xref: /openbmc/linux/mm/slab.h (revision aa0dc6a7)
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 #ifdef CONFIG_SLOB
9 /*
10  * Common fields provided in kmem_cache by all slab allocators
11  * This struct is either used directly by the allocator (SLOB)
12  * or the allocator must include definitions for all fields
13  * provided in kmem_cache_common in their definition of kmem_cache.
14  *
15  * Once we can do anonymous structs (C11 standard) we could put a
16  * anonymous struct definition in these allocators so that the
17  * separate allocations in the kmem_cache structure of SLAB and
18  * SLUB is no longer needed.
19  */
20 struct kmem_cache {
21 	unsigned int object_size;/* The original size of the object */
22 	unsigned int size;	/* The aligned/padded/added on size  */
23 	unsigned int align;	/* Alignment as calculated */
24 	slab_flags_t flags;	/* Active flags on the slab */
25 	unsigned int useroffset;/* Usercopy region offset */
26 	unsigned int usersize;	/* Usercopy region size */
27 	const char *name;	/* Slab name for sysfs */
28 	int refcount;		/* Use counter */
29 	void (*ctor)(void *);	/* Called on object slot creation */
30 	struct list_head list;	/* List of all slab caches on the system */
31 };
32 
33 #endif /* CONFIG_SLOB */
34 
35 #ifdef CONFIG_SLAB
36 #include <linux/slab_def.h>
37 #endif
38 
39 #ifdef CONFIG_SLUB
40 #include <linux/slub_def.h>
41 #endif
42 
43 #include <linux/memcontrol.h>
44 #include <linux/fault-inject.h>
45 #include <linux/kasan.h>
46 #include <linux/kmemleak.h>
47 #include <linux/random.h>
48 #include <linux/sched/mm.h>
49 
50 /*
51  * State of the slab allocator.
52  *
53  * This is used to describe the states of the allocator during bootup.
54  * Allocators use this to gradually bootstrap themselves. Most allocators
55  * have the problem that the structures used for managing slab caches are
56  * allocated from slab caches themselves.
57  */
58 enum slab_state {
59 	DOWN,			/* No slab functionality yet */
60 	PARTIAL,		/* SLUB: kmem_cache_node available */
61 	PARTIAL_NODE,		/* SLAB: kmalloc size for node struct available */
62 	UP,			/* Slab caches usable but not all extras yet */
63 	FULL			/* Everything is working */
64 };
65 
66 extern enum slab_state slab_state;
67 
68 /* The slab cache mutex protects the management structures during changes */
69 extern struct mutex slab_mutex;
70 
71 /* The list of all slab caches on the system */
72 extern struct list_head slab_caches;
73 
74 /* The slab cache that manages slab cache information */
75 extern struct kmem_cache *kmem_cache;
76 
77 /* A table of kmalloc cache names and sizes */
78 extern const struct kmalloc_info_struct {
79 	const char *name[NR_KMALLOC_TYPES];
80 	unsigned int size;
81 } kmalloc_info[];
82 
83 #ifndef CONFIG_SLOB
84 /* Kmalloc array related functions */
85 void setup_kmalloc_cache_index_table(void);
86 void create_kmalloc_caches(slab_flags_t);
87 
88 /* Find the kmalloc slab corresponding for a certain size */
89 struct kmem_cache *kmalloc_slab(size_t, gfp_t);
90 #endif
91 
92 gfp_t kmalloc_fix_flags(gfp_t flags);
93 
94 /* Functions provided by the slab allocators */
95 int __kmem_cache_create(struct kmem_cache *, slab_flags_t flags);
96 
97 struct kmem_cache *create_kmalloc_cache(const char *name, unsigned int size,
98 			slab_flags_t flags, unsigned int useroffset,
99 			unsigned int usersize);
100 extern void create_boot_cache(struct kmem_cache *, const char *name,
101 			unsigned int size, slab_flags_t flags,
102 			unsigned int useroffset, unsigned int usersize);
103 
104 int slab_unmergeable(struct kmem_cache *s);
105 struct kmem_cache *find_mergeable(unsigned size, unsigned align,
106 		slab_flags_t flags, const char *name, void (*ctor)(void *));
107 #ifndef CONFIG_SLOB
108 struct kmem_cache *
109 __kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
110 		   slab_flags_t flags, void (*ctor)(void *));
111 
112 slab_flags_t kmem_cache_flags(unsigned int object_size,
