * As p[i] might get tagged, memset and kmemleak hook come after KASAN.
791 */
792 for (i = 0; i < size; i++) {
793 p[i] = kasan_slab_alloc(s, p[i], flags, kasan_init);
794 if (p[i] && init && (!kasan_init || !kasan_has_integrated_init()))
795 memset(p[i], 0, zero_size);
796 kmemleak_alloc_recursive(p[i], s->object_size, 1,
797 s->flags, flags);
798 kmsan_slab_alloc(s, p[i], flags);
799 }
800
801 memcg_slab_post_alloc_hook(s, objcg, flags, size, p);
802 }
803
804 /*
805 * The slab lists for all objects.
806 */
807 struct kmem_cache_node {
808 #ifdef CONFIG_SLAB
809 raw_spinlock_t list_lock;
810 struct list_head slabs_partial; /* partial list first, better asm code */
811 struct list_head slabs_full;
812 struct list_head slabs_free;
813 unsigned long total_slabs; /* length of all slab lists */
814 unsigned long free_slabs; /* length of free slab list only */
815 unsigned long free_objects;
816 unsigned int free_limit;
817 unsigned int colour_next; /* Per-node cache coloring */
818 struct array_cache *shared; /* shared per node */
819 struct alien_cache **alien; /* on other nodes */
820 unsigned long next_reap; /* updated without locking */
821 int free_touched; /* updated without locking */
822 #endif
823
824 #ifdef CONFIG_SLUB
825 spinlock_t list_lock;
826 unsigned long nr_partial;
827 struct list_head partial;
828 #ifdef CONFIG_SLUB_DEBUG
829 atomic_long_t nr_slabs;
830 atomic_long_t total_objects;
831 struct list_head full;
832 #endif
833 #endif
834
835 };
836
get_node(struct kmem_cache * s,int node)837 static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
838 {
839 return s->node[node];
840 }
841
842 /*
843 * Iterator over all nodes. The body will be executed for each node that has
844 * a kmem_cache_node structure allocated (which is true for all online nodes)
845 */
846 #define for_each_kmem_cache_node(__s, __node, __n) \
847 for (__node = 0; __node < nr_node_ids; __node++) \
848 if ((__n = get_node(__s, __node)))
849
850
851 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
852 void dump_unreclaimable_slab(void);
853 #else
dump_unreclaimable_slab(void)854 static inline void dump_unreclaimable_slab(void)
855 {
856 }
857 #endif
858
859 void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr);
860
861 #ifdef CONFIG_SLAB_FREELIST_RANDOM
862 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
863 gfp_t gfp);
864 void cache_random_seq_destroy(struct kmem_cache *cachep);
865 #else
cache_random_seq_create(struct kmem_cache * cachep,unsigned int count,gfp_t gfp)866 static inline int cache_random_seq_create(struct kmem_cache *cachep,
867 unsigned int count, gfp_t gfp)
868 {
869 return 0;
870 }
cache_random_seq_destroy(struct kmem_cache * cachep)871 static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { }
872 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
873
slab_want_init_on_alloc(gfp_t flags,struct kmem_cache * c)874 static inline bool slab_want_init_on_alloc(gfp_t flags, struct kmem_cache *c)
875 {
876 if (static_branch_maybe(CONFIG_INIT_ON_ALLOC_DEFAULT_ON,
877 &init_on_alloc)) {
878 if (c->ctor)
879 return false;
880 if (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON))
881 return flags & __GFP_ZERO;
882 return true;
883 }
884 return flags & __GFP_ZERO;
885 }
886
slab_want_init_on_free(struct kmem_cache * c)887 static inline bool slab_want_init_on_free(struct kmem_cache *c)
888 {
889 if (static_branch_maybe(CONFIG_INIT_ON_FREE_DEFAULT_ON,
890 &init_on_free))
891 return !(c->ctor ||
892 (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)));
893 return false;
894 }
895
896 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG)
897 void debugfs_slab_release(struct kmem_cache *);
898 #else
debugfs_slab_release(struct kmem_cache * s)899 static inline void debugfs_slab_release(struct kmem_cache *s) { }
900 #endif
901
902 #ifdef CONFIG_PRINTK
903 #define KS_ADDRS_COUNT 16
904 struct kmem_obj_info {
905 void *kp_ptr;
906 struct slab *kp_slab;
907 void *kp_objp;
908 unsigned long kp_data_offset;
909 struct kmem_cache *kp_slab_cache;
910 void *kp_ret;
911 void *kp_stack[KS_ADDRS_COUNT];
912 void *kp_free_stack[KS_ADDRS_COUNT];
913 };
914 void __kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab);
915 #endif
916
917 void __check_heap_object(const void *ptr, unsigned long n,
918 const struct slab *slab, bool to_user);
919
920 #ifdef CONFIG_SLUB_DEBUG
921 void skip_orig_size_check(struct kmem_cache *s, const void *object);
922 #endif
923
924 #endif /* MM_SLAB_H */
925