113 	slab_flags_t flags, const char *name);
114 #else
115 static inline struct kmem_cache *
116 __kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
117 		   slab_flags_t flags, void (*ctor)(void *))
118 { return NULL; }
119 
120 static inline slab_flags_t kmem_cache_flags(unsigned int object_size,
121 	slab_flags_t flags, const char *name)
122 {
123 	return flags;
124 }
125 #endif
126 
127 
128 /* Legal flag mask for kmem_cache_create(), for various configurations */
129 #define SLAB_CORE_FLAGS (SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA | \
130 			 SLAB_CACHE_DMA32 | SLAB_PANIC | \
131 			 SLAB_TYPESAFE_BY_RCU | SLAB_DEBUG_OBJECTS )
132 
133 #if defined(CONFIG_DEBUG_SLAB)
134 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
135 #elif defined(CONFIG_SLUB_DEBUG)
136 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
137 			  SLAB_TRACE | SLAB_CONSISTENCY_CHECKS)
138 #else
139 #define SLAB_DEBUG_FLAGS (0)
140 #endif
141 
142 #if defined(CONFIG_SLAB)
143 #define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \
144 			  SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \
145 			  SLAB_ACCOUNT)
146 #elif defined(CONFIG_SLUB)
147 #define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
148 			  SLAB_TEMPORARY | SLAB_ACCOUNT)
149 #else
150 #define SLAB_CACHE_FLAGS (0)
151 #endif
152 
153 /* Common flags available with current configuration */
154 #define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
155 
156 /* Common flags permitted for kmem_cache_create */
157 #define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \
158 			      SLAB_RED_ZONE | \
159 			      SLAB_POISON | \
160 			      SLAB_STORE_USER | \
161 			      SLAB_TRACE | \
162 			      SLAB_CONSISTENCY_CHECKS | \
163 			      SLAB_MEM_SPREAD | \
164 			      SLAB_NOLEAKTRACE | \
165 			      SLAB_RECLAIM_ACCOUNT | \
166 			      SLAB_TEMPORARY | \
167 			      SLAB_ACCOUNT)
168 
169 bool __kmem_cache_empty(struct kmem_cache *);
170 int __kmem_cache_shutdown(struct kmem_cache *);
171 void __kmem_cache_release(struct kmem_cache *);
172 int __kmem_cache_shrink(struct kmem_cache *);
173 void slab_kmem_cache_release(struct kmem_cache *);
174 
175 struct seq_file;
176 struct file;
177 
178 struct slabinfo {
179 	unsigned long active_objs;
180 	unsigned long num_objs;
181 	unsigned long active_slabs;
182 	unsigned long num_slabs;
183 	unsigned long shared_avail;
184 	unsigned int limit;
185 	unsigned int batchcount;
186 	unsigned int shared;
187 	unsigned int objects_per_slab;
188 	unsigned int cache_order;
189 };
190 
191 void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo);
192 void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s);
193 ssize_t slabinfo_write(struct file *file, const char __user *buffer,
194 		       size_t count, loff_t *ppos);
195 
196 /*
197  * Generic implementation of bulk operations
198  * These are useful for situations in which the allocator cannot
199  * perform optimizations. In that case segments of the object listed
200  * may be allocated or freed using these operations.
201  */
202 void __kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
203 int __kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
204 
205 static inline enum node_stat_item cache_vmstat_idx(struct kmem_cache *s)
206 {
207 	return (s->flags & SLAB_RECLAIM_ACCOUNT) ?
208 		NR_SLAB_RECLAIMABLE_B : NR_SLAB_UNRECLAIMABLE_B;
209 }
210 
211 #ifdef CONFIG_SLUB_DEBUG
212 #ifdef CONFIG_SLUB_DEBUG_ON
213 DECLARE_STATIC_KEY_TRUE(slub_debug_enabled);
214 #else
215 DECLARE_STATIC_KEY_FALSE(slub_debug_enabled);
216 #endif
217 extern void print_tracking(struct kmem_cache *s, void *object);
218 long validate_slab_cache(struct kmem_cache *s);
219 #else
220 static inline void print_tracking(struct kmem_cache *s, void *object)
221 {
222 }
223 #endif
224 
225 /*
226  * Returns true if any of the specified slub_debug flags is enabled for the
227  * cache. Use only for flags parsed by setup_slub_debug() as it also enables
228  * the static key.
229  */
230 static inline bool kmem_cache_debug_flags(struct kmem_cache *s, slab_flags_t flags)
231 {
232 #ifdef CONFIG_SLUB_DEBUG
233 	VM_WARN_ON_ONCE(!(flags & SLAB_DEBUG_FLAGS));
234 	if (static_branch_unlikely(&slub_debug_enabled))
235 		return s->flags & flags;
236 #endif
237 	return false;
238 }
239 
240 #ifdef CONFIG_MEMCG_KMEM
241 int memcg_alloc_page_obj_cgroups(struct page *page, struct kmem_cache *s,
242 				 gfp_t gfp, bool new_page);
243 void mod_objcg_state(struct obj_cgroup *objcg, struct pglist_data *pgdat,
244 		     enum node_stat_item idx, int nr);
245 
246 static inline void memcg_free_page_obj_cgroups(struct page *page)
247 {
248 	kfree(page_objcgs(page));
249 	page->memcg_data = 0;
250 }
251 
252 static inline size_t obj_full_size(struct kmem_cache *s)
253 {
254 	/*
255 	 * For each accounted object there is an extra space which is used
256 	 * to store obj_cgroup membership. Charge it too.
257 	 */
258 	return s->size + sizeof(struct obj_cgroup *);
259 }
260 
261 /*
262  * Returns false if the allocation should fail.
263  */
264 static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s,
265 					     struct obj_cgroup **objcgp,
266 					     size_t objects, gfp_t flags)
267 {
268 	struct obj_cgroup *objcg;
269 
270 	if (!memcg_kmem_enabled())
271 		return true;
272 
273 	if (!(flags & __GFP_ACCOUNT) && !(s->flags & SLAB_ACCOUNT))
274 		return true;
275 
276 	objcg = get_obj_cgroup_from_current();
277 	if (!objcg)
278 		return true;
279 
280 	if (obj_cgroup_charge(objcg, flags, objects * obj_full_size(s))) {
281 		obj_cgroup_put(objcg);
282 		return false;
283 	}
284 
285 	*objcgp = objcg;
286 	return true;
287 }
288 
289 static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
290 					      struct obj_cgroup *objcg,
291 					      gfp_t flags, size_t size,
292 					      void **p)
293 {
294 	struct page *page;
295 	unsigned long off;
296 	size_t i;
297 
298 	if (!memcg_kmem_enabled() || !objcg)
299 		return;
300 
301 	for (i = 0; i < size; i++) {
302 		if (likely(p[i])) {
303 			page = virt_to_head_page(p[i]);
304 
305 			if (!page_objcgs(page) &&
306 			    memcg_alloc_page_obj_cgroups(page, s, flags,
307 							 false)) {
308 				obj_cgroup_uncharge(objcg, obj_full_size(s));
309 				continue;
310 			}
311 
312 			off = obj_to_index(s, page, p[i]);
313 			obj_cgroup_get(objcg);
314 			page_objcgs(page)[off] = objcg;
315 			mod_objcg_state(objcg, page_pgdat(page),
316 					cache_vmstat_idx(s), obj_full_size(s));
317 		} else {
318 			obj_cgroup_uncharge(objcg, obj_full_size(s));
319 		}
320 	}
321 	obj_cgroup_put(objcg);
322 }
323 
324 static inline void memcg_slab_free_hook(struct kmem_cache *s_orig,
325 					void **p, int objects)
326 {
327 	struct kmem_cache *s;
328 	struct obj_cgroup **objcgs;
329 	struct obj_cgroup *objcg;
330 	struct page *page;
331 	unsigned int off;
332 	int i;
333 
334 	if (!memcg_kmem_enabled())
335 		return;
336 
337 	for (i = 0; i < objects; i++) {
338 		if (unlikely(!p[i]))
339 			continue;
340 
341 		page = virt_to_head_page(p[i]);
342 		objcgs = page_objcgs(page);
343 		if (!objcgs)
344 			continue;
345 
346 		if (!s_orig)
347 			s = page->slab_cache;
348 		else
349 			s = s_orig;
350 
351 		off = obj_to_index(s, page, p[i]);
352 		objcg = objcgs[off];
353 		if (!objcg)
354 			continue;
355 
356 		objcgs[off] = NULL;
357 		obj_cgroup_uncharge(objcg, obj_full_size(s));
358 		mod_objcg_state(objcg, page_pgdat(page), cache_vmstat_idx(s),
359 				-obj_full_size(s));
360 		obj_cgroup_put(objcg);
361 	}
362 }
363 
364 #else /* CONFIG_MEMCG_KMEM */
365 static inline struct mem_cgroup *memcg_from_slab_obj(void *ptr)
366 {
367 	return NULL;
368 }
369 
370 static inline int memcg_alloc_page_obj_cgroups(struct page *page,
371 					       struct kmem_cache *s, gfp_t gfp,
372 					       bool new_page)
373 {
374 	return 0;
375 }
376 
377 static inline void memcg_free_page_obj_cgroups(struct page *page)
378 {
379 }
380 
381 static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s,
382 					     struct obj_cgroup **objcgp,
383 					     size_t objects, gfp_t flags)
384 {
385 	return true;
386 }
387 
388 static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
389 					      struct obj_cgroup *objcg,
390 					      gfp_t flags, size_t size,
391 					      void **p)
392 {
393 }
394 
395 static inline void memcg_slab_free_hook(struct kmem_cache *s,
396 					void **p, int objects)
397 {
398 }
399 #endif /* CONFIG_MEMCG_KMEM */
400 
401 static inline struct kmem_cache *virt_to_cache(const void *obj)
402 {
403 	struct page *page;
404 
405 	page = virt_to_head_page(obj);
406 	if (WARN_ONCE(!PageSlab(page), "%s: Object is not a Slab page!\n",
407 					__func__))
408 		return NULL;
409 	return page->slab_cache;
410 }
411 
412 static __always_inline void account_slab_page(struct page *page, int order,
413 					      struct kmem_cache *s,
414 					      gfp_t gfp)
415 {
416 	if (memcg_kmem_enabled() && (s->flags & SLAB_ACCOUNT))
417 		memcg_alloc_page_obj_cgroups(page, s, gfp, true);
418 
419 	mod_node_page_state(page_pgdat(page), cache_vmstat_idx(s),
420 			    PAGE_SIZE << order);
421 }
422 
423 static __always_inline void unaccount_slab_page(struct page *page, int order,
424 						struct kmem_cache *s)
425 {
426 	if (memcg_kmem_enabled())
427 		memcg_free_page_obj_cgroups(page);
428 
429 	mod_node_page_state(page_pgdat(page), cache_vmstat_idx(s),
430 			    -(PAGE_SIZE << order));
431 }
432 
433 static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x)
434 {
435 	struct kmem_cache *cachep;
436 
437 	if (!IS_ENABLED(CONFIG_SLAB_FREELIST_HARDENED) &&
438 	    !kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS))
439 		return s;
440 
441 	cachep = virt_to_cache(x);
442 	if (WARN(cachep && cachep != s,
443 		  "%s: Wrong slab cache. %s but object is from %s\n",
444 		  __func__, s->name, cachep->name))
445 		print_tracking(cachep, x);
446 	return cachep;
447 }
448 
449 static inline size_t slab_ksize(const struct kmem_cache *s)
450 {
451 #ifndef CONFIG_SLUB
452 	return s->object_size;
453 
454 #else /* CONFIG_SLUB */
455 # ifdef CONFIG_SLUB_DEBUG
456 	/*
457 	 * Debugging requires use of the padding between object
458 	 * and whatever may come after it.
459 	 */
460 	if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
461 		return s->object_size;
462 # endif
463 	if (s->flags & SLAB_KASAN)
464 		return s->object_size;
465 	/*
466 	 * If we have the need to store the freelist pointer
467 	 * back there or track user information then we can
468 	 * only use the space before that information.
469 	 */
470 	if (s->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_STORE_USER))
471 		return s->inuse;
472 	/*
473 	 * Else we can use all the padding etc for the allocation
474 	 */
475 	return s->size;
476 #endif
477 }
478 
479 static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s,
480 						     struct obj_cgroup **objcgp,
481 						     size_t size, gfp_t flags)
482 {
483 	flags &= gfp_allowed_mask;
484 
485 	might_alloc(flags);
486 
487 	if (should_failslab(s, flags))
488 		return NULL;
489 
490 	if (!memcg_slab_pre_alloc_hook(s, objcgp, size, flags))
491 		return NULL;
492 
493 	return s;
494 }
495 
496 static inline void slab_post_alloc_hook(struct kmem_cache *s,
497 					struct obj_cgroup *objcg, gfp_t flags,
498 					size_t size, void **p, bool init)
499 {
500 	size_t i;
501 
502 	flags &= gfp_allowed_mask;
503 
504 	/*
505 	 * As memory initialization might be integrated into KASAN,
506 	 * kasan_slab_alloc and initialization memset must be
507 	 * kept together to avoid discrepancies in behavior.
508 	 *
509 	 * As p[i] might get tagged, memset and kmemleak hook come after KASAN.
510 	 */
511 	for (i = 0; i < size; i++) {
512 		p[i] = kasan_slab_alloc(s, p[i], flags, init);
513 		if (p[i] && init && !kasan_has_integrated_init())
514 			memset(p[i], 0, s->object_size);
515 		kmemleak_alloc_recursive(p[i], s->object_size, 1,
516 					 s->flags, flags);
517 	}
518 
519 	memcg_slab_post_alloc_hook(s, objcg, flags, size, p);
520 }
521 
522 #ifndef CONFIG_SLOB
523 /*
524  * The slab lists for all objects.
525  */
526 struct kmem_cache_node {
527 	spinlock_t list_lock;
528 
529 #ifdef CONFIG_SLAB
530 	struct list_head slabs_partial;	/* partial list first, better asm code */
531 	struct list_head slabs_full;
532 	struct list_head slabs_free;
533 	unsigned long total_slabs;	/* length of all slab lists */
534 	unsigned long free_slabs;	/* length of free slab list only */
535 	unsigned long free_objects;
536 	unsigned int free_limit;
537 	unsigned int colour_next;	/* Per-node cache coloring */
538 	struct array_cache *shared;	/* shared per node */
539 	struct alien_cache **alien;	/* on other nodes */
540 	unsigned long next_reap;	/* updated without locking */
541 	int free_touched;		/* updated without locking */
542 #endif
543 
544 #ifdef CONFIG_SLUB
545 	unsigned long nr_partial;
546 	struct list_head partial;
547 #ifdef CONFIG_SLUB_DEBUG
548 	atomic_long_t nr_slabs;
549 	atomic_long_t total_objects;
550 	struct list_head full;
551 #endif
552 #endif
553 
554 };
555 
556 static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
557 {
558 	return s->node[node];
559 }
560 
561 /*
562  * Iterator over all nodes. The body will be executed for each node that has
563  * a kmem_cache_node structure allocated (which is true for all online nodes)
564  */
565 #define for_each_kmem_cache_node(__s, __node, __n) \
566 	for (__node = 0; __node < nr_node_ids; __node++) \
567 		 if ((__n = get_node(__s, __node)))
568 
569 #endif
570 
571 void *slab_start(struct seq_file *m, loff_t *pos);
572 void *slab_next(struct seq_file *m, void *p, loff_t *pos);
573 void slab_stop(struct seq_file *m, void *p);
574 int memcg_slab_show(struct seq_file *m, void *p);
575 
576 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
577 void dump_unreclaimable_slab(void);
578 #else
579 static inline void dump_unreclaimable_slab(void)
580 {
581 }
582 #endif
583 
584 void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr);
585 
586 #ifdef CONFIG_SLAB_FREELIST_RANDOM
587 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
588 			gfp_t gfp);
589 void cache_random_seq_destroy(struct kmem_cache *cachep);
590 #else
591 static inline int cache_random_seq_create(struct kmem_cache *cachep,
592 					unsigned int count, gfp_t gfp)
593 {
594 	return 0;
595 }
596 static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { }
597 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
598 
599 static inline bool slab_want_init_on_alloc(gfp_t flags, struct kmem_cache *c)
600 {
601 	if (static_branch_maybe(CONFIG_INIT_ON_ALLOC_DEFAULT_ON,
602 				&init_on_alloc)) {
603 		if (c->ctor)
604 			return false;
605 		if (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON))
606 			return flags & __GFP_ZERO;
607 		return true;
608 	}
609 	return flags & __GFP_ZERO;
610 }
611 
612 static inline bool slab_want_init_on_free(struct kmem_cache *c)
613 {
614 	if (static_branch_maybe(CONFIG_INIT_ON_FREE_DEFAULT_ON,
615 				&init_on_free))
616 		return !(c->ctor ||
617 			 (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)));
618 	return false;
619 }
620 
621 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG)
622 void debugfs_slab_release(struct kmem_cache *);
623 #else
624 static inline void debugfs_slab_release(struct kmem_cache *s) { }
625 #endif
626 
627 #ifdef CONFIG_PRINTK
628 #define KS_ADDRS_COUNT 16
629 struct kmem_obj_info {
630 	void *kp_ptr;
631 	struct page *kp_page;
632 	void *kp_objp;
633 	unsigned long kp_data_offset;
634 	struct kmem_cache *kp_slab_cache;
635 	void *kp_ret;
636 	void *kp_stack[KS_ADDRS_COUNT];
637 	void *kp_free_stack[KS_ADDRS_COUNT];
638 };
639 void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct page *page);
640 #endif
641 
642 #endif /* MM_SLAB_H */
